Aiming at the high-dimensional and small sample problems of functional Magnetic Resonance Imaging (fMRI) data, a Semi-Supervised Feature Selection by Spectral Clustering (SS-FSSC) model was proposed. Firstly, the prior brain region template was used to extract the time series signal. Then, the Pearson correlation coefficient and the Order Statistics Correlation Coefficient (OSCC) were selected to describe the functional connection features between the brain regions, and spectral clustering was performed to the features. Finally, the feature importance criterion based on Constraint score was adopted to select feature subsets, and the subsets were input into the Support Vector Machine (SVM) classifier for classification. By 100 times of five-fold cross-validation on the COBRE (Center for Biomedical Research Excellence) schizophrenia public dataset in the experiments, it is found that when the number of retained features is 152, the highest average accuracy of the proposed model to schizophrenia is about 77%, and the highest accuracy of the proposed model to schizophrenia is 95.83%. Experimental result analysis shows that by only retaining 16 functional connection features for classifier training, the model can stably achieve an average accuracy of more than 70%. In addition, in the results obtained by the proposed model, Intracalcarine Cortex has the highest occurrence frequency among the 10 brain regions corresponding to the functional connections, which is consistent to the existing research state about schizophrenia.

Aiming at the problem of the present mainstream adversarial attack algorithm that the attack invisibility is reduced by disturbing the global image features, an untargeted attack algorithm named PS-MIFGSM (Perceptual-Sensitive Momentum Iterative Fast Gradient Sign Method) was proposed. Firstly, the areas of the image focused by Convolutional Neural Network (CNN) in the classification task were captured by using Grad-CAM algorithm. Then, MI-FGSM (Momentum Iterative Fast Gradient Sign Method) was used to attack the classification network to generate the adversarial disturbance, and the disturbance was applied to the focus areas of the image with the non-focus areas of the image unchanged, thereby, a new adversarial sample was generated. In the experiment, based on three image classification models Inception_v1, Resnet_v1 and Vgg_16, the effects of PS-MIFGSM and MI-FGSM on single model attack and set model attack were compared. The results show that PS-MIFGSM can effectively reduce the difference between the real sample and the adversarial sample with the attack success rate unchanged.

The equivalent channel matrix dimension of Generalized Frequency Division Multiplexing with Multiple Input Multiple Output (MIMO-GFDM) system is very large, and the traditional Multiple Input Multiple Output (MIMO) detection algorithm has high complexity and poor performance. Aiming at those problems, Reactive Tabu Search (RTS) detection algorithm in massive MIMO systems was applied to MIMO-GFDM system, and the high complexity problem of the initial value in RTS algorithm was also solved. Firstly, by using the positive definite symmetry of the matrix used in Minimum Mean Squared Error (MMSE) detection algorithm, Cholesky decomposition was applied to the matrix, and Sherman-Morrison formula was combined to iteratively calculate the initial value, reducing high complexity of the initial value inversion. Then, with the result of the improved MMSE detection as the initial value of RTS algorithm, the optimum solution was searched globally from the initial value. Finally, the iteration numbers and Bit Error Rate (BER) performance were researched through simulations. Theoretical analysis and simulation results show that, in MIMO-GFDM, the improved RTS signal detection algorithm has much lower BER than traditional signal detection algorithms. In 4 Quadrature Amplitude Modulation (4QAM), the RTS algorithm has approximately 6 dB lower signal-to-noise performance gain than MMSE detection (when BER is 10 ^-3). In 16QAM, the RTS algorithm has approximately 4 dB lower signal-to-noise performance gain than MMSE detection (when BER is 10 ^-2). Compared with the traditional RTS algorithm, the proposed algorithm has lower complexity without affecting the BER performance.

Because of complexity of background replacement, the traditional method is inefficient and the accuracy is difficult to improve. To solve these problems, a fast image background replacement method based on dilated convolution, called FABRNet, was proposed. First of all, the first three parts of VGG (Visual Geometry Group network) model were used for convolution and pooling operations of input images. Secondly, the combination of multiple sets of dilated convolutions were embedded into convolution neural network to make the network have a large and fine enough receptive field; meanwhile, the residual network structure was used to ensure the accuracy of the information distribution in the convolution process. Finally, the image was scaled to the original size and output by bilinear interpolation algorithm. Compared with three classical methods such as KNN (K-Nearest Neighbors) matting, Portrait matting and Deep matting, the experimental results show that FABRNet can effectively complete the background automatic replacement, and has advantages in running speed.

