In order to address the failure of boundary protection measures caused by the evolution with complexity， dynamics and fragmentation of network architecture， and to cope with the challenge for network data security caused by the continuous emergence of vulnerabilities in non-autonomous controllable systems， software， hardware and cryptographic algorithms， the following tasks were performed. Firstly， a zero trust network architecture implementation model was designed on the basis of zero trust concept. Secondly， a zero trust network security protection framework was proposed， which integrated concept of zero trust security， Chinese cryptographic algorithm system， and trusted computing technology in links such as identity management and authentication， authorization and access， data processing and transmission， framework processes such as Chinese cryptographic certificate application and issuance， business data secure processing and transmission were designed， and functional components such as identity and access management module， terminal trusted network access proxy device were designed and implemented. Finally， a network platform based on the security protection framework was built， which provided new frameworks， technologies and tools for network data security protection and zero trust security practices. Security analysis and performance test results show that with the proposed platform， the signing and signature verification performance of the SM2 reaches 1 118.72 and 441.43 times per second respectively， the encryption and decryption performance of SM4 reaches 10.05 MB/s and 9.96 MB/s respectively， and the secure data access/response performance reaches 7.23 MB/s， demonstrating that the proposed framework can provide stable support for data security.

Domain adaptation algorithms are widely used for cross-corpus speech emotion recognition. However， many domain adaptation algorithms lose the discrimination of target domain samples while pursuing the minimization of domain discrepancy， resulting in their presence at the decision boundary of the model in a high-density form， which degrades the performance of the model. Based on the above problem， a Decision Boundary Optimized Domain Adaptation （DBODA） method based cross-corpus speech emotion recognition was proposed. Firstly， the features were processed by using convolutional neural networks. Then， the features were fed into the Maximum Nuclear-norm and Mean Discrepancy （MNMD） module to maximize the nuclear norm of the sentiment prediction probability matrix of the target domain while reducing the inter-domain discrepancy， thereby enhancing the discrimination of the target domain samples and optimize the decision boundary. In six sets of cross-corpus experiments set up on the basis of Berlin， eNTERFACE and CASIA speech databases， the average recognition accuracy of the proposed method is 1.68 to 11.01 percentage points ahead of those of the other algorithms， indicating that the proposed model effectively reduces the sample density around the decision boundary and improves the prediction accuracy.

Aiming at the problem of artificial artifacts due to phase disorder in frequency-domain speech enhancement algorithms， which limits the denoising performance and decreases the speech quality， a speech enhancement algorithm based on Multi-Scale Ladder-type Time-Frequency Conformer Generative Adversarial Network （MSLTF-CMGAN） was proposed. Taking the real part， imaginary part and magnitude spectrum of the speech spectrogram as input， the generator first learned the local and global feature dependencies between temporal and frequency domains by using time-frequency Conformer at multiple scales. Secondly， the Mask Decoder branch was used to learn the amplitude mask， and the Complex Decoder branch was directly used to learn the clean spectrogram， and the outputs of the two decoder branches were fused to obtain the reconstructed speech. Finally， the metric discriminator was used to judge the scores of speech evaluation metrics， and high-quality speech was generated by the generator through minimax training. Comparison experiments with various types of speech enhancement models were conducted on the public dataset VoiceBank+Demand by subjective evaluation Mean Opinion Score （MOS） and objective evaluation metrics.Experimental results show that compared with current state-of-the-art speech enhancement method CMGAN （Comformer-based MetricGAN）， MSLTF-CMGAN improves MOS prediction of the signal distortion （CSIG） and MOS predictor of intrusiveness of background noise （CBAK） by 0.04 and 0.07 respectively， even though its Perceptual Evaluation of Speech Quality （PESQ） and MOS prediction of the overall effect （COVL） are slightly lower than that of CMGAN， it still outperforms other comparison models in several subjective and objective speech evaluation metrics.

