Table of Content

10 October 2019, Volume 39 Issue 10

Previous Issue    Next Issue

Artificial intelligence

Multiple autonomous underwater vehicle task allocation policy based on robust Restless Bandit model

LI Xinbin, ZHANG Shoutao, YAN Lei, HAN Song

2019, 39(10):  2795-2801.  DOI: 10.11772/j.issn.1001-9081.2019020341

Asbtract ( )   PDF (1025KB) ( )  

References | Related Articles | Metrics

The problem of multiple Autonomous Underwater Vehicles (AUV) collaborative task allocation for information acquisition in the underwater detection network was researched. Firstly, a comprehensive model of underwater acoustic monitoring network system was constructed considering the influence of network system sensor nodes status and communication channel status synthetically. Secondly, because of the multi-interference factors under water, with the inaccuracy of the model generation considered, and the multi-AUV task allocation system was modeled as a robust Restless Bandits Problem (RBP) based on the theory of reinforce learning. Lastly, the robust Whittle algorithm was proposed to solve the RBP problem to get the task allocation policy of multi-AUV. Simulation results show that when the system selected 1, 2 and 3 targets, the system cumulative return performance of the robust allocation policy improves by 5.5%, 12.3% and 9.6% respectively compared with that of the allocation strategy without interference factors considered, proving the effectiveness of the proposed approaches.

Compression method of super-resolution convolutional neural network based on knowledge distillation

GAO Qinquan, ZHAO Yan, LI Gen, TONG Tong

2019, 39(10):  2802-2808.  DOI: 10.11772/j.issn.1001-9081.2019030516

Asbtract ( )   PDF (1103KB) ( )  

References | Related Articles | Metrics

Aiming at the deep structure and high computational complexity of current network models based on deep learning for super-resolution image reconstruction, as well as the problem that the networks can not operate effectively on resource-constrained devices caused by the high storage space requirement for the network models, a super-resolution convolutional neural network compression method based on knowledge distillation was proposed. This method utilizes a teacher network with large parameters and good reconstruction effect as well as a student network with few parameters and poor reconstruction effect. Firstly the teacher network was trained; then knowledge distillation method was used to transfer knowledge from teacher network to student network; finally the reconstruction effect of the student network was improved without changing the network structure and the parameters of the student network. The Peak Signal-to-Noise Ratio (PSNR) was used to evaluate the quality of reconstruction in the experiments. Compared to the student network without knowledge distillation method, the student network using the knowledge distillation method has the PSNR increased by 0.53 dB, 0.37 dB, 0.24 dB and 0.45 dB respectively on four public test sets when the magnification times is 3. Without changing the structure of student network, the proposed method significantly improves the super-resolution reconstruction effect of the student network.

Improved elastic network model for deep neural network

FENG Minghao, ZHANG Tianlun, WANG Linhui, CHEN Rong, LIAN Shaojing

2019, 39(10):  2809-2814.  DOI: 10.11772/j.issn.1001-9081.2019040624

Asbtract ( )   PDF (886KB) ( )  

References | Related Articles | Metrics

Deep neural networks tend to suffer from overfitting problem because of the high complexity of the model. To reduce the adverse eeffects of the problem on the network performance, an improved elastic network model based deep learning optimization method was proposed. Firstly, considering the strong correlation between the variables, the adaptive weights were assigned to different variables of L1-norm in elastic network model, so that the linear combination of the L2-norm and the adaptively weighted L1-norm was obtained. Then, the solving process of neural network parameters under this new regularization term was given by combining improved elastic network model with the deep learning optimization model. Moreover, the robustness of this proposed model was theoretically demonstrated by showing the grouping selection ability and Oracle property of the improved elastic network model in the optimization of neural network. At last, in regression and classification experiments, the proposed model was compared with L1-norm, L2-norm and elastic network regularization term, and had the regression error decreased by 87.09, 88.54 and 47.02 and the classification accuracy improved by 3.98, 2.92 and 3.58 percentage points respectively. Thus, theory and experimental results prove that the improved elastic network model can effectively improve the generalization ability of deep neural network model and the performance of optimization algorithm, and solve the overfitting problem of deep learning.

Fast feature selection method based on mutual information in multi-label learning

XU Hongfeng, SUN Zhenqiang

2019, 39(10):  2815-2821.  DOI: 10.11772/j.issn.1001-9081.2019030483

Asbtract ( )   PDF (965KB) ( )  

References | Related Articles | Metrics

Concerning the high time complexity of traditional heuristic search-based multi-label feature selection algorithm, an Easy and Fast Multi-Label Feature Selection (EF-MLFS) method was proposed. Firstly, Mutual Information (MI) was used to measure the features and the correlations between the labels of each dimension; then, the obtained correlations were added up and ranked; finally, feature selection was performed according to the total correlation. The proposed method was compared to six existing representative multi-label feature selection methods such as Max-Dependency and Min-Redundancy (MDMR) algorithm, Multi-Label Naive Bayes (MLNB) method. Experimental results show that the average precision, coverage, Hamming Loss and other common multi-label classification indicators are optimal after feature selection and classificationby using EF-MLFS method. In addition, global search is not required in the method, so the time complexity is significantly reduced compared with MDMR and Pairwise Mutli-label Utility (PMU).

Semi-supervised self-training positive and unlabeled learning based on new spy technology

LI Tingting, LYU Jia, FAN Weiya

2019, 39(10):  2822-2828.  DOI: 10.11772/j.issn.1001-9081.2019040606

Asbtract ( )   PDF (1083KB) ( )  

References | Related Articles | Metrics

Spy technology in Positive and Unlabeled (PU) learning is easily susceptible to noise and outliers, which leads to the impurity of reliable positive instances, and the mechanism of selecting spy instances in the initial positive instances randomly tends to cause inefficiency in dividing reliable negative instances. To solve these problems, a PU learning framework combining new spy technology and semi-supervised self-training was proposed. Firstly, the initial labeled instances were clustered and the instances closer to the cluster center were selected to replace the spy instances. These instances were able to map the distribution structure of unlabeled instances effectively, so as to better assist to the selection of reliable negative instances. Then, the reliable positive instances divided by spy technology were purified by self-training, and the reliable negative instances which were divided as positive instances mistakenly were corrected by secondary training. The proposed framework can solve the problem of PU learning that the classification efficiency of traditional spy technology is susceptible to data distribution and random spy instances. The experiments on nine standard data sets show that the average classification accuracy and F-measure of the proposed framework are higher than those of Basic PU-learning algorithm (Basic_PU), PU-learning algorithm based on spy technology (SPY), Self-Training PU learning algorithm based on Naive Bayes (NBST) and Iterative pruning based PU learning (Pruning) algorithm.

Euclidean embedding recommendation algorithm by fusing trust information

XU Lingling, QU Zhijian, XU Hongbo, CAO Xiaowei, LIU Xiaohong

2019, 39(10):  2829-2833.  DOI: 10.11772/j.issn.1001-9081.2019040597

Asbtract ( )   PDF (819KB) ( )  

References | Related Articles | Metrics

To solve the sparse and cold start problems of recommendation system, a Trust Regularization Euclidean Embedding (TREE) algorithm by fusing trust information was proposed. Firstly, the Euclidean embedding model was employed to embed the user and project in the unified low-dimensional space. Secondly, to measure the trust information, both the project participation degree and user common scoring factor were brought into the user similarity calculation formula. Finally, a regularization term of social trust relationship was added to the Euclidean embedding model, and trust users with different preferences were used to constrain the location vectors of users and generate the recommendation results. In the experiments, the proposed TREE algorithm was compared with the Probabilistic Matrix Factorization (PMF), Social Regularization (SoReg), Social Matrix Factorization (SocialMF) and Recommend with Social Trust Ensemble (RSTE) algorithms. When dimensions are 5 and 10, TREE algorithm has the Root Mean Squared Error (RMSE) decreased by 1.60% and 5.03% respectively compared with the optimal algorithm RSTE on the dataset Filmtrust.While on the dataset Epinions, the RMSE of TREE algorithm was respectively 1.12% and 1.29% lower than that of the optimal algorithm SocialMF. Experimental results show that TREE algorithm further alleviate the sparse and cold start problems and improves the accuracy of scoring prediction.

