For the shortcomings of poor interpretation ability and instability in neural network training, a Chambolle- Pock (CP) algorithm optimized denoising network based on Total Variational (TV) regularization, CPTV-Net, was proposed to solve the denoising problem of Low-Dose Computed Tomography (LDCT) images. Firstly, the TV constraint term was introduced into the L1 regularization term model to preserve the structural information of the image. Secondly, the CP algorithm was used to solve the denoising model and obtain specific iterative steps to ensure the convergence of the algorithm. Finally, the shallow CNN (Convolutional Neural Network) was used to learn the iterative formula of the primal dual variables of the linear operation. The neural network was used to calculate the solution of the model, and the network parameters were collected to optimize the combined data. The experimental results on simulated and real LDCT datasets show that compared with five advanced denoising methods such as REDCNN (Residual Encoder-Decoder Convolutional Neural Network) and TED-Net (Transformer Encoder-decoder Dilation Network), CPTV-Net has the best Peak Signal-to-Noise Ratio (PSNR), Structural SIMilarity (SSIM), and Visual Information Fidelity (VIF) evaluation values, and can generate LDCT images with significant denoising effect and the most details preserved.
A supervisor of a discrete-event system can prohibit controllable events to ensure the safety and liveness specifications of the system. However, the supervisor does not actively select the controllable events that are allowed to occur, so it is possible that several controllable events occur simultaneously. In practice, such as traffic scheduling and robot path planning, the system is required to allow at most one controllable event to occur in each state. In response to the above problem, an optimal mechanism was introduced to quantify control cost, and an optimal supervisory control algorithm of discrete-event systems was proposed, which not only can guarantee the safety and liveness of the system, but also can minimize the cumulative cost of event execution. Firstly, the automata model of controlled system and behavioral constraints was given, and a nonblocking supervisor with maximum allowable behaviors was solved on the basis of the supervisory control theory of Ramadge and Wonham. Secondly, a cost function was defined to assign the corresponding cost to the execution of each event in the supervisor. Finally, an optimal directed supervisor was calculated iteratively based on dynamic programming to achieve the goals of at most one controllable event occurring in each state and minimizing the cumulative cost of event execution. To verify the effectiveness and correctness of the proposed algorithm, a one-way train guideway example and a multi-track train control example were used. For the above two examples, the cumulative cost of the event execution required for the directed supervisor solved by the proposed algorithm to reach the target state is 26.0 and 14.0 respectively, which is lower than the 27.5 and 16.0 of greedy algorithm and the 26.5 and 14.0 of Q-learning.
To solve the long training time and slow convergence problems when applying SAC (Soft Actor-Critic) algorithm to the local path planning of mobile robots, a PER-SAC algorithm was proposed by introducing the Prioritized Experience Replay (PER) technique. Firstly, to improve the convergence speed and stability of the robot training process, a priority strategy was applied to extract samples from the experience pool instead of the traditional random sampling and the network prioritized the training of samples with larger errors. Then, the calculation of Temporal-Difference (TD) error was optimized, and the training deviation was reduced. Next, the transfer learning was used to train the robot from a simple environment to a complex one gradually in order to improve the training speed. In addition, an improved reward function was designed to increase the intrinsic reward of robots, and therefore, the sparsity problem of environmental reward was solved. Finally, the simulation was carried out on the ROS (Robot Operating System) platform, and the simulation results show that PER-SAC algorithm outperforms the original algorithm in terms of convergence speed and length of the planned path in different obstacle environments. Moreover, the PER-SAC algorithm can reduce the training time and is significantly better than the original algorithm on path planning performance.