For nine certificateless signature schemes proposed by Y L Tang， et al. （TANG Y L， WANG F F， YE Q， et al. Improved provably secure certificateless signature scheme. Journal of Beijing University of Posts and Telecommunications， 2016， 39（1）： 112-116）， firstly， the linearized equation analysis method was used. It was found that there was a linear relationship between the public keys in all schemes. This defect was exploited to complete a signature forgery attack on all schemes. Secondly， in order to break the linear relationship between the public keys， the method of modifying the parameters of hash function was used to improve the scheme， and the security of the improved scheme was proved under the random oracle model. Thirdly， a public key construction format of certificateless signature scheme was proposed. The signature scheme constructed by this format could not be attacked by adversaries using public key replacement. Finally， the efficiency of the improved scheme was compared with those of the existing certificateless signature schemes through simulation. Experimental results show that the improved scheme promotes the security without reducing the computational efficiency.

Rate-Distortion （R-D） optimization is a crucial technique in video encoders. However， the widely used independent R-D optimization is far from being global optimal. In order to further improve the compression performance of High Efficiency Video Coding （HEVC）， a two-pass encoding algorithm combined with both R-D dependency and R-D characteristic was proposed. Firstly， the current frame was encoded with the original method in HEVC， and the number of bits consumed by the current frame and the R-D model parameters of each Coding Tree Unit （CTU） were obtained. Then， combined with time domain dependent rate distortion optimization， the optimal Lagrange multiplier and quantization parameter for each CTU were determined according to the information including current frame bit budget and R-D model parameters. Finally， the current frame was re-encoded， where each CTU had different optimization goal according to its Lagrange multiplier. Experimental results show that the proposed algorithm achieves significant rate-distortion performance improvement. Specifically， the proposed algorithm saves 3.5% and 3.8% bitrate at the same coding quality， compared with the original HEVC encoder， under the coding configurations of low-delay B and P frames.

Aiming at problems such as high memory bandwidth demand, high computational complexity, long design and exploration cycle, and inter-layer computing delay of cascade convolution in two-dimensional Winograd convolution algorithm, a double-buffer 5×5 convolutional layer design method based on two-dimensional Winograd algorithm was proposed. Firstly, the column buffer structure was used to complete the data layout, so as to reuse the overlapping data between adjacent blocks and reduce the memory bandwidth demand. Then, the repeated intermediate calculation results in addition process of Winograd algorithm were precisely searched and reused to reduce the computational cost of addition, so that the energy consumption and the design area of the accelerator system were decreased. Finally, according to the calculation process of Winograd algorithm, the design of 6-stage pipeline structure was completed, and the efficient calculation for 5×5 convolution was realized. Experimental results show that, on the premise that the prediction accuracy of the Convolutional Neural Network (CNN) is basically not affected, this calculation method of 5×5 convolution reduces the multiplication computational cost by 83% compared to the traditional convolution, and has the acceleration ratio of 5.82; compared with the method of cascading 3×3 two-dimensional Winograd convolutions to generate 5×5 convolutions, the proposed method has the multiplication computational cost reduced by 12%, the memory bandwidth demand decreased by about 24.2%, and the computing time reduced by 20%.

In order to improve the exploration efficiency of the design space for floating-point expression, a design space exploration method based on heuristic search was proposed. The design space of non-dominated expression was explored firstly during each iteration. At the same time, the non-dominated expression and the dominated expression were added to the non-dominated list and the dominated list respectively. Then the expression in the dominated list was explored after the iteration, the non-dominated expression in the dominated list was selected, and the neighborhood of the non-dominated expression in the dominated list was explored. And the new non-dominated expression was added to the non-dominated list, effectively improving the diversity and randomness of the non-dominated expression. Finally, the non-dominated list was explored again to obtain the final equivalent expression and further improve the performance of optimal expression. Compared with the existing design space exploration methods for floating-point expression, the proposed method has the calculation accuracy increased by 2% to 9%, the calculation time reduced by 5% to 19% and the resource consumption reduced by 4% to 7%. Experimental results show that the proposed method can effectively improve the efficiency of design space exploration.