Recommendation algorithm based on probability matrix factorization and fusing trust

TIAN Baojun, YANG Huyun, FANG Jiandong

2019, 39(10):  2834-2840.  DOI: 10.11772/j.issn.1001-9081.2019030583

Asbtract ( )   PDF (933KB) ( )  

References | Related Articles | Metrics

For the problems of low recommendation accuracy, data sparsity and malicious recommendation, a new recommendation model based on Probability Matrix Factorization (PMF) and fusing trust was proposed. Firstly, by establishing a Collaborative Filtering Model based on Trust Similarity (CFMTS), the improved trust mechanism was integrated into the collaborative filtering recommendation algorithm. The trust value was obtained through global trust and local trust calculation. The local trust was obtained by calculating the direct trust value and the indirect trust value of the user by the trust propagation mechanism, the global trust was calculated by the trust directed graph. Then, the trust value was combined with the score similarity to solve the problems of data sparsity and malicious recommendation. At the same time, CFMTS was integrated into the PMF model to establish a new recommendation model-Model based on Probability Matrix Factorization and Fusing Trust (MPMFFT). The user feature vectors and the project feature vectors were calculated by the gradient descent algorithm to generate the predicted scores, further improving the accuracy of the recommender system. Through experiments, the proposed MPMFFT was compared with the classical models such as PMF, Social Matrix Factorization (SocialMF), Social Recommendation (SoRec) and Recommendations with Social Trust Ensemble (RSTE). The proposed model has the Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) decreased by 2.9% and 1.5% respectively compared with the optimal model RSTE on the open real dataset Epinions, and has the MAE and RMSE decreased by 1.1% and 1.8% respectively compared with the optimal SocialMF model on open real dataset Ciao, verifying that the proposed model is significantly improved on the above indicators. The results confirme that the propose model can resolve the problem of data sparseness and malicious recommendation to some extent, and effectively improved the recommendation quality.

Chinese text sentiment analysis based on CNN-BiGRU network with attention mechanism

WANG Liya, LIU Changhui, CAI Dunbo, LU Tao

2019, 39(10):  2841-2846.  DOI: 10.11772/j.issn.1001-9081.2019030579

Asbtract ( )   PDF (909KB) ( )  

References | Related Articles | Metrics

In the traditional Convolutional Neural Network (CNN), the information cannot be transmitted to each other between the neurons of the same layer, the feature information at the same layer cannot be fully utilized, making the lack of the representation of the characteristics of the sentence system. As the result, the feature learning ability of model is limited and the text classification effect is influenced. Aiming at the problem, a model based on joint network CNN-BiGRU and attention mechanism was proposed. In the model, the CNN-BiGRU joint network was used for feature learning. Firstly, deep-level phrase features were extracted by CNN. Then, the Bidirectional Gated Recurrent Unit (BiGRU) was used for the serialized information learning to obtain the characteristics of the sentence system and strengthen the association of CNN pooling layer features. Finally, the effective feature filtering was completed by adding attention mechanism to the hidden state weighted calculation. Comparative experiments show that the method achieves 91.93% F1 value and effectively improves the accuracy of text classification with small time cost and good application ability.

Simultaneous localization and semantic mapping of indoor dynamic scene based on semantic segmentation

XI Zhihong, HAN Shuangquan, WANG Hongxu

2019, 39(10):  2847-2851.  DOI: 10.11772/j.issn.1001-9081.2019040711

Asbtract ( )   PDF (735KB) ( )  

References | Related Articles | Metrics

To address the problem that dynamic objects affect pose estimation in indoor Simultaneous Localization And Mapping (SLAM) systems, a semantic segmentation based SLAM system in dynamic scenes was proposed. Firstly, an image was semantically segmented by the Pyramid Scene Parsing Network (PSPNet) after being captured by the camera. Then image feature points were extracted, feature points distributed in the dynamic object were removed, and camera pose was estimated by using static feature points. Finally, the semantic point cloud map and semantic octree map were constructed. Results of multiple comparison tests on five dynamic sequences of public datasets show that compared with the SLAM system using SegNet network, the proposed system has the standard deviation of absolute trajectory error improved by 6.9%-89.8%, and has the standard deviation of translation and rotation drift improved by 73.61% and 72.90% respectively in the best case in high dynamic scenes. The results show that the improved method can significantly reduce the error of pose estimation in dynamic scenes, and can correctly estimate the camera pose in dynamic scenes.

Attribute reduction of relative indiscernibility relation and discernibility relation in relation decision system

LI Xu, RONG Zijing, RUAN Xiaoxi

2019, 39(10):  2852-2858.  DOI: 10.11772/j.issn.1001-9081.2019030438

Asbtract ( )   PDF (980KB) ( )  

References | Related Articles | Metrics

Corresponding reduction algorithms for relative indiscernibility and discernibility relation were proposed. Firstly, considering the reduction of the relative indiscernibility relation in equivalence relation, the corresponding discernibility matrix was proposed and a reduction algorithm was proposed based on the matrix. Then, a reduction algorithm for relative discernibility relation was proposed according to the complementary relationship of the relation. Secondly, the concepts such as relative indiscernibility relation were expanded to the general relation. The corresponding discernibility matrix was proposed for the relative indiscernibility relation reduction in the relation decision system, and the corresponding discernibility matrix for the relative discernibility relation reduction was obtained by using the complementary relationship of the relation, so the reduction algorithms for both relations were obtained. Finally, the proposed algorithms were verified on the selected UCI datasets. In the equivalence relation, the algorithm of the relative EQuivalence INDiscernibility relation reduction based on absolute reduction (EQIND) and the algorithm of the relative BInary INDiscernibility relation reduction (BⅡND) have the same results. The algorithm of the relative EQuivalence DIScernibility relation reduction based on absolute reduction (EQDIS) and the algorithm of the relative BInary DIScernibility relation reduction (BIDIS) have the same results. Meanwhile, BⅡND and BIDIS are suitable for the incomplete decision table. The feasibility of the proposed algorithms were verified by the experimental results.

Path planning of mobile robot based on multi-objective grasshopper optimization algorithm

HUANG Chao, LIANG Shengtao, ZHANG Yi, ZHANG Jie

2019, 39(10):  2859-2864.  DOI: 10.11772/j.issn.1001-9081.2019040722

Asbtract ( )   PDF (873KB) ( )  

References | Related Articles | Metrics

In the mobile robot path planning problem in static multi-obstacle environment, Particle Swarm Optimization (PSO) algorithm has the disadvantages of easy premature convergence and poor local optimization ability, resulting in low accuracy of robot path planning. To solve the problem, a Multi-Objective Grasshopper Optimization Algorithm (MOGOA) was proposed. The path length, smoothness and security were taken as path optimization targets according to the mobile robot path planning requirements, and the corresponding mathematical model of multi-objective optimization problem was established. In the process of population search, the curve adaptive strategy was introduced to speed up the convergence of the algorithm, and the Pareto optimal criterion was used to solve the coexistence problem of the above three targets. Experimental results show that the proposed algorithm finds shorter paths and shows better convergence while solving the above problems. Compared with the Multi-Objective Particle Swarm Optimization (MOPSO) algorithm, the proposed algorithm has the path length reduced by about 2.01 percentage, and the number of iterations reduced by about 19.34 percentage.

Narrow channel path planning based on bidirectional rapidly-exploring random tree

FU Jiupeng, ZENG Guohui, HUANG Bo, FANG Zhijun

2019, 39(10):  2865-2869.  DOI: 10.11772/j.issn.1001-9081.2019030508

Asbtract ( )   PDF (813KB) ( )  

References | Related Articles | Metrics

In the process of mobile robot path planning, it is difficult for the Rapidly-exploration Random Tree (RRT) algorithm to sample narrow channels. In order to deal with this problem, an improved bridge detection algorithm was proposed, which is dedicated to narrow channel sampling. Firstly, the environment map was pre-processed and the obstacle edge coordinate set was extracted as the sampling space for the bridge detection algorithm, thus avoiding a large number of invalid sampling points and making the sampling points distribution of the narrow channel more rational. Secondly, the process for bridge endpoint construction was improved, and the operation efficiency of the bridge detection algorithm was increased. Finally, a slight variant Connect algorithm was used to expand the narrow channel sample points rapidly. For the narrow channel environment map in the experiment, the improved algorithm has the success rate increased from 68% to 92% compared with the original RRT-Connect algorithm. Experimental results show that the proposed algorithm can sample the narrow channel well and improve the efficiency of path planning.