In Spatial Modulation (SM) system, the Maximum Likelihood (ML) detection algorithm with the best performance has high complexity, while the complexity can be reduced by M-algorithm based on QR decomposition (QRD-M) of channel matrix. However, when the traditional QRD-M algorithm is used, fixed M nodes were chosen at each layer, which leads to additional computation. Therefore, for the problem of the traditional QRD-M algorithm, a Low-Complexity QR-Decomposition M-algorithm with dynamic value of M (LC-QRD-dM) was proposed. In LC-QRD-dM, by comparing the designed threshold with the cumulative branch metrics, the number of reserved nodes that does not exceed M was adaptively selected at each layer, thus reducing the computational complexity with the cost of a small amount of performance. Then, concerning the high bit error rate of LC-QRD-dM with deep channel fading, QR-Decomposition M-algorithm with dynamic value of M based on Channel State (CS-QRD-dM) was further proposed. Based on the principle of LC-QRD-dM, the number of reserved nodes that do not less than M was selected by the threshold value at each layer when the Signal-to-Noise Ratio (SNR) is not high; and the number of reserved nodes that do not exceed M was selected by the threshold value at each layer when the SNR is high. Theoretic analysis and simulation results show that, compared with the traditional QRD-M algorithm, CS-QRD-dM achieves about 1.3 dB SNR advantage (when the bit error rate is 10 ^-2) at low SNR, which can significantly improve the detection performance at the cost of small complexity increase; and its detection performance and complexity are the same as LC-QRD-dM at high SNR.

Due to the weak development ability, Particle Swarm Optimization (PSO) algorithms have the shortages of low precision and slow convergence. Comparatively weak exploration ability of Differential Evolution (DE) algorithm, might further lead to a trap in the local extremum. A particle swarm-differential evolution fusion algorithm based on fuzzy Gaussian learning strategy was proposed. On the basis of the standard particle swarm algorithm, the elite particle population was selected, and the fusion mechanism of elite particle swarm-evolution was constructed by using mutation, crossover and selection evolution operators to improve particle diversity and convergence. A fuzzy Gaussian learning strategy according with human thinking characteristics was introduced to improve particle optimization ability, and further generate an elite particle swarm and differential evolution fusion algorithm based on fuzzy Gaussian learning strategy. Nine benchmark functions were calculated and analyzed in this thesis. The results show that the mean values of the functions Schwefel.1.2, Sphere, Ackley, Griewank and Quadric Noise are respectively 1.5E-39, 8.5E-82, 9.2E-13, 5.2E-17, 1.2E-18, close to the minimum values of the algorithm. The convergences of Rosenbrock, Rastrigin, Schwefel and Salomon functions are 1~3 orders of magnitude higher than those of four contrast particle swarm optimization algorithms. At the same time, the convergence of the proposed algorithm is 5%-30% higher than that of the contrast algorithms. The proposed algorithm has significant effects on improving convergence speed and precision, and has strong capabilities in escaping from the local extremum and global searching.

In the distributed storage system, Disk space Utilization (DU) is generally used to measure the load balance of each storage node. When given the equal disk space utilization to each node, the balance of storage load is achieved in the whole distributed storage system. However, in practice, the storage node with relatively low disk I/O speed and reliability becomes a bottleneck for the performance of data I/O in the whole storage system. Therefore in heterogeneous distributed storage system and specially the system which has great differences in disk I/O speed and reliability of each storage node, the speed of data I/O is definitely limited when disk space utilization is the only evaluation criteria of storage load balance. A new idea based on read-write efficiency was proposed to measure the storage load balance in the distributed storage system. According to the definition of Storage Entropy (SE) given by the theory of load balance and entropy, a kind of load balance algorithm based on SE was proposed. With system load and single node load determination as well as load shifting, the quantitative adjustment for storage load of the distributed storage system was achieved. The proposed algorithm was tested and compared with the load balance algorithm based on disk space utilization. Experimental results show that the proposed algorithm can balance storage load well in the distributed storage system, which effectively restrains the system load imbalance and improves the overall efficiency of reading and writing of the distributed storage system.

