When the Unmanned Aerial Vehicle （UAV） cluster attacks ground targets， it will be divided into two formations： a strike UAV cluster that attacks the targets and a auxiliary UAV cluster that pins down the enemy. When auxiliary UAVs choose the action strategy of aggressive attack or saving strength， the mission scenario is similar to a public goods game where the benefits to the cooperator are less than those to the betrayer. Based on this， a decision method for cooperative combat of UAV clusters based on deep reinforcement learning was proposed. First， by building a public goods game based UAV cluster combat model， the interest conflict problem between individual and group in cooperation of intelligent UAV clusters was simulated. Then， Muti-Agent Deep Deterministic Policy Gradient （MADDPG） algorithm was used to solve the most reasonable combat decision of the auxiliary UAV cluster to achieve cluster victory with minimum loss cost. Training and experiments were performed under conditions of different numbers of UAV. The results show that compared to the training effects of two algorithms — IDQN （Independent Deep Q-Network） and ID3QN （Imitative Dueling Double Deep Q-Network）， the proposed algorithm has the best convergence， its winning rate can reach 100% with four auxiliary UAVs， and it also significantly outperforms the comparison algorithms with other UAV numbers.

Human pose estimation is one of the basic tasks in computer vision， which can be applied to the fields such as action recognition， games， and animation production. The current designs of deep network model mostly use deepening the network to obtain better performance. As a result， the demand for computing resources is beyond the computing power of embedded devices and mobile devices， and the requirements of actual applications can not be met. In order to solve the problems， a new lightweight network model integrating Ghost module structure was proposed， that is， the Ghost module was used to replace the basic module in the original high-resolution network， thereby reducing the number of network parameters. In addition， a non-local high-resolution network was designed， that is， the non-local network module was fused in the 1/32 resolution stage of the network， so that the network has the ability to obtain global features， thereby improving the accuracy of human pose estimation， and reducing the network parameters while ensuring the accuracy of model. Experiments were carried out on the human pose estimation datasets such as Max Planck Institut Informatik （MPII） and the Common Objects in COntext （COCO）.Experimental results indicate that， compared with the original high-resolution network， the proposed network model has the accuracy of human pose estimation increased by 1.8 percentage points with the number of network parameters reduced by 40%.

Point-Of-Interest (POI) recommendation aims to recommend places that users do not visit but may be interested in, which is one of the important location-based services. In POI recommendation, time is an important factor, but it is not well considered in the existing POI recommendation models. Therefore, the Time-incorporated User-based Collaborative Filtering POI recommendation (TUCF) algorithm was proposed to improve the performance of POI recommendation by considering time factor. Firstly, the users' check-in data of Location-Based Social Network (LBSN) was analyzed to explore the time relationship of users' check-ins. Then, the time relationship was used to smooth the users' check-in data, so as to incorporate time factor and alleviate data sparsity. Finally, according to the user-based collaborative filtering method, different POIs were recommended to the users at different times. Experimental results on real check-in datasets showed that compared with the User-based collaborative filtering (U) algorithm, TUCF algorithm had the precision and recall increased by 63% and 69% respectively, compared with the U with Temporal preference with smoothing Enhancement (UTE) algorithm, TUCF algorithm had the precision and recall increased by 8% and 12% respectively. And TUCF algorithms reduced the Mean Absolute Error (MAE) by 1.4% and 0.5% respectively, compared with U and UTE algorithms.