Motion planning model based on deep cascaded neural network for autonomous driving

BAI Liyun, HU Xuemin, SONG Sheng, TONG Xiuchi, ZHANG Ruohan

2019, 39(10):  2870-2875.  DOI: 10.11772/j.issn.1001-9081.2019040629

Asbtract ( )   PDF (992KB) ( )  

References | Related Articles | Metrics

To address the problems that rule-based motion planning algorithms under constraints need pre-definition of rules and temporal features are not considered in deep learning-based methods, a motion planning model based on deep cascading neural networks was proposed. In this model, the two classical deep learning models, Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) network, were combined to build a novel cascaded neural network, the spatial and temporal features of the input images were extracted respectively, and the nonlinear relationship between the input sequential images and the output motion parameters were fit to achieve the end-to-end planning from the input sequential images to the output motion parameters. In experiments, the data of simulated environment were used for training and testing. Results show that the Root Mean Squared Error (RMSE) of the proposed model in four scenes including country road, freeway, tunnel and mountain road is less than 0.017, and the stability of the prediction results of the proposed model is better than that of the algorithm without using cascading neural network by an order of magnitude. Experimental results show that the proposed model can effectively learn human driving behaviors, eliminate the effect of cumulative errors and adapt to different scenes of a variety of road conditions with good robustness.

Motion planning algorithm of robot for crowd evacuation based on deep Q-network

ZHOU Wan, HU Xuemin, SHI Chenyin, WEI Jieling, TONG Xiuchi

2019, 39(10):  2876-2882.  DOI: 10.11772/j.issn.1001-9081.2019030507

Asbtract ( )   PDF (1195KB) ( )  

References | Related Articles | Metrics

Aiming at the danger and unsatisfactory effect of dense crowd evacuation in public places in emergency, a motion planning algorithm of robots for crowd evacuation based on Deep Q-Network (DQN) was proposed. Firstly, a human-robot social force model was constructed by adding human-robot interaction to the original social force model, so that the motion state of crowd was able to be influenced by the robot force on pedestrians. Then, a motion planning algorithm of robot was designed based on DQN. The images of the original pedestrian motion state were input into the network and the robot motion behavior was output. In this process, the designed reward function was fed back to the network to enable the robot to autonomously learn from the closed-loop process of "environment-behavior-reward". Finally, the robot was able to learn the optimal motion strategies at different initial positions to maximize the total number of people evacuated after many iterations. The proposed algorithm was trained and evaluated in the simulated environment. Experimental results show that the proposed algorithm based on DQN increases the evacuation efficiency by 16.41%, 10.69% and 21.76% respectively at three different initial positions compared with the crowd evacuation algorithm without robot, which proves that the algorithm can significantly increase the number of people evacuated per unit time with flexibility and effectiveness.

Forest fire smoke detection model based on deep convolution long short-term memory network

WEI Xin, WU Shuhong, WANG Yaoli

2019, 39(10):  2883-2887.  DOI: 10.11772/j.issn.1001-9081.2019040707

Asbtract ( )   PDF (774KB) ( )  

References | Related Articles | Metrics

Since the smoke characteristics of each sampled frame have great similarity, and the forest fire smoke dataset is relatively small and monotonous, in order to make full use of the static and dynamic information of smoke to prevent forest fires, a Deep Convolution Integrated Long Short-Term Memory network (DC-ILSTM) model was proposed. Firstly, VGG-16 networks pre-trained on ImageNet dataset were used for feature transfer based on isomorphic data to effectively extract smoke characteristics. Secondly, an Integrated Long Short-Term Memory network (ILSTM) based on pooling layer and Long Short-Term Memory network (LSTM) was proposed, and ILSTM was used for segmental fusion of smoke characteristics. Finally, a trainable deep neural network model was built for forest fire smoke detection. In the smoke detection experiment, compared with Deep Convolution Long Recursive Network (DCLRN), DC-ILSTM can detect smoke with 10 frames advantage under the optimal efficiency and has the test accuracy increased by 1.23 percentage points. The theoretical analysis and simulation results show that DC-ILSTM has good applicability in forest fire smoke detection.

Classification of online loan based on improved cost-sensitive decision tree

GUO Bingnan, WU Guangchao

2019, 39(10):  2888-2892.  DOI: 10.11772/j.issn.1001-9081.2019020827

Asbtract ( )   PDF (657KB) ( )  

References | Related Articles | Metrics

In the online loan user data set, there is a serious imbalance between the number of successful and failed loan users. The traditional machine learning algorithm pays attention to the overall classification accuracy when solving such problems, which leads to lower prediction accuracy of successful loan users. In order to solve this problem, the class distribution was added to the calculation of cost-sensitive decision tree sensitivity function, in order to weaken the impact of positive and negative samples on the misclassification cost, and an improved cost-sensitive decision tree based on ID3 (ID3cs)was constructed. With the improved cost-sensitive decision tree as the base classifier and the classification accuracy as the criterion, the base classifiers with better performance were selected and integrated with the classifier generated in the last stage to obtain the final classifier. Experimental results show that compared with the existing algorithms to solve such problems (such as MetaCost algorithm, cost-sensitive decision tree, AdaCost algorithm), the improved cost-sensitive decision tree can reduce the overall misclassification rate of online loan users and has stronger generalization ability.

Partial occlusion face recognition based on structured occlusion coding and extreme learning machine

ZHANG Fangyan, WANG Xin, XU Xinzheng

2019, 39(10):  2893-2898.  DOI: 10.11772/j.issn.1001-9081.2019051176

Asbtract ( )   PDF (865KB) ( )  

References | Related Articles | Metrics

An algorithm combining Structured Occlusion Coding (SOC) with Extreme Learning Machine (ELM) was proposed to deal with the occlusion problem in face recognition. Firstly, the SOC was used to remove the occlusion from the image and separate the oclusion from the human face. At the same time, the position of the occlusion was estimated by the Local Constraint Dictionary (LCD), and an occlusion dictionary and a face dictionary were established. Then, the established face dictionary matrix was normalized, and the ELM was used to classify and identify the normalized data. Finally, the simulation results on the AR face database show that the proposed method has higher recognition rate and stronger robustness for different types of occlusions and images with different regions occluded.

RGB-NIR image demosaicing based on deep learning

XIE Changjiang, YANG Xiaomin, YAN Binyu, LU Lu

2019, 39(10):  2899-2904.  DOI: 10.11772/j.issn.1001-9081.2019040614

Asbtract ( )   PDF (1000KB) ( )  

References | Related Articles | Metrics

Spectral interference in Red Green Blue-Near InfRared (RGB-NIR) images captured by single sensor results in colour distortion and detail information ambiguity of the reconstructed standard Red Green Blue (RBG) and Near InfRared (NIR) images. To resolve this problem, a demosaicing method based on deep learning was proposed. In this method, the grandient dppearance and dispersion problems were solved by introducing long jump connection and dense connection, the network was easier to be trained, and the fitting ability of the network was improved. Firstly, the low-level features such as pixel correlation and channel correlation of the mosaic image were extracted by the shallow feature extraction layer. Secondly, the obtained shallow feature graph was input into successive and multiple residual dense blocks to extract the high-level semantic features aiming at the demosaicing. Thirdly, to make full use of the low-level features and high-level features, the features extracted by multiple residual dense blocks were combined. Finally, the RGB-NIR image was reconstructed by the global long jump connection. Experiments were performed on the deep learning framework Tensorflow using three public data sets, the Common Image and Visual Representation Group (IVRG) dataset, the Outdoor Multi-Spectral Images with Vegetation (OMSIV) dataset, and the Forest dataset. The experimental results show that the proposed method is superior to the RGB-NIR image demosaicing methods based on multi-level adaptive residual interpolation, convolutional neural network and deep residual U-shaped network.

Remote sensing image segmentation method based on deep learning model

XU Yue, FENG Mengru, PI Jiatian, CHEN Yong

2019, 39(10):  2905-2914.  DOI: 10.11772/j.issn.1001-9081.2019030529

Asbtract ( )   PDF (1531KB) ( )  

References | Related Articles | Metrics

To detect surface object information quickly and accurately by using remote sensing images is a current research hot spot. In order to solve the problems of inefficiency of the traditional manual visual interpretation segmentation method as well as the low accuracy and a lot of background noise of the existing remote sensing image segmentation based on deep learning in complex scenes, an image segmentation algorithm based on improved U-net network architecture and fully connected conditional random field was proposed. Firstly, a new network model was constructed by integrating VGG16 and U-net to effectively extract the features of remote sensing images with highly complex background. Then, by selecting the appropriate activation function and convolution method, the image segmentation accuracy was improved while the model prediction time was significantly reduced. Finally, on the basis of guaranteeing the segmentation accuracy, the segmentation result was further improved by using fully connected conditional random field. The simulation test on the standard dataset Potsdam provided by ISPRS showed that the accuracy, recall and the Mean Intersection over Union (MIoU) of the proposed algorithm were increased by 15.06 percentage points, 29.11 percentage points and 0.3662 respectively, and the Mean Absolute Error (MAE) of the algorithm was reduced by 0.02892 compared with those of U-net. Experimental results verify that the proposed algorithm is an effective and robust algorithm for extracting surface objects from remote sensing images.