A Byzantine Fault Tolerance (BFT) schema was proposed to solve the problem that most Byzantine fault tolerance protocols were not suitable for service-oriented computing and other emerging computing models because of the assumption that the services were passive and independent. Service replicas were created on both sides of service requester and service provider. State machine replication algorithm was used to reach agreement on the ID and the content of the request after three rounds of communications among service replicas. After receiving a request, replicas submitted the request to upper application logic. After receiving the reply, replicas on service requester reached agreement on the ID and the content of the reply after three rounds of communications among services replicas and then accepted the reply. To deal with the problem of only having simple correctness reasoning and lacking of formal verification, an I/O automaton was used to model the protocol and simulation relation method was used as a tool to prove the correctness of the protocol more formally and rigorously. A highly abstract simple I/O automata S was constructed, which meeted safety and liveness. The parties of the protocol were broken down into several simple member I/O automata including front-end automaton, back-end automaton and multicast channel automaton. It is proved that the system composed of member I/O automata realizes the automata S. I/O automaton can accurately describe the protocol, which makes the correctness proof more standard than inductive reasoning.

Concerning the problem that traditional recommendation algorithm ignores the time factors, according to the similarity of individuals' short-term behavior, a calculation method of item correlation by using time decay function based on users' interest was proposed. And based on this method, a new item similarity was proposed. At the same time, the TItemRank algorithm was proposed which is an improved ItemRank algorithm by combining with the user interest-based item correlation. The experimental results show that: the improved algorithms have better recommendation effects than classical ones when the recommendation list is small. Especially, when the recommendation list has 20 items, the precision of user interest-based item similarity is 21.9% higher than Cosin similarity and 6.7% higher than Jaccard similarity. Meanwhile, when the recommendation list has 5 items, the precision of TItemRank is 2.9% higher than ItemRank.

为了克服传统密度估计方法受限于算法配置工作量高、高等级密度样本数量有限等因素无法大规模应用的缺点,提出一种基于监控视频的全景密度估计方法。首先,通过自动构建场景的权重图消除成像过程中射影畸变造成的影响,该过程针对不同的场景自动鲁棒地学习出对应的权值图,从而有效降低算法配置工作量;其次,利用仿真模拟方法通过低密度等级样本构建大量高密度等级样本;最后,提取训练样本的面积、周长等特征用于训练支持向量回归机(SVR)来预测每个场景的密度等级。在测试过程中,还通过二维图像与全景地理信息系统(GIS)地图的映射,实时展示全景密度分布情况。在北京北站广场地区的深度应用结果表明,所提全景密度估计方法可以准确、快速、有效地估计复杂场景中人群密度动态变化。

Most existing cloud storage systems are based on the model, which leads to a full dataset scan for multi-dimensional queries and low query efficiency. A KD-tree and R-tree based multi-dimensional cloud data index named KD-R index was proposed. KD-R index adopted two-layer architecture: a KD-tree based global index was built in the global server and R-tree based local indexes were built in local server. A cost model was used to adaptively select appropriate R-tree nodes to publish into global KD-tree index. The experimental results show that, compared with R-tree based global index, KD-R index is efficient for multi-dimensional range queries, and it has high availability in the case of server failure.

In this paper, the grey constrained nonlinear programming problems were investigated. With the help of mean, this paper firstly transformed the original grey optimization problem into a determinate constrained nonlinear programming problem. Then, based on the estimation of distribution algorithm, a stochastic search method was developed to solve the determinate constrained nonlinear programming problem. The key technique of the proposed method was explained in detail and the steps of the proposed method were described concretely. Finally, the elementary numerical examples show the proposed stochastic search method is feasible and effective.

Under the condition of coherent multipath environment, the algorithms of temporal smoothing and reconstruction of Toeplitz matrix were adopted respectively to estimate the DoAs (Direction of Arrivals) of complex coherent signals, furthermore these two algorithms were compared through mathematical analysis and computer simulation, and the following conclusion could be obtained: Both of these two algorithms for DoA estimation of complex coherent signals have the effectivity, the performance of algorithm based on reconstruction of Toeplitz matrix is relatively better but it will lose the array aperture, the algorithm based on temporal smoothing will not lose array aperture and can estimate M-1 coherent multi-paths (M is the number of array elements), however, it has a slightly larger calculation load.

The chip of smart card stores private sensitive defense data of user, which concerns the benefits of customers and mobile phone service providers. To enhance the security and correctness of data with a limited capacity chip is a critical problem. The security module architecture of chip operating system in evolution data of CDMA2000 was designed, and the network access authentication algorithm was put forward with a space optimization method. The file access control, message authentication and data encryption for communication were designed, and the power down protection while modifying flash memory or update several files was introdued. Experimental results prove that the system can run stably, effectively and safely, and ensures the integrity, validity, authenticity of data in storage and transmission between smart card and network.