Existing learning-based single-image deraining networks mostly focus on the effect of rain streaks in rainy images on visual imaging, while ignoring the effect of fog on visual imaging due to the increase of humidity in the air in rainy environments, thus causing problems such as low generation quality and blurred texture detail information in the derained images. To address these problems, an asymmetric unsupervised end-to-end image deraining network model was proposed. It mainly consists of rain and fog removal network, rain and fog feature extraction network and rain and fog generation network, which form two different data domain mapping conversion modules: Rain-Clean-Rain and Clean-Rain-Clean. The above three sub-networks constituted two parallel transformation paths: the rain removal path and the rain-fog feature extraction path. In the rain-fog feature extraction path, a rain-fog-aware extraction network based on global and local attention mechanisms was proposed to learn rain-fog related features by using the global self-similarity and local discrepancy existing in rain-fog features. In the rain removal path, a rainy image degradation model and the above extracted rain-fog related features were introduced as priori knowledge to enhance the ability of rain-fog image generation, so as to constrain the rain-fog removal network and improve its mapping conversion capability from rain data domain to rain-free data domain. Extensive experiments on different rain image datasets show that compared to state-of-the-art deraining method CycleDerain, the Peak Signal-to-Noise Ratio (PSNR) is improved by 31.55% on the synthetic rain-fog dataset HeavyRain. The proposed model can adapt to different rainy scenarios, has better generalization, and can better recover the details and texture information of images.
Aiming at the high computational complexity and large memory consumption of the existing super-resolution reconstruction networks, a lightweight image super-resolution reconstruction network based on Transformer-CNN was proposed, which made the super-resolution reconstruction network more suitable to be applied on embedded terminals such as mobile platforms. Firstly, a hybrid block based on Transformer-CNN was proposed, which enhanced the ability of the network to capture local-global depth features. Then, a modified inverted residual block, with special attention to the characteristics of the high-frequency region, was designed, so that the improvement of feature extraction ability and reduction of inference time were realized. Finally, after exploring the best options for activation function, the GELU (Gaussian Error Linear Unit) activation function was adopted to further improve the network performance. Experimental results show that the proposed network can achieve a good balance between image super-resolution performance and network complexity, and reaches inference speed of 91 frame/s on the benchmark dataset Urban100 with scale factor of 4, which is 11 times faster than the excellent network called SwinIR (Image Restoration using Swin transformer), indicates that the proposed network can efficiently reconstruct the textures and details of the image and reduce a significant amount of inference time.
In relation extraction tasks, distant supervision is a common method for automatic data labeling. However, this method will introduce a large amount of noisy data, which affects the performance of the model. In order to solve the problem of noisy data, a relation extraction method based on negative training and transfer learning was proposed. Firstly, a noisy data recognition model was trained through negative training method. Then, the noisy data were filtered and relabeled according to the predicted probability value of the sample, Finally, a transfer learning method was used to solve the domain shift problem existing in distant supervision tasks, and the precision and recall of the model were further improved. Based on Thangka culture, a relation extraction dataset with national characteristics was constructed. Experimental results show that the F1 score of the proposed method reaches 91.67%, which is 3.95 percentage points higher than that of SENT (Sentence level distant relation Extraction via Negative Training) method, and is much higher than those of the relation extraction methods based on BERT (Bidirectional Encoder Representations from Transformers), BiLSTM+ATT(Bi-directional Long Short-Term Memory and Attention), and PCNN (Piecewise Convolutional Neural Network).
As a semantic knowledge base, Knowledge Graph (KG) uses structured triples to store real-world entities and their internal relationships. In order to infer the missing real triples in the knowledge graph, considering the strong triple representation ability of relational memory network and the powerful feature processing ability of capsule network, a knowledge graph embedding model of capsule network based on relational memory was proposed. First, the encoding embedding vectors were formed through the potential dependencies between encoding entities and relationships and some important information. Then, the embedding vectors were convolved with the filter to generate different feature maps, and the corresponding capsules were recombined. Finally, the connections from the parent capsule to the child capsule was specified through the compression function and dynamic routing, and the confidence coefficient of the current triple was estimated by the inner product score between the child capsule and the weight. Link prediction experimental results show that compared with CapsE model, on the Mean Reciprocal Rank (MRR) and Hit@10 evaluation indicators, the proposed model has the increase of 7.95% and 2.2 percentage points respectively on WN18RR dataset, and on FB15K-237 dataset, the proposed model has the increase of 3.82% and 2 percentage points respectively. Experiments results show that the proposed model can more accurately infer the relationship between the head entity and the tail entity.