Automatic recognition algorithm for cervical lymph nodes using cascaded fully convolutional neural networks

QIN Pinle, LI Pengbo, ZENG Jianchao, ZHU Hui, XU Shaowei

2019, 39(10):  2915-2922.  DOI: 10.11772/j.issn.1001-9081.2019030510

Asbtract ( )   PDF (1267KB) ( )  

References | Related Articles | Metrics

The existing automatic recognition algorithms for cervical lymph nodes have low efficiency, and the overall false positive removal are unsatisfied, so a cervical lymph node detection algorithm using cascaded Fully Convolutional Neural Networks (FCNs) was proposed. Firstly, combined with the prior knowledge of doctors, the cascaded FCNs were used for preliminary identification, that was, the first FCN was used for extracting the cervical lymph node region from the Computed Tomography (CT) image of head and neck. Then, the second FCN was used to extract the lymph node candidate samples from the region, and merging them at the three-dimensional (3D) level to generate a 3D image block. Finally, the proposed feature block average pooling method was introduced into the 3D classification network, and the 3D input image blocks with different scales were classified into two classes to remove false positives. On the cervical lymph node dataset, the recall of cervical lymph nodes identified by cascaded FCNs is up to 97.23%, the classification accuracy of the 3D classification network with feature block average pooling can achieve 98.7%. After removing false positives, the accuracy of final result reaches 93.26%. Experimental results show that the proposed algorithm can realize the automatic recognition of cervical lymph nodes with high recall and accuracy, which is better than the current methods reported in the literatures; it is simple and efficient, easy to extend to other tasks of 3D medical images recognition.

Gastric tumor cell image recognition method based on radial transformation and improved AlexNet

GAN Lan, GUO Zihan, WANG Yao

2019, 39(10):  2923-2929.  DOI: 10.11772/j.issn.1001-9081.2019040709

Asbtract ( )   PDF (1200KB) ( )  

References | Related Articles | Metrics

When using AlexNet to implement image classification of gastric tumor cells, there are problems of small dataset, slow model convergence and low recognition rate. Aiming at the above problems, a Data Augmentation (DA) method based on Radial Transformation (RT) and improved AlexNet was proposed. The original dataset was divided into test set and training set. In the test set, cropping was used to increase the data. In the training set, cropping, rotation, flipping and brightness conversion were employed to obtain the enhanced image set, and then some of them were selected for RT processing to achieve the enhanced effect. In addition, the replacement activation of functions and normalization layers was used to speed up the convergence and improve the generalization performance of AlexNet. Experimental results show that the proposed method can implement the recognition of gastric tumor cell images with faster convergence and higher recognition accuracy. On the test set, the highest accuracy is 99.50% and the average accuracy is 96.69%, and the F1 scores of categories:canceration, normal and hyperplasia are 0.980, 0.954 and 0.958 respectively, indicating that the proposed method can implement the recognition of gastric tumor cell images well.

Multi-scale grape image recognition method based on convolutional neural network

QIU Jinyi, LUO Jun, LI Xiu, JIA Wei, NI Fuchuan, FENG Hui

2019, 39(10):  2930-2936.  DOI: 10.11772/j.issn.1001-9081.2019040594

Asbtract ( )   PDF (1038KB) ( )  

References | Related Articles | Metrics

Grape quality inspection needs the identification of multiple categories of grapes, and there are many scenes such as depth of field changes and multiple strings in the grape images. Grape recognition is ineffective due to the limitations of single pretreatment method. The research objects were 15 kinds of natural scene grape images collected in the greenhouse, and the corresponding image dataset Vitis-15 was established. Aiming at the large intra-class differences and small inter-class of differences grape images, a multi-scale grape image recognition method based on Convolutional Neural Network (CNN) was proposed. Firstly, the data in Vitis-15 dataset were pre-processed by three methods, including the image rotating based data augmentation method, central cropping based multi-scale image method and data fusion method of the above two. Then, transfer learning method and convolution neural network method were adopted to realiize the classification and recognition. The Inception V3 network model pre-trained on ImageNet was selected for transfer learning, and three types of models-AlexNet, ResNet and Inception V3 were selected for convolution neural network. The multi-scale image data fusion classification model MS-EAlexNet was proposed, which was suitable for Vitis-15. Experimental results show that with the same learning rate on the same test dataset, compared with the augmentation and multi-scale image method, the data fusion method improves nearly 1% testing accuracy on MS-EAlexNet model with 99.92% accuracy, meanwhile the proposed method has higher efficiency in classifying small sample datasets.

Angular interval embedding based end-to-end voiceprint recognition model

WANG Kang, DONG Yuanfei

2019, 39(10):  2937-2941.  DOI: 10.11772/j.issn.1001-9081.2019040757

Asbtract ( )   PDF (827KB) ( )  

References | Related Articles | Metrics

An end-to-end model with angular interval embedding was constructed to solve the problems of complicated multiple steps and weak generalization ability in the traditional voiceprint recognition model based on the combination of identity vector (i-vector) and Probabilistic Linear Discriminant Analysis (PLDA). A deep convolutional neural network was specially designed to extract deep speaker embedding from the acoustic features of voice data. The Angular Softmax (A-Softmax), which is based on angular improvement, was employed as the loss function to keep the angular interval between the different classes of features learned by the model and make the clustering of the similar features closer in the angle space. Compared with the method combining i-vector and PLDA, it shows that the proposed model has the identification accuracy of Top-1 and Top-5 increased by 58.9% and 30% respectively and has the minimum detection cost and equal error rate reduced by 47.9% and 45.3% respectively for speaker verification on the public dataset VoxCeleb2. The results verify that the proposed end-to-end model is more suitable for learning class-discriminating features from multi-channel and large-scale datasets.

Data science and technology

Mass data clean system based on regular expression

CHANG Zheng, LYU Yong

2019, 39(10):  2942-2947.  DOI: 10.11772/j.issn.1001-9081.2019030492

Asbtract ( )   PDF (866KB) ( )  

References | Related Articles | Metrics

Based on the current mainstream Extract Transform Load (ETL) tools for data and the disadvantages of some applications in restricted environments, a Regular Expression Mass-data Cleaning System (REMCS) was proposed for the specific requirements in the restricted application scenarios. Firstly, the data features of six typical problems including ultra-long error data, batch fusion of data source files, automatic sorting of data source files, were discovered. And the appropriate regular expressions and pre-processing algorithms were put forward according to the data features. Then, data pre-processing was completed by using the algorithm model to remove the errors in data. At the same time, the system logical structure, common problems, and corresponding solutions, and code implementation scheme of REMCS were described in detail. Finally, the comparison experiments of several common data processing problems were carried out with the following four aspects:the compatible data source file formats, the soveble problem types, the problem processing time and the data processing limit value. Compared with the traditional ETL tools, REMCS can support nine typical file formats such as csv format, json format, dump format, and can address all six common problems with shorter processing time and larger supportable data limit value. Experimental results show that REMCS has better applicability and high accuracy for common data processing problems in restricted application scenarios.

Multi-keyword parallel ciphertext retrieval scheme in distributed environment

DAI Houle, YANG Geng, MIN Zhao'e

2019, 39(10):  2948-2954.  DOI: 10.11772/j.issn.1001-9081.2019020376

Asbtract ( )   PDF (1151KB) ( )  

References | Related Articles | Metrics

For searchable encryption, balancing the security and retrieval efficiency of data is important. Aiming at the low retrieval performance and the lack of single keyword search mode in SSE-1 ciphertext retrieval scheme, and the problems such as the limitation of single-machine resources in the traditional single-server architecture, a multi-keyword parallel ciphertext retrieval system was designed and implemented. Different index encryption strategies were used to improve the ciphertext retrieval performance. The block search of the inverted index was realized by partitioning the ciphertext inverted index, which solves the limitation of single-machine resources and improves the retrieval efficiency. The traditional single-machine retrieval architecture was extended and the parallel retrieval of multiple keywords was realized by combining the characteristic of distribution. Experimental results show that compared with the SSE-1 scheme, the proposed scheme has the efficiency of retrieval and update operations improved under the premise of ensuring ciphertext data security and realizes multi-keyword retrieval. At the same time, the distributed architecture of the system is dynamically expanded to improve the system load capacity.