To solve the problems of too long time of pathfinding and collision and blocking during movement in real-time strategy games， a combined improved flow field pathfinding algorithm was proposed. Firstly， the red-black tree was used to store data to improve the speed of data access. Secondly， by using the penalty function， the calculation of the integration field cost was simplified through transforming the nonlinear partial differential equation problem into a linear unconstrained problem. Finally， a pre-adjacency node was introduced to generate the flow direction. Compared with the flow field pathfinding algorithm without improvement， the improved algorithm has the path calculation time reduced by 20%， and the average moving time is stable at 20 s. Experimental results show that the improved flow field pathfinding algorithm can effectively shorten the pathfinding time， increase the moving speed of Agents and improve the level of game artificial intelligence in real-time strategy games.

For the security problem that the SDN (Software Defined Network) controller can not guarantee the network strategy issued by itself to be correctly executed on the forwarding devices, a new forwarding path monitoring security solution was proposed. Firstly, based on the overall view capability of the controller, a path credential interaction processing mechanism based on OpenFlow was designed. Secondly, Hash chain and message authentication code were introduced as the key technologies for generating and processing the forwarding path credential information. Thirdly, on this basis, Ryu controller and Open vSwitch open-source switch were deeply optimized,with credential processing flow added, constructing a lightweight path security mechanism. The test results show that the proposed mechanism can effectively guarantee the security of data forwarding path, and its throughput consumption is reduced by more than 20% compared with SDNsec, which means it is more suitable for the network environment with complex routes, but its fluctuates of latency and CPU usage are more than 15%, which needs further optimization.

To solve the data leakage problem of data plane in Software Defined Network (SDN), a new data security processing mechanism based on OpenFlow protocol was proposed. Firstly, the flow table structure of OpenFlow protocol was reconstructed, the OpenFlow data security policies including safe matching fields, safe actions were designed and implemented. Secondly, a centralized management controller was designed to sense changes in the network in a timely manner through the development of multiple functional modules, which effectively controlled the global network, maintained and distributed data encryption/decryption keys and data security policies. Thirdly, the open virtual switch OVS (Open vSwitch) architecture was reconstructed deeply, the complete process including data security strategy matching and data security processing was designed, and the extraction interface of data payload information was programmed. Through the development of multiple functional modules, OVS can match the data packets according to the fine-grained granularity of data security policies, and perform complete data security processing operations on matched data packets. Finally, by building the hardware and software platform, the results of the encryption and decryption mechanisms, and the time delay, throughput and CPU utilization rate were tested and compared. The experimental results show that the proposed mechanism can accurately operate data encryption and decryption. The latency and throughput of the proposed mechanism are at normal levels, but its CPU usage rate is between 45% and 60%, which indicates that it needs to be optimized furtherer.

Aiming at the problem that multi-label classification algorithms cannot fully utilize the correlation between labels, a new multi-label classification algorithm based on gravitational model namely MLBGM was proposed, by establishing the positive and negative correlation matrices of labels to mine different correlations among labeled. Firstly, by traversing all samples in the training set, k nearest neighbors for each training sample were obtain. Secondly, according to the distribution of labels in all neighbors of each sample, positive and negative correlation matrices were established for each training sample. Then, the neighbor density and neighbor weights for each training sample were calculated. Finally, a multi-label classification model was constructed by calculating the interaction between data particles. The experimental results show that the HammingLoss of MLBGM is reduced by an average of 15.62% compared with 5 contrast algorithms that do not consider negative correlation between labels; on the MicroF1, the average increase is 7.12%; on the SubsetAccuracy, the average increase is 14.88%. MLBGM obtains effective experimental results and outperforms comparison algorithms as it makes full use of the different correlations between labels.