2D/3D medical image registration is a key technology in 3D real-time navigation of orthopedic surgery. However, the traditional 2D/3D registration methods based on optimization iteration require multiple iterative calculations, which cannot meet the requirements of doctors for real-time registration during surgery. To solve this problem, a pose regression network based on autoencoder was proposed. In this network, the geometric pose information was captured through hidden space decoding, thereby quickly regressing the 3D pose of preoperative spine pose corresponding to the intraoperative X-ray image, and the final registration image was generated through reprojection. By introducing new loss functions, the model was constrained by “Rough to Fine” combined registration method to ensure the accuracy of pose regression. In CTSpine1K spine dataset, 100 CT scan image sets were extracted for 10-fold cross-validation. Experimental results show that the registration result image generated by the proposed model has the Mean Absolute Error (MAE) with the X-ray image of 0.04, the mean Target Registration Error (mTRE) with the X-ray image of 1.16 mm, and the single frame consumption time of 1.7 s. Compared to the traditional optimization based method, the proposed model has registration time greatly shortened. Compared with the learning-based method, this model ensures a high registration accuracy with quick registration. Therefore, the proposed model can meet the requirements of intraoperative real-time high-precision registration.
In the research of image classification tasks in deep learning, the phenomenon of adversarial attacks brings severe challenges to the secure application of deep learning models, which arouses widespread attention of researchers. Firstly, around the adversarial attack technologies for generating the adversarial perturbations, the important white-box adversarial attack algorithms in the image classification tasks were introduced in detail, and the advantages and disadvantages of different attack algorithms were analyzed. Then, from three realistic application scenarios: mobile application, face recognition and autonomous driving, the application status of the white-box adversarial attack technologies was illustrated. Additionally, some typical white-box adversarial attack algorithms were selected to perform experiments on different target models, and the experimental results were analyzed. Finally, the white-box adversarial attack technologies were summarized, and their valuable research directions were prospected.
Aiming at the problem of insufficient representation ability of features extracted by the existing vehicle re-identification methods based on convolution Neural Network (CNN), a vehicle re-identification method based on the combination of wavelet features and attention mechanism was proposed. Firstly, the single-layer wavelet module was embedded in the convolution module to replace the pooling layer for subsampling, thereby reducing the loss of fine-grained features. Secondly, a new local attention module named Feature Extraction Module (FEM) was put forward by combining Channel Attention (CA) mechanism and Pixel Attention (PA) mechanism, which was embedded into CNN to weight and strengthen the key information. Comparison experiments with the benchmark residual convolutional network ResNet-50 and ResNet-101 were conducted on VeRi dataset. Experimental results show that increasing the number of wavelet decomposition layers in ResNet-50 can improve mean Average Precision (mAP). In the ablation experiment, although ResNet-50+Discrete Wavelet Transform (DWT) has the mAP reduced by 0.25 percentage points compared with ResNet-101, it has the number of parameters and computational complexity lower than those of ResNet-101, and has the mAP, Rank-1 and Rank-5 higher than those of ResNet-50 without DWT, verifying that the proposed model can effectively improve the accuracy of vehicle retrieval in vehicle re-identification.
Aiming at the problem that the nearest neighbor parameters need to be set manually in density peak clustering algorithm based on shared nearest neighbor, a density peak clustering algorithm based on adaptive nearest neighbor parameters was proposed. Firstly, the proposed nearest neighbor parameter search algorithm was used to automatically obtain the nearest neighbor parameters. Then, the clustering centers were selected through the decision diagram. Finally, according to the proposed allocation strategy of representative points, all sample points were clustered through allocating the representative points and the non-representative points sequentially. The clustering results of the proposed algorithm was compared with those of the six algorithms such as Shared-Nearest-Neighbor-based Clustering by fast search and find of Density Peaks (SNN?DPC), Clustering by fast search and find of Density Peaks (DPC), Affinity Propagation (AP), Ordering Points To Identify the Clustering Structure (OPTICS), Density-Based Spatial Clustering of Applications with Noise (DBSCAN), and K-means on the synthetic datasets and UCI datasets. Experimental results show that, the proposed algorithm is better than the other six algorithms on the evaluation indicators such as Adjusted Mutual Information (AMI), Adjusted Rand Index (ARI) and Fowlkes and Mallows Index (FMI). The proposed algorithm can automatically obtain the effective nearest neighbor parameters, and can better allocate the sample points in the edge region of the cluster.