False trend time series detection based on randomness analysis

LI Jianxun, MA Meiling, GUO Jianhua, YAN Jun

2019, 39(10):  2955-2959.  DOI: 10.11772/j.issn.1001-9081.2019030573

Asbtract ( )   PDF (805KB) ( )  

References | Related Articles | Metrics

Focusing on the detection problem of false data that conform to a certain pattern or rule, a false trend time series detection method based on randomness analysis was proposed. Based on the analysis of time series composition, firstly the simple forgery method and complex forgery method of false trend time series were explored, and decomposed into two parts:false trendness and false randomness. Then the false trend of time series was extracted by the approximation of base function, the false random of time series was analyzed with the randomness theory. Finally, monobit frequency and frequency within a block were adopted to test whether the false random part has randomness, which established a detection method of false time series with a certain trend feature. The simulation results show that proposed method can decompose the false time series and extract the false trend part effectively, meanwhile realize the detection of simple and complex forged data. It also supports the authenticity analysis for the numerical data obtained by means of observation or monitoring equipment, which improves the discrimination range of false data with average detection accuracy of 74.7%.

Measurement of spatial straightness of train axle

WANG Hua, HOU Daishuang, ZHANG Shuang, GAO Jingang

2019, 39(10):  2960-2965.  DOI: 10.11772/j.issn.1001-9081.2019020318

Asbtract ( )   PDF (858KB) ( )  

References | Related Articles | Metrics

In order to accurately and quickly measure the spatial straightness of train axle, a measurement system of train axle spatial straightness was constructed and the algorithms of spatial circle fitting, spatial straight line fitting and straightness measurement were studied. Firstly, the spatial circle fitting algorithm based on spatial plane and spatial sphere tangent was introduced according to the characteristics of the object under test. Then, the RANdom SAmple Consensus (RANSAC) algorithm was used to iterate out the best point set of the model. On the basis of the data obtained from the spatial circle fitting of the train axle section, the data of the train axle section spatial circle center was analyzed. And the wolf colony algorithm was used to fit the spatial straight line, that is, according to the circle center coordinates of the spatial circle of the train axle section at the position of the space section, the spatial straight line of the train axle was fitted. Finally, the wolf colony algorithm was used to measure the spatial straightness of train axle, and the measured data were compared with the data of laser tracker. Experimental results show that the accuracy of measuring the spatial straightness of train axle based on wolf colony algorithm is 0.01 mm, which can meet the requirements of high accuracy, high stability and repeatability in measurement of the spatial straightness of train axle.

Cyber security

Network negative energy propagation dynamics model and simulation

LIU Chao, HUANG Shiwen, YANG Hongyu, CAO Qiong, LIU Xiaoyang

2019, 39(10):  2966-2972.  DOI: 10.11772/j.issn.1001-9081.2019040664

Asbtract ( )   PDF (1008KB) ( )  

References | Related Articles | Metrics

In view of the problem that the existing researches do not consider the refinement of the factors affecting the network negative energy propagation and construct a propagation dynamics model for analysis, a Weak-Strong-Received-Infected-Evil (WSRIE) model of network negative energy propagation was proposed. Firstly, considering the difference of negative energy immunity and propagation ability of network users, the vulnerable states were divided into weak immunity and strong immunity, and the infectious states were divided into weak infection, strong infection and malicious propagation with unchanged scale. Secondly, according to the negative energy infection mechanism of the network, the state transition law was proposed. Finally, a dynamics model of network negative energy propagation for complex networks was constructed. The simulation comparison experiments on NW small world network and BA scale-free network were carried out. The simulation results show that under the same parameters, the weak immune node density of the NW network is 9 percentage points lower than that of the BA network, indicating that the network with small world characteristics is more susceptible to negative energy. In the BA network, the density of infected nodes with the malicious node degree of 200 is 5 percentage points higher than that with the node degree of 0, indicating that the greater the node degree of the network red opinion leader, the more network users affected by the network negative energy.

Selective encryption scheme based on Logistic and Arnold transform in high efficiency video coding

ZHOU Yizhao, WANG Xiaodong, ZHANG Lianjun, LAN Qiongqiong

2019, 39(10):  2973-2979.  DOI: 10.11772/j.issn.1001-9081.2019040742

Asbtract ( )   PDF (1054KB) ( )  

References | Related Articles | Metrics

In order to effectively protect video information, according to the characteristics of H.265/HEVC (High Efficiency Video Coding), a scheme combining transform coefficient scrambling and syntax element encryption was proposed. For Transform Unit (TU), the TU with the size of 4×4 was scrambled by Arnold transform. At the same time, a shift cipher was designed, and the cipher was initialized according to the approximate distribution rule of the Direct Current (DC) coefficient of the TU, and the DC coefficients of TU with the size of 8×8, 16×16 and 32×32 were shifting encrypted using encryption map generated by Arnold transform. For some of the syntax elements with bypass coding used in the entropy coding process, the Logistic chaotic sequence was used for encryption. After encryption, the Peak Signal-to-Noise Ratio (PSNR) and Structual Similarity (SSIM) of the video were decreased by 26.1 dB and 0.51 respectively on average, while the compression ratio was only decreased by 1.126% and the coding time was only increased by 0.17%. Experimental results show that under the premise of ensuring better encryption effect and less impact on bit rate, the proposed scheme has less extra coding overhead and is suitable for real-time video applications.

Advanced computing

Scheduling algorithm for periodic tasks with low energy consumption based on heterogeneous mult-core platforms

XIA Jun, YUAN Shuai, YANG Yi

2019, 39(10):  2980-2984.  DOI: 10.11772/j.issn.1001-9081.2019040665

Asbtract ( )   PDF (842KB) ( )  

References | Related Articles | Metrics

Concerning at the high energy consumption of heterogeneous multi-core platforms, an algorithm for solving the optimal energy allocation scheme of periodic tasks by using optimization theory was proposed. The optimal energy consumption problem of periodic tasks was modeled and added constraints to the model. According to the optimization theory, the binary integer programming problem was relaxed to obtain the convex optimization problem. The interior point method was used to solve the optimization problem and the relaxed distribution matrix was obtained. The allocation scheme for partial tasks was obtained after the judgement processing of the decision matrix. On this basis, the iterative method was used to find the allocation scheme for the remaining tasks. Experimental results show that the energy consumption of this distribution scheme is reduced by about 1.4% compared with the similar optimization theory algorithm, and compared with the optimization theory algorithm with the similar energy consumption, the execution time of this scheme is reduced by 86%. At the same time, the energy consumption of the scheme is only 2.6% higher than the theoretically optimal energy consumption.

Best and worst coyotes strengthened coyote optimization algorithm and its application to quadratic assignment problem

ZHANG Xinming, WANG Doudou, CHEN Haiyan, MAO Wentao, DOU Zhi, LIU Shangwang

2019, 39(10):  2985-2991.  DOI: 10.11772/j.issn.1001-9081.2019030454

Asbtract ( )   PDF (1090KB) ( )  

References | Related Articles | Metrics

In view of poor performance of Coyote Optimization Algorithm (COA), a Best and Worst coyotes strengthened COA (BWCOA) was proposed. Firstly, for growth of the worst coyote in the group, a global optimal coyote guiding operation was introduced on the basis of the optimal coyote guidance to improve the social adaptability (local search ability) of the worst coyote. Then, a random perturbation operation was embedded in the growth process of the optimal coyote in the group, which means using the random perturbation between coyotes to promote the development of the coyotes and make full play of the initiative of each coyotes in the group to improve the diversity of the population and thus to enhance the global search ability, while the growing pattern of the other coyotes kept unchanged. BWCOA was applied to complex function optimization and Quadratic Assignment Problem (QAP) using hospital department layout as an example. Experimental results on CEC-2014 complex functions show that compared with COA and other state-of-the-art algorithms, BWCOA obtains 1.63 in the average ranking and 1.68 rank mean in the Friedman test, both of the results are the best. Experimental results on 6 QAP benchmark sets show that BWCOA obtains the best mean values for 5 times. These prove that BWCOA is more competitive.

Imperialist competitive algorithm based on multiple search strategy for solving traveling salesman problem

CHEN Menghui, LIU Junlin, XU Jianfeng, LI Xiangjun

2019, 39(10):  2992-2996.  DOI: 10.11772/j.issn.1001-9081.2019030434

Asbtract ( )   PDF (802KB) ( )  

References | Related Articles | Metrics

The imperialist competitive algorithm is a swarm intelligence optimization algorithm with strong local search ability, but excessive local search will lead to the loss of diversity and fall into local optimum. Aiming at this problem, an Imperialist Competitive Algorithm based on Multiple Search Strategy (MSSICA) was proposed. The country was defined as a feasible solution and the kingdoms were defined as four mechanisms of combinatorial artificial chromosome with different characteristics. The block mechanism was used to retain the dominant solution fragment during search and differentiated mechanisms of combinatorial artificial chromosome was used for different empires to search the effective and feasible solution information of different solution spaces. When it come to the local optimum, the multiple search strategy was used to inject a uniformly distributed feasible solution to replace a less advantageous solution to enhance the diversity. Experimental results show that the multiple search strategy can effectively improve diversity of the imperialist competitive algorithm and improve the quality and stability of the solution.