In order to solve the occlusion problem of student expression recognition in complex classroom scenes, and give full play to the advantages of deep learning in the application of intelligent teaching evaluation,a student expression recognition model and an intelligent teaching evaluation algorithm based on deep attention network in classroom teaching videos were proposed. A video library, an expression library and a behavior library for classroom teaching were constructed, then, multi-channel facial images were generated by cropping and occlusion strategies. A multi-channel deep attention network was built and self-attention mechanism was used to assign different weights to multiple channel networks. The weight distribution of each channel was restricted by a constrained loss function, then the global feature of the facial image was expressed as the quotient of the sum of the product of the feature times its attention weight of each channel divided by the sum of the attention weights of all channels. Based on the learned global facial feature, the student expressions in classroom were classified, and the student facial expression recognition under occlusion was realized. An intelligent teaching evaluation algorithm that integrates the student facial expressions and behavior states in classroom was proposed, which realized the recognition of student facial expressions and intelligent teaching evaluation in classroom teaching videos. By making experimental comparison and analysis on the public dataset FERplus and self-built classroom teaching video datasets, it is verified that the student facial expressions recognition model in classroom teaching videos achieves high accuracy of 87.34%, and the intelligent teaching evaluation algorithm that integrates the student facial expressions and behavior states in classroom achieves excellent performance on the classroom teaching video dataset.
In view of the problems that classroom teaching scene is obscured seriously and has numerous students, the current video action recognition algorithm is not suitable for classroom teaching scene, and there is no public dataset of student classroom action, a classroom teaching video library and a student classroom action library were constructed, and a real-time multi-person student classroom action recognition algorithm based on deep spatiotemporal residual convolution neural network was proposed. Firstly, combined with real-time object detection and tracking to get the real-time picture stream of each student, and then the deep spatiotemporal residual convolution neural network was used to learn the spatiotemporal characteristics of each student’s action, so as to realize the real-time recognition of classroom behavior for multiple students in classroom teaching scenes. In addition, an intelligent teaching evaluation model was constructed, and an intelligent teaching evaluation system based on the recognition of students’ classroom actions was designed and implemented, which can help improve the teaching quality and realize the intelligent education. By making experimental comparison and analysis on the classroom teaching video dataset, it is verified that the proposed real-time classroom action recognition model for multiple students in classroom teaching video can achieve high accuracy of 88.5%, and the intelligent teaching evaluation system based on classroom action recognition has also achieved good results in classroom teaching video dataset.
Concerning the problem that the manual design of airport arrival procedures is time consuming and it is difficult to optimize the path length quantitatively, a three-dimensional automatic optimization design method of multiple arrival procedures was proposed. Firstly, based on the specifications of RNAV (Rules for implementation of area NAVigation), the geometric configuration and the merging structure of the arrival procedures were modeled. Then, considering airport layout and aircraft operation constraints such as obstacle avoidance and route separation, with the goal of minimizing the total length of arrival procedures, a complete mathematical model was established. Finally, a hybrid algorithm based on simulated annealing algorithm and improved A* algorithm was developed to automatically optimize the merging structure of arrival procedures. Simulation results show that, in the experiment based on Sweden Arlanda Airport, compared with the existing related integer programming method, the hybrid simulated annealing algorithm can shorten the total path length by 3% and reduce the computing time by 87%. In the experiment based on Shanghai Pudong Airport, compared with the actual arrival procedures, the length of the routes designed by the proposed algorithm is reduced by 6.6%. These results indicate that the proposed algorithm can effectively design multiple three-dimensional arrival procedures, and can provide preliminary decision support for the procedure designers.
Most traditional community detection methods are limited to single relational network, and their applicability and accuracy are relatively poor. In order to solve the problems, a community detection method for multiple relationship networks was proposed. Firstly, for modeling the multiple relational network, the third-order adjacency tensor was used, in which each slice of the tensor represented an adjacency matrix corresponding to a type of relationship between participants. From the perspective of data representation, by interpreting the multiple relational network as a third-order tensor is helpful to use the factorization method as a learning method. Then, RESCAL decomposition was used as a relational learning tool to reveal the unique implicit representation of participants. Finally, the evolutionary K-means clustering algorithm was applied to the results obtained in the previous step to determine the community structure in multiple dimensions. The experiments were conducted on a synthetic dataset and two public datasets. The experimental results show that, compared with Contextual Information-based Community Detection (CICD) method, Memetic method and Local Spectral Clustering (LSC) method, the proposed method has the purity at least 5 percentage points higher, the Overlapping Normalized Mutual Information (ONMI) at least 2 percentage points higher, and the F score at least 3 percentage points higher. And it is proved that the proposed method has fast convergence speed.