Aiming at the problems that traditional interference alignment algorithm based on the maximum Signal to Interference and Noise Ratio (SINR) in Multiple-Input Multiple-Output (MIMO) cognitive network is hard to converge when sending multiple data streams and the interference between them is prominent, an interference alignment algorithm that considers data stream interference and iterative limit was proposed. Firstly, the secondary users eliminated interference between primary users and secondary users through coding design. Then, when eliminating the interference between the primary users and the secondary users, the Generalized Rayleigh Entropy (GRE) was used to calculate the precoding and interference suppression matrix based on the maximum SINR algorithm according to channel reciprocity, and in the iterative process, each iteration always made precoding and interference suppression matrix firstly satisfy that the interference power in the expected signal space was minimal. Finally, combined with the MIMO interference channel between the secondary users, the interference channel between primary and secondary users and the necessity of interference alignment of secondary usernetwork, the secondary users' reachable upper bound of degree of freedom was deduced. The experimental results show that compared with the traditional maximum SINR algorithm, the proposed algorithm has no significant improvement in the total capacity of the secondary users when the signal to noise ratio is low, but with the increase of signal to noise ratio, the advantages of the proposed algorithm are more and more obvious. When convergence is reached, the iterative times of the proposed algorithm are reduced by 40% compared with the conventional maximum SINR algorithm. Therefore, the proposed algorithm can improve system capacity and accelerate convergence.

In order to solve the bottleneck problems of memory capacity and CPU processing speed when the traditional K-Medoids clustering algorithm is used to deal with massive data, based on the in-depth study of K-Medoids algorithm, a parallel K-Medoids algorithm based on the MapReduce programming model was proposed. The part of Map function is to calculate the distance of each data object to the center point of the cluster and (re)allocation of their respective clusters, and the part of Reduce function is to calculate the new center point of each cluster according to the intermediate results of the Map section. The experimental results show that the parallel K-Medoids algorithm in the Hadoop cluster based on the MapReduce running has good clustering results and scalability, and for large data sets, the algorithm may get close to linear speedup.

A new Web service composition method based on service communities and service chains was proposed in this paper to improve the time efficiency of service composition. In the method, a service network was constructed for the Web service collection, the service community discovery algorithm based on information center was applied to find service clubs in the service network, and then the community service chain discovery algorithm and Web service composition algorithm based on service chain were built. With these algorithms, all of service interface associations in a service club were changed into service chains, and the Web service composition process based on community service chains and Quality of Service (QoS) pruning was implemented. The experimental results indicate that, compared with the traditional service composition method based on graph depth traversal, the response time on five test sets in the service composition method with community service chains is on average improved by 42%, and up to 67%. Community service chains can effectively reduce the service search space for the current service request and improve the time efficiency of service composition.

In brain functional Magnetic Resonance Imaging (fMRI) experiment, motion correction is an important step in data preprocessing. Results of motion correction affect the follow up analysis such as detecting the functional activation area and functional connectivity. There are some simplified hypotheses for head motion in the common analysis software packages. Due to large volume of data, the correction error is also large. In order to reduce the correction error, a novel motion correction method was proposed based on local rigid transform. This method first used adjacent weighted slices to construct local volumetric data for each slice in a multi-slice echo planar imaging volume data, and then estimated the space position of each slice by the registration of local volumetric data using the modified Gauss-Newton optimization algorithm. Finally the image stack was re-sliced using Delaunay triangulation method. Results of implementation based on this method during phantom data and human vision experiments reveal that it is effective to reduce the correction error, which leads to accurate realignment.

In the course of tracking IR motive targets, a large number of false alarm signals might appear in the target detection stage of an IR ATR system because of the non-prediction of IR targets signature. So the false alarm signals had to be filtered in the stage of holding down background clutters. A new FLIR image segmentation technique was presented based on wavelet neural networks, aiming to fusing both local characteristic of wavelet time-frequency and adaptive learning by neural networks, and resulting in the powerful abilities of approximation and tolerate error in IR image segmentation. This new algorithm was applied in a FLIR-ATR system, and got favorable results in achieving IR target contours and damping background noises.