Construction and characteristic analysis of Chebyshev mapping system based on homogenized distribution

HUANG Bin, BAO Liyong, DING Hongwei

2019, 39(10):  2997-3001.  DOI: 10.11772/j.issn.1001-9081.2019020255

Asbtract ( )   PDF (719KB) ( )  

References | Related Articles | Metrics

Concerning the bimodal distribution characteristics of the range boundary presented by the traditional Chebyshev mapping, in order to meet the requirements of homogenized distribution of sequences in optimization theory, the mathematical equation was given by using the probability density function of Chebyshev mapping, and a new system was constructed by combining with the original mapping into a new system. The comparative study shows that the system has good homogenized distribution characteristic, ergodic characteristic, balance and low complexity, and the random error of the generated sequences is small and the similarity is high. Finally, the system is applied to the initialization population stage of the optimization algorithm, and it is further shown that the homogenized distribution system has a significant effect on improving the homogenized distribution characteristic of the original mapping.

Network and communications

Credibility analysis method of online user behavior based on non-interference theory

DONG Haiyan, YU Feng, CHENG Ke, HUANG Shucheng

2019, 39(10):  3002-3006.  DOI: 10.11772/j.issn.1001-9081.2019040660

Asbtract ( )   PDF (863KB) ( )  

References | Related Articles | Metrics

Focusing on the difficulty in monitoring and judging the credibility of user behaviors in online applications and the problem of weak theorey of user behavior credibility analysis, a credibility analysis method of online user behavior was proposed based on non-interference theory. Firstly, the static credibility of single behavior was defined from three aspects-the behavioral entity identity, state and environment of the single behavior, and the static credibility verification strategy was given. Thereafter, dynamic behavioral credibility was defined from the perspectives of execution process and result, and dynamic credibility verification strategy was given. Finally, the user behavior process was constructed based on the single behavior, and the credibility determination theorem of user behavior process was proposed based on the idea of credibility extension, and the theorem was proved by using non-interference theory. The correctness and validity of the proposed method were verified by the provement process and result.

Precoding based on improved conjugate gradient algorithm in massive multi-input multi-output system

BAI He, LIU Ziyan, ZHANG Jie, WAN Peipei, MA Shanshan

2019, 39(10):  3007-3012.  DOI: 10.11772/j.issn.1001-9081.2019040638

Asbtract ( )   PDF (825KB) ( )  

References | Related Articles | Metrics

To solve the problems of high complexity of precoding and difficulty of linear matrix inversion in downlink Massive Multi-Input Multi-Output (Massive MIMO) system, a precoding algorithm based on low-complexity Symmetric Successive Over Relaxation Preconditioned Conjugate Gradient (SSOR-PCG) was proposed. Based on preconditioned Conjugate Gradient Precoding (PCG) algorithm, a Symmetric Successive Over Relaxation (SSOR) algorithm was used to preprocess the matrix to reduce its condition number, accelerating the convergence speed and the decreasing the complexity. Simulation results demonstrate that compared with PCG algorithm, the proposed algorithm has running time of around 88.93% shortened and achieves convergence when the Signal-to-Noise Ratio (SNR) is 26 dB. Furthermore, compared to zero-forcing precoding algorithm, the proposed algorithm requires only two iterations capacity-approaching performance,the overall complexity is reduced by one order of magnitude, and the bit error rate is decreased by about 49.94%.

Gardner timing recovery algorithm for improved loop structure

LI Wei, JIANG Hong, WU Chun, DENG Haowen

2019, 39(10):  3013-3017.  DOI: 10.11772/j.issn.1001-9081.2019040636

Asbtract ( )   PDF (744KB) ( )  

References | Related Articles | Metrics

Aiming at the problems of long synchronization setup time and poor synchronization stability in classical Gardner timing recovery algorithms, a Gardner timing synchronization recovery algorithm with improved loop structure was proposed. Firstly, two interpolation filters with cubic interpolation and piece wise parabolic interpolation were used to obtain two optimal interpolation sequences. Secondly, the timing errors corresponding to the two interpolation sequences were calculated respectively and the weighted average value was obtained to gain the timing error of the loop. Finally, the weighted average value of two optimal interpolation sequences was used as the loop output. The simulation experiments of two modulated signals of Quadrature Phase Shift Keying (QPSK) and 16 Quadrature Amplitude Modulation (16QAM) were performed. Simulation results show that the synchronization stability of the proposed algorithm is better on QPSK signal. Compared with performing on 16QAM signal, the number of sequences corresponding to the position of the symbols when the loop starts the synchronization is obviously reduced. Additionally by using the propposed algorithm, the convergence radius of the QPSK constellation is about 0.26 when the SNR is -5 dB. Compared with the improved Gardner timing recovery algorithm similar to Frequency and Phase Lock Loop (FPLL), the convergence radius is reduced by 0.08. This algorithm effectively shortens the synchronization setup time, improves the stability of the loop, and can be widely applied in high-speed demodulation system.

Computer software technology

Correctness verification of static taint analysis results for Android application

QIN Biao, GUO Fan, TU Fengtao

2019, 39(10):  3018-3027.  DOI: 10.11772/j.issn.1001-9081.2019040644

Asbtract ( )   PDF (1509KB) ( )  

References | Related Articles | Metrics

Many false positives are generated when an Android application is detected by static taint analysis to discover potential privacy-leak bugs. For that, a context-sensitive, path-sensitive and field-sensitive semi-auto analysis method was proposed to verify if a potential bug is a true positive by only traversing a few executable paths. Firstly, a seed Trace covering both Source and Sink was obtained manually by running the instrumented application. Then, a Trace-based taint analysis method was used to verify if there was a taint propagating path in the Trace. If there was a taint propagating path, it meaned a real privacy leak bug existed. If not, the conditioin set and taint information of the Trace were further collected, and by combining the live-variable analysis and the program transformation approach based on conditional inversion, a constraint selection policy was designed to prune most executable paths irrelevant to taint propagation. Finally, remaining executable paths were traversed and corresponding Traces were analyzed to verify if the bug is a false positive. Seventy-five applications of DroidBench and ten real applications were tested by a prototype system implemented on FlowDroid. Results show that only 15.09% paths traversed averagely in each application, the false positive rate decreases 58.17% averagely. Experimental results demonstrate the analysis can effectively reduce the false positives generated by static taint analysis.

Virtual reality and multimedia computing

Non-rigid point set registration based on global and local similarity measurement

PENG Lei, YANG Xiuyun, ZHANG Yufei, LI Guangyao

2019, 39(10):  3028-3033.  DOI: 10.11772/j.issn.1001-9081.2019040681

Asbtract ( )   PDF (861KB) ( )  

References | Related Articles | Metrics

In the non-rigid point set registration algorithm, whether the correct correspondence can be found plays an important role. Generally the corresponding points in two point sets have similar neighborhood structures besides the close distance. Therefore, a non-rigid point set registration algorithm based on global and local similarity measurement was proposed. Firstly, the Coherent Point Drift (CPD) algorithm was used as the registration framework, and the Gaussian mixture model was used to model the point sets. Secondly, the global and local mixture distance was improved to form the global and local similarity measurement criterion. Finally, the correspondence and the transformation formula were solved by the Expectation Maximization (EM) algorithm. In the initial stage of the iteration, the proportion of local similarity was larger so that the correct correspondence was able to be found rapidly; with the progress of the iteration, the proportion of global similarity was increased to ensure the smaller registration error. Experimental results show that compared with the Thin Plate Spline Robust Point Matching (TPS-RPM) algorithm, the Gaussian Mixture Models point set REGistration (GMMREG) algorithm, the Robust Point Matching algorithm based on L₂E estimation (RPM-L2E), the Global and Local Mixture Distance and Thin Plate Spline based point set registration algorithm (GLMDTPS) and the CPD algorithm, the proposed algorithm has the Root Mean Squared Error (RMSE) decreased by 39.93%, 42.45%, 32.51%, 22.36% and 11.76% respectively, indicating the proposed algorithm has better registration performance.