Concerning the problems of artifacts and loss of image details in the analytically reconstructed image by time-domain filters, a new time-frequency domain Computed Tomography (CT) reconstruction algorithm based on Convolutional Neural Network (CNN) was proposed. Firstly, a filter network based on a convolutional neural network was constructed in the frequency domain to achieve the frequency-domain filtering of the projection data. Secondly, the back-projection operator was used to perform domain conversion on the frequency-domain filtered result to obtain a reconstructed image. A network was constructed in the image domain to process the image from the back-projection layer. Finally, a multi-scale structural similarity loss function was introduced on the basis of the minimum mean square error loss function to form a composite loss function, which reduced the blur effect of the neural network on the result image and preserved the details of the reconstructed image. The image domain network and the projection domain filter network worked together to finally get the reconstructed result. The effectiveness of the proposed algorithm was verified on the clinical dataset. Compared with the Filtered Back Projection (FBP) algorithm, the Total Variation (TV) algorithm and the image domain Residual Encoder-Decoder CNN (RED-CNN) algorithm, when the number of projections is respectively 180 and 90, the proposed algorithm achieved the reconstructed result image with highest Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity (SSIM), and the least Normalized Mean Square Error (NMSE).When the number of projections is 360,the proposed algorithm is second only to TV algorithm. The experimental results show that the proposed algorithm can improve the reconstructed image quality of CT image, and it is feasible and effective.
Aiming at the problem of big clustering error of the Sparse Subspace Clustering (SSC) methods, an SSC method based on random blocking was proposed. First, the original problem dataset was divided into several subsets randomly to construct several sub-problems. Then, after obtaining the coefficient matrices of several sub-problems by the sparse subspace Alternating Direction Method of Multipliers (ADMM) respectively, these coefficient matrices were expanded into coefficient matrices of the same size as the original problem and integrated into a coefficient matrix. Finally, a similarity matrix was calculated according to the coefficient matrix obtained by the integration, and the clustering result of the original problem was obtained by using the Spectral Clustering (SC) algorithm. The SSC method based on random blocking has the subspace clustering error reduced by 3.12 percentage points on average compared with the optional algorithm among SSC, Stochastic Sparse Subspace Clustering via Orthogonal Matching Pursuit with Consensus (S3COMP-C), scalable Sparse Subspace Clustering by Orthogonal Matching Pursuit (SSCOMP), SC and K-Means algorithms, and has all the mutual information, Rand index and entropy significantly better than comparison algorithms. Experimental results show that the SSC method based on random blocking can significantly reduce subspace clustering error, and improve the clustering performance.
The traditional wireless communication system is composed of transmitters and receivers. The information to be transmitted is transmitted through antenna after channel coding, modulation, and shaping. Due to the influence of factors such as channel fading, noise, and interference, signals arriving at the receiver will have serious distortion, and the receiver needs to recover original information from distorted signals as much as possible. To solve this problem, a Multi-Input Multi-Output (MIMO) intelligent receiver model based on multi-label classification neural network was proposed. In this model, Deep Neural Network (DNN) was used to replace the entire information recovery link of receiver from signals to information, and multi-label classification algorithm was used to replace multiple binary classifiers to achieve multi-bit information flow recovery. The training dataset has quadrature signals that contains two modulation modes including Binary Phase Shift Keying (BPSK) and Quadrature Phase Shift Keying (QPSK) as well as two coding modes of Hamming coding and cyclic coding. Experimental results show that under conditions such as noise, Rayleigh fading, and interference, when the Bit Error Rate (BER) of receiver using the traditional Alamouti decoding method is 1E-3, the intelligent receiver realizes the recovered information with the BER of 0. While maintaining the same BER performance, the proposed multi-label classification algorithm reduces the training time of each batch by about 4 min compared with the multiple binary classifier algorithms of the comparison model.