Feature selection plays an important role in the classification accuracy and generalization performance of classifiers. The existing multi-label feature selection algorithms mainly use the maximum relevance and minimum redundancy criterion to perform feature selection in all feature sets without considering expert features, therefore, the multi-label feature selection algorithm has the disadvantages of long running time and high complexity. Actually, in real life, experts can directly determine the overall prediction direction based on a few or several key features. Paying attention to and extracting this information will inevitably reduce the calculation time of feature selection and even improve the performance of classifier. Based on this, a multi-label feature selection algorithm based on conditional mutual information of expert feature was proposed. Firstly, the expert features were combined with the remaining features, and then the conditional mutual information was used to obtain a feature sequence of strong to weak relativity with the label set. Finally, the subspaces were divided to remove the redundant features. The experimental comparison was performed to the proposed algorithm on 7 multi-label datasets. Experimental results show that the proposed algorithm has certain advantages over the other feature selection algorithms, and the statistical hypothesis testing and the stability analysis further illustrate the effectiveness and the rationality of the proposed algorithm.
The traditional 3D face recognition and classification algorithms require multiple samples for training. However, the recognition performance will be seriously degraded on single sample training. To resolve the above problem, Fuzzy Adaptive Resonance theory MAP (Fuzzy ARTMAP) algorithm was used to classify the 3D face database. Firstly, the features of the 3D face deep image were extracted by Local Binary Pattern (LBP). Then the frequency-domain features of LBP features extracted by Log-Gabor wavelet were used as the input vectors for training. Finally the set of feature vectors were sent to Fuzzy ARTMAP classifier for recognition. The experiments compared with Probabilistic Neural Network (PNN) and Extreme Learning Machine (ELM) were conducted on FRGC v2.0 database, the recognition rate of the proposed algorithm reached 87.15%, the classifier training time was 24.88s, the matching time of single sample to single registered face was 0.0015s, and the searching time of a new face sample in the database was 1.08s. The experimental results show that the proposed method outperforms to PNN and ELM, it achieves a higher recognition rate with shorter training time, and has stable time performance with strong controllability.
For the vital arc problem of maximum dynamic flow in time-capacitated network, the classic Ford-Fulkerson maximum dynamic flow algorithm was analyzed and simplified. Thus an improved algorithm based on minimum cost augmenting path to find the vital arc of the maximum dynamic flow was proposed. The shared minimum augmenting paths were retained when computing maximum dynamic flow in new network and the unnecessary computation was removed in the algorithm. Finally, the improved algorithm was compared with the original algorithm and natural algorithm. The numerical analysis shows that the improved algorithm is more efficient than the natural algorithm
This study begins with a Location-free Topology Construction (LTC) algorithm to construct a virtual backbone based on connected dominating tree, in consideration of the characteristics of densely deployed Wireless Sensor Network (WSN). On this basis, the energy consumption of backbone nodes and the data transmission delay were analyzed. Then, a density control factor and a rate control factor for data transmission were introduced to balance energy consumption of the virtual backbone construction, and a Location-free and Energy-balanced Topology Control (LETC) algorithm, as an extension of LTC, was proposed. In accordance with the amount of data transmission in difference regions, LETC adjusted arrangement density of virtual backbone nodes, and increased the node transmission rate of nodes to reduce network latency. Both theoretical analysis and simulation results demonstrate that LETC algorithm can effectively balance energy consumption, extending the network lifetime by 24.1%, and reducing the transimisson delay by 28.1% compared to LTC. 〖BP(〗In the case of data transmission delay, the reduction achieved is up to 28.1%.〖BP)〗
For the problems of long runtime, ignoring the difference between classes of sample, the paper put forward an algorithm called Global Weighted Sparse Locality Preserving Projection (GWSLPP) based on Sparse Preserving Projection (SPP). The algorithm made sample have good identification ability while maintaining the sparse reconstruction relations of the samples. The algorithm processed the samples though sparse reconstruction, then made the sample on the projection and maximized the divergence between classes of sample. It got the projection and classified the sample at last. The algorithm made the experiments on FERET face database and YALE face database. The experimental results show the GWSLPP algorithm is superior to the Locality Preserving Projection (LPP), SPP and FisherFace algorithm in both execution time and recognition rate. The execution time is only 25s and the recognition rate can reach more than 95%. The experimental data prove the effectiveness of the algorithm.
In view of the sensor network boundary identification in 3D environment, this paper presented a distributed algorithm for boundary node identification based on flipping finite plane. Based on three known adjacent nodes, the finite plane took each edge of triangle as axis to flip, the first node scanned is the new boundary node, this node and two nodes on the axis construct a new triangle. Above process was carried out iteratively, eventually the boundary contour was got and the boundary nodes were identified. The experimental result shows that, compared with Alpha-shape3D algorithm, the proposed algorithm can greatly reduce the redundant boundary nodes.