Mesh parameterization method based on limiting distortion

CAI Xingquan, SUN Chen, GE Yakun

2019, 39(10):  3034-3039.  DOI: 10.11772/j.issn.1001-9081.2019030550

Asbtract ( )   PDF (991KB) ( )  

References | Related Articles | Metrics

Aiming at the low efficiency and serious mapping distortion of current mesh parameterization, a mesh parameterization method with limiting distortion was proposed. Firstly, the original mesh model was pre-processed. After inputting the original 3D mesh model, the Half-Edge data structure was used to reorganize the mesh and the corresponding seams were generated by cutting the mesh model. The Tutte mapping was constructed to map the 3D mesh to a 2D convex polygon domain, that is to construct the 2D mesh model. Then, the mesh parameterization calculation with limiting distortion was performed. The Tutte-mapped 2D mesh model was used as the initial data for limiting distortion calculation, and the distortion metric function relative to the original 3D model mesh was established. The minimum value points of the metric function were obtained, which form the mapped mesh coordinate set. The mapped mesh was used as the input mesh to limit the distortion mapping, and the iteration termination condition was set. The iteration was performed cyclically until the termination condition was satisfied, and the convergent optimal mesh coordinates were obtained. In calculating the mapping distortion, the Dirichlet energy function was used to measure the isometric mapping distortion, and the Most Isometric Parameterizations (MIPS) energy function was used for the conformal mapping distortion. The minimum of the mapping distortion energy function was solved by proxy function combining assembly-Newton method. Finally, this method was implemented and a prototype system was developed. In the prototype system, mesh parameterization experiments for limiting isometric distortion and limiting conformal distortion were designed respectively, statistics and comparisons of program execution time and distortion energy falling were performed, and the corresponding texture mapping effects were displayed. Experimental results show that the proposed method has high execution efficiency, fast falling speed of mapping distortion energy and stable quality of optimal value convergence. When texture mapping is performed, the texture is evenly colored, close laid and uniformly lined, which meets the practical application standards.

Image super-resolution algorithm based on adaptive anchored neighborhood regression

YE Shuang, YANG Xiaomin, YAN Bin'yu

2019, 39(10):  3040-3045.  DOI: 10.11772/j.issn.1001-9081.2019040760

Asbtract ( )   PDF (1001KB) ( )  

References | Related Articles | Metrics

Among the dictionary-based Super-Resolution (SR) algorithms, the Anchored Neighborhood Regression (ANR) algorithm has been attracted widely attention due to its superior reconstruction speed and quality. However, the anchored neighborhood projections of ANR are unstable to cover varieties of mapping relationships. Aiming at the problem, an image SR algorithm based on adaptive anchored neighborhood regression was proposed, which adaptively calculated the neighborhood center based on the distribution of samples in order to pre-estimate the projection matrix based on more accurate neighborhood. Firstly, K-means clustering algorithm was used to cluster the training samples into different clusters with the image patches as centers. Then, the dictionary atoms were replaced with the cluster centers to calculate the corresponding neighborhoods. Finally, the neighborhoods were applied to pre-compute the projection matrix from LR space to HR space. Experimental results show that the average reconstruction performance of the proposed algorithm on Set14 is better than that of other state-of-the-art dictionary-based algorithms with 31.56 dB of Peak Signal-to-Noise Ratio (PSNR) and 0.8712 of Structural SIMilarity index (SSIM), and even is superior to the Super-Resolution Convolutional Neural Network (SRCNN) algorithm. At the same time, in terms of the subjective performance, the proposed algorithm produces sharp edges in reconstruction results with little artifacts.

Low-illumination image enhancement algorithm based on multi-scale gradient domain guided filtering

LI Hong, WANG Ruiyao, GENG Zexun, HU Haifeng

2019, 39(10):  3046-3052.  DOI: 10.11772/j.issn.1001-9081.2019040642

Asbtract ( )   PDF (1112KB) ( )  

References | Related Articles | Metrics

An improved low-illumination image enhancement algorithm was proposed to solve the problems that the overall intensity of low-illumination color image is low, the color in the enhanced image is easy to be distorted, and some enhanced image details are drowned in the pixels with low gray value. Firstly, an image to be processed was converted to the Hue Saturation Intensity (HSI) color space, and the nonlinear global intensity correction was carried out for the intensity component. Then, an intensity enhancement model based on multi-scale guided gradient domain filtering was put forward to enhance the corrected intensity component, and the intensity correction was further performed to avoid color distortion. Finally, the image was converted back into Red Green Blue (RGB) color space. Experimental results show that the enhanced images have the intensity increased by more than 90.0% on average, and the sharpness increased by more than 123.8% on average, which are mainly due to the better intensity smoothing and enhancement ability of multi-scale gradient domain guided filtering. At the same time, due to the reduction of color distortion, the detail performance of enhanced images increases by more than 18.2% on average. The proposed low-illumination image enhancement algorithm is suitable for enhancing color images under night and other weak light source conditions, because of using intensity enhancement model based on multi-scale gradient domain guided filtering and histogram adaptive intensity correction algorithm.

Fast image mosaic algorithm based on adaptive elimination of stitching seam and panorama alignment

YANG Chunde, CHENG Yanfei

2019, 39(10):  3053-3059.  DOI: 10.11772/j.issn.1001-9081.2019030544

Asbtract ( )   PDF (1150KB) ( )  

References | Related Articles | Metrics

Aiming at the phenomenon that image mosaic, to a certain extent, has uneven chromatic aberration, distortion and low efficiency, an adaptive elimination of image stitching seam and panorama alignment based fast image mosaic algorithm was proposed. Firstly, the Scale-Invariant Feature Transform (SIFT) was used to extract feature points of the specified area of the image and image registration was performed by using bidirectional K-Nearest Neighbor (KNN) algorithm, effectively improving the algorithm efficiency. Secondly, focusing on the uneven chromatic aberration transition of stitching seam, an adaptive formula for finding the optimal stitching seam was proposed based on dynamic programming, and then the seam was adaptively eliminated by image fusion. Finally, for the phenomenon of panoramic tilt caused by accumlated stitching error, an adaptive fitting quadrilateral alignment model based on edge detection algorithm was proposed to make the original panorama into a completely new panorama. Compared with the image mosaic algorithm based on block and the image mosaic algorithm based on binary tree, the proposed algorithm has the image quality improved by 5.84%-7.83% and the stitching time shortened to only 50%-70% of the original. Experimental results show that the proposed algorithm not only reduces the unevenness of chromatic aberration transition of stitching seam in different image backgrounds through adaptive update mechanism, so as to improve the image quality, but also increases the stitching efficiency and reduces the distortion degree of panorama.

Image stitching by combining deformation function and power function weight

LI Jialiang, JIANG Pinqun

2019, 39(10):  3060-3064.  DOI: 10.11772/j.issn.1001-9081.2019020239

Asbtract ( )   PDF (802KB) ( )  

References | Related Articles | Metrics

An image stitching method based on Scale-Invariant Feature Transform (SIFT), thin-plate spline function and power function was proposed to solve the problem of low efficiency, mismatching of feature points, ghosting and stitching seam in image stitching algorithm. The point mapping relationship and overlapping area between the images were calculated by sampling and matching the input images. The local distortion model of the image was calculated by the feature point set, and the deformation of the image was completed by image interpolation. The power function weighting model was used to realize smooth transaction of the pixels in the deformed image to complete the image stitching. Experimental results show that the proposed method improves the registration efficiency of the feature points approximately by 59.78% and obtains more pairs of feature points compared to the traditional SIFT algorithm. Moreover, compared with the classical image stitching algorithm, the method solves the problems of image ghosting and stitching seam, and improves the score of image quality evaluation index.

Monaural speech enhancement algorithm based on mask estimation and optimization

GE Wanying, ZHANG Tianqi

2019, 39(10):  3065-3070.  DOI: 10.11772/j.issn.1001-9081.2019030486

Asbtract ( )   PDF (892KB) ( )  

References | Related Articles | Metrics

Monaural speech enhancement algorithms obtain enhanced speech by estimating and negating the noise components in speech with noise. However, the over-estimation and the error of the introduction to make up the over-estimation of noise power make detrimental effect on the enhanced speech. To constrain the distortion caused by noise over-estimation, a time-frequency mask estimation and optimization algorithm based on Computational Auditory Scene Analysis (CASA) was proposed. Firstly, Decision Directed (DD) algorithm was used to estimate the priori Signal-to-Noise Ratio (SNR) and calculate the initial mask. Secondly, the Inter-Channel Correlation (ICC) factor between noise and speech with noise in each Gammatone filterbank channel was used to calculate the noise presence probability, the new noise estimation was obtained by the probability combining with the power spectrum of speech with noise, and the over-estimation of the primary estimated noise was decreased. Thirdly, the initial mask was iterated by the optimization algorithm to reduce the error caused by the noise over-estimation and raise the target speech components in the mask, and the new mask was obtained when the iteration stopped with the conditions met. Finally, the optimization method was used to optimize the estimated mask. The enhanced speech was composed by using the new mask. Experimental results demonstrate that the new mask has higher Perceptual Evaluation of Speech Quality (PESQ) and Short-Time Objective Intelligibility measure (STOI) values of the enhanced speech in comparison with the mask before optimization, improving the intelligibility and listening feeling of speech.