Aiming at the problem that the key entity information in the police field is difficult to recognize, a neural network model based on BERT (Bidirectional Encoder Representations from Transformers), namely BERT-BiLSTM-Attention-CRF, was proposed to recognize and extract related named entities, in the meantime, the corresponding entity annotation specifications were designed for different cases. In the model ,the BERT pre-trained word vectors were used to replace the word vectors trained by the traditional methods such as Skip-gram and Continuous Bag of Words (CBOW), improving the representation ability of the word vector and solving the problem of word boundary division in Chinese corpus trained by the character vectors. And the attention mechanism was used to improve the architecture of classical Named Entity Recognition (NER) model BiLSTM-CRF. BERT-BiLSTM-Attention-CRF model has an accuracy of 91% on the test set, which is 7% higher than that of CRF++ Baseline, and 4% higher than that of BiLSTM-CRF model. The F1 values of the entities are all higher than 0.87.
Accurate background model is the paramount base for object extracting and tracing. In response to swing objects which part quasi-periodically changed in intricate scene, based on multi-Gaussian background model, a new Quasi-Periodic Background Algorithm (QPBA) was proposed to suppress the swing objects and establish an accurate and stable background model. The specific process included: According to multi-Gaussian background model, the object classification in scene was set up, and the effect on Gaussian model's parameters caused by swing objects was analyzed. By using color distribution values as samples to establish Gaussian model to keep swing pixels, the swing model in swing pixels was integrated into background model with weight factors of occurrence frequency and time interval. Comparison among QPBA and the classical background modeling algorithms such as GMM (Gaussian Mixture Model), ViBe (Visual Background extractor) and CodeBook was put forward, and the results were assessed in aspects of quality, quantity and efficiency. It shows that QPBA has a more obvious suppression on swing objects, and its fall-out ratio is less than 1%, so that it can handle the scene with swing objects. At the same time, its correct detection number is consistent with other algorithms, thus the moving objects can be reserved perfectly. In addition, the efficiency of QPBA is high, and its resolving time is approximate to CodeBook, which can satisfy the requirements of real-time computation.
Aiming at the problem of high computational cost of the BlackBox of the engineering optimizations, a River-based Dynamic Service-oriented Optimization Computing Platform (R-DSOCP) was proposed to calculate the BlackBox in a distributed and parallel way. Firstly, the running pattern of BlackBox in the optimization algorithms was analyzed. Conforming to the dynamic service-oriented architecture and surrounding the functions of service release and lookup of River, the kernel services required for building R-DSOCP were designed. Secondly, an ACO-based BlackBox Schedule Problem (BSP) algorithm was devised. Depending on it, the scheduling service could not only choose the best computing services for BlackBox quickly but also balance the load of R-DSOCP. At Last, the experimental results show that the BlackBox can be parallel performed on the platform effectively after separating the BlackBox’s computation from the execution of the optimization algorithm. Comparing with a single computing machine, the average computing efficiency is advanced nearly n times. n is the parallel factor. Thus, with the help of High Performance Computing (HPC) technology, R-DSOCP can offer a feasible scheme for accelerating the optimization algorithm and reducing the computational expenses in the field of engineering optimization.
Concerning the present situation that Quality of Service (QoS) evaluation methods ignore the implicit service quality assessment and lead to inaccurate results, a service evaluation method that comprehensively considered explicit and implicit quality attributes was put forward. Explicit quality attributes were expressed in vector form, using service quality assessment model, after quantization, normalization, then evaluation values were calculated; and implicit quality attributes were expressed according to the evaluation on similar users' recommendation. The users' credibility and difference between old and new users were considered in the evaluation process. Finally the explicit and implicit quality evaluation was regarded as the QoS evaluation results. The experiments were performed in comparison with three algorithms by using one million Web Service QoS data. The simulation results show that the proposed method has certain feasibility and accuracy.
Concerning the low efficiency of present methods of IP lookup, a new data lookup algorithm based on Multi-Bit Priority Tries (MBPT) was proposed in this paper. By storing the prefixes with higher priority in dummy nodes of multi-bit tries in proper order and storing the prefixes for being extended in an auxiliary storage structure,this algorithm tried to make the structure find the longest matching prefix in the internal node instead of the leaf node. Meanwhile, the algorithm avoided the reconstruction of router-table when it needed to be updated. The simulation results show that the proposed algorithm can effectively minimize the number of memory accesses for dynamic router-table operations, including lookup, insertion and deletion, which significantly improves the speed of router-table lookup as well as update.