Aiming at the problem that the cross-modality interaction and the impact of the contribution of each modality on the final sentiment classification results are not considered in multimodal sentiment analysis of video, a multimodal sentiment analysis model of Attention Mechanism based feature Fusion-Bidirectional Gated Recurrent Unit (AMF-BiGRU) was proposed. Firstly, Bidirectional Gated Recurrent Unit (BiGRU) was used to consider the interdependence between utterances in each modality and obtain the internal information of each modality. Secondly, through the cross-modality attention interaction network layer, the internal information of the modalities were combined with the interaction between modalities. Thirdly, an attention mechanism was introduced to determine the attention weight of each modality, and the features of the modalities were effectively fused together. Finally, the sentiment classification results were obtained through the fully connected layer and softmax layer. Experiments were conducted on open CMU-MOSI (CMU Multimodal Opinion-level Sentiment Intensity) and CMU-MOSEI (CMU Multimodal Opinion Sentiment and Emotion Intensity) datasets. The experimental results show that compared with traditional multimodal sentiment analysis methods (such as Multi-Attention Recurrent Network (MARN)), the AMF-BiGRU model has the accuracy and F1-Score on CMU-MOSI dataset improved by 6.01% and 6.52% respectively, and the accuracy and F1-Score on CMU-MOSEI dataset improved by 2.72% and 2.30% respectively. AMF-BiGRU model can effectively improve the performance of multimodal sentiment classification.

Node deployment is one of the important problems in Wireless Sensor Network (WSN). Concerning the problem of energy hole in the process of node employment, a Node Redeployment Based on the Firefly Algorithm (NRBFA) strategy was proposed. Firstly, the k-means algorithm was used to cluster nodes and the redundant nodes were introduced into the sensor network where nodes are randomly deployed. Then, the Firefly Algorithm (FA) was used to move the redundant nodes to share the load of Cluster Heads (CHs) and balance the energy consumption of nodes in the network. Finally, the redundant nodes were updated after finding the target node by reusing the FA. In the proposed strategy, the reduction of moving distances of nodes and the decrease of the network energy consumption were achieved through moving the redundant nodes effectively. Experimental results show that the proposed strategy can alleviate the "energy hole" problem effectively. Compared with the partition node redeployment algorithm based on virtual force, the proposed strategy reduces the complexity of the algorithm, and can better improve the energy efficiency of the network, balance the network load, as well as prolong the network lifetime by nearly 10 times.

In order to make full use of appearance and motion information in video abnormal behavior detection, a Siamese network model that can capture appearance and motion information at the same time was proposed. The two branches of the network were composed of the same autoencoder structure. Several consecutive frames of RGB images were used as the input of the appearance sub-network to predict the next frame, while RGB frame difference image was used as the input of the motion sub-network to predict the future frame difference. In addition, considering one of the reasons that affected the detection effect of the prediction-based method, that is the diversity of normal samples, and the powerful "generation" ability of the autoencoder network, that is it has a good prediction effect on some abnormal samples. Therefore, a memory enhancement module that learns and stores the "prototype" features of normal samples was added between the encoder and the decoder, so that the abnormal samples were able to obtain greater prediction error. Extensive experiments were conducted on three public anomaly detection datasets Avenue, UCSD-ped2 and ShanghaiTech. Experimental results show that, compared with other video abnormal behavior detection methods based on reconstruction or prediction, the proposed method achieves better performance. Specifically, the average Area Under Curve (AUC) of the proposed method on Avenue, UCSD-ped2 and ShanghaiTech datasets reach 88.2%, 97.5% and 73.0% respectively.

Band selection can effectively reduce the spatial redundancy of hyperspectral data and provide effective support for subsequent classification. Multi-kernel fuzzy rough set model is able to analyze numerical data containing uncertainty and approximate description, and grasshopper optimization algorithm can solve optimization problem with strong exploration and development capabilities. Multi-kernelized fuzzy rough set model was introduced into hyperspectral uncertainty analysis modeling, grasshopper optimization algorithm was used to select the subset of bands, then a hyperspectral band selection algorithm based on multi-kernel fuzzy rough set and grasshopper optimization algorithm was proposed. Firstly, the multi-kernel operator was used to measure the similarity in order to improve the adaptability of the model to data distribution. The correlation measure of bands based on the kernel fuzzy rough set was determined, and the correlation between bands was measured by the lower approximate distribution