Frontier & interdisciplinary applications

Siting model of electric taxi charging station based on GPU parallel computing

WU Xuchen, PIAO Chunhui, JIANG Xuehong

2019, 39(10):  3071-3078.  DOI: 10.11772/j.issn.1001-9081.2019040762

Asbtract ( )   PDF (1285KB) ( )  

References | Related Articles | Metrics

Aiming at the problem of optimal siting of charging station for electric taxis, a siting model of charging station for electric taxis was established with the unmet charging demand and the minimum fixed cost of constructing new charging station as objective functions, and a model solution method based on improved multi-objective particle swarm optimization was proposed. In order to solve the performance bottleneck of computing unmet charging demand, a Graphics Processing Unit (GPU)-based unmet charging demand parallel algorithm was designed. Experimental results demonstrat that the running time of the parallel algorithm is about 10%-12% of that of CPU-based serial algorithm. Beijing was taken as an example of applying the proposed charging station siting model, and related multi-source data was collected and processed. The results show that the proposed optimal siting scheme for charging station is feasible.

Urban traffic networks collaborative optimization method based on two-layered complex networks

CHEN Xiaoming, LI Yinzhen, SHEN Qiang, JU Yuxiang

2019, 39(10):  3079-3087.  DOI: 10.11772/j.issn.1001-9081.2019030538

Asbtract ( )   PDF (1344KB) ( )  

References | Related Articles | Metrics

In order to solve the problems in the transfer process connection and collaboration of metro-bus two-layered network faced by the passengers making route selection in the urban transportation network, such as the far distance between some transfer stations, the unclear connection orientation and the imbalance between supply and demand in local transfer, a collaborative optimization method for urban traffic networks based on two-layered complex networks was presented. Firstly, the logical network topology method was applied to the topology of the urban transportation network, and the metro-bus two-layered network model was established by the complex network theory. Secondly, with the transfer station as research object, a node importance evaluation method based on K-shell decomposition method and central weight distribution was presented. This method was able to realize coarse and fine-grained divison and identification of metro and bus stations in large-scale networks. And a collaborative optimization method for two-layered urban traffic network with mutual encouragement was presented, that is to say the method in the complex network theory to identify and filter the node importance in network topology was introduced to the two-layered network structure optimization. The two-layered network structure was updated by identifying high-aggregation effects and locating favorable nodes in the route selection to optimize the layout and connection of stations in the existing network. Finally, the method was applied to the Chengdu metro-bus network, the existing network structure was optimized to obtain the optimal optimized node location and number of existing network, and the effectiveness of the method was verified by the relevant index system. The results show that the global efficiency of the network is optimized after 32 optimizations, and the optimization effect of the average shortest path is 15.89% and 16.97%, respectively, and the passenger transfer behavior is increased by 57.44 percentage points, the impact on the accessibility is the most obvious when the travel cost is 8000-12000 m with the optimization effect of 23.44% on average. At the same time, with the two-layered network speed ratio and unit transportation cost introduced, the response and sensitivity difference of the traffic network to the collaborative optimization process under different operational conditions are highlighted.

Industrial X-ray image enhancement algorithm based on gradient field

ZHOU Chong, LIU Huan, ZHAO Ailing, ZHANG Pengcheng, LIU Yi, GUI Zhiguo

2019, 39(10):  3088-3092.  DOI: 10.11772/j.issn.1001-9081.2019040694

Asbtract ( )   PDF (843KB) ( )  

References | Related Articles | Metrics

In the detection of components with uneven thickness by X-ray, the problems of low contrast or uneven contrast and low illumination often occur, which make it difficult to observe and analyze some details of components in the images obtained. To solve this problem, an X-ray image enhancement algorithm based on gradient field was proposed. The algorithm takes gradient field enhancement as the core and is divided into two steps. Firstly, an algorithm based on logarithmic transformation was proposed to compress the gray range of an image, remove redundant gray information of the image and improve image contrast. Then, an algorithm based on gradient field was proposed to enhance image details, improve local image contrast and image quality, so that the details of components were able to be clearly displayed on the detection screen. A group of X-ray images of components with uneven thickness were selected for experiments, and the comparisons with algorithms such as Contrast Limited Adaptive Histogram Equalization (CLAHE) and homomorphic filtering were carried out. Experimental results show that the proposed algorithm has more obvious enhancement effect and can better display the detailed information of the components. The quantitative evaluation criteria of calculating average gradient and No-Reference Structural Sharpness (NRSS) texture analysis further demonstrate the effectiveness of this algorithm.

Deep in vivo quantitative photoacoustic imaging based on improved fixed point iterative method

LIU Fangyan, MENG Jing, SI Guangtao

2019, 39(10):  3093-3099.  DOI: 10.11772/j.issn.1001-9081.2019010076

Asbtract ( )   PDF (1116KB) ( )  

References | Related Articles | Metrics

Focusing on the reconstruction artifact of photoacoustic images in restricted view, an improved fixed-point iterative quantitative photoacoustic imaging method was proposed. Firstly, the original photoacoustic pressure data detected by the detector were reconstructed by the traditional back projection reconstruction algorithm to obtain the original photoacoustic pressure image. Secondly, the original photoacoustic pressure image was filtered to remove the reconstruction artifact by adaptive Wiener filtering algorithm. Thirdly, the optical transmission model was used to solve the optical flux of the target imaging region. And finally, iterative calculation was performed to obtain the optical absorption coefficient of the target tissue. In addition, Toast++ software was introduced in the process of solving the optical flux to realize the forward solution of the optical transmission model, which improved the efficiency and accuracy of quantitative imaging. The phantom and in vivo experiments show that compared with the traditional fixed-point iterative method, the proposed method can obtain photoacoustic images with higher quality and there are fewer artifacts in the deep quantitative photoacoustic images reconstructed by the method. The optical absorption coefficient of the quantitatively reconstructed deep target tissue is very close to the optical absorption coefficient of the shallow target tissue, the former is about 70% of the latter. As a result, the quantitative reconstruction of the optical absorption coefficient of the deep biological tissue can be implemented by the proposed method.

Manifold regularized sparse constraint nonnegative matrix factorization with superpixel algorithm for hyperspectral unmixing

LI Denggang, CHEN Xiangxiang, LI Huali, WANG Zhongmei

2019, 39(10):  3100-3106.  DOI: 10.11772/j.issn.1001-9081.2019030534

Asbtract ( )   PDF (1048KB) ( )  

References | Related Articles | Metrics

For the problems such as poor unmixing results and sensitivity to noise of traditional Nonnegative Matrix Factorization (NMF) applied to hyperspectral unmixing, a Manifold Regularized Sparse NMF with superpixel (MRS-NMF) algorithm for hyperspectral unmixing was proposed. Firstly, the manifold structure of hyperspectral image was constructed by superpixel segmentation based on entropy. The original image was divided into k-superpixel blocks, and the data points in each superpixel block with same property were labeled the same label. Weight matrices were defined between any two data points with the similar label in a superpixel block, and then the weight matrices were applied to the objective function of NMF to construct the manifold regularization constraint. Secondly, a quadratic parabola function was added to the objective function to complete the sparse constraint. Finally, the multiplicative iterative update rule was used to solve the objective function to obtain the solution formulas of endmember matrix and abundance matrix. At the same time, maximum iteration times and tolerate error threshold were set to get the final results by iterative operation. The proposed method makes full use of spectral and spatial information of hyperspectral images. Experimental results show that on synthetic data the unmixing accuracies of endmember and abundance based on proposed MRS-NMF are 0.016-0.063 and 0.01-0.05 respectively higher than those based on traditional methods like Graph-regularized L1/2-Nonnegative Matrix Factorization (GLNMF), L1/2NMF and Vertex Component Analysis-Fully Constrained Least Squares (VCA-FCLS); while on real hyperspectral images, the average unmixing accurary of endmember based on proposed MRS-NMF is 0.001-0.0437 higher than that of traditional GLNMF, Vertex Component Analysis (VCA) and Minimum Volume Constrained Nonnegative Matrix Factorization (MVCNMF). This proposed algorithm improves the accuracy of unmixing effectively with good robustness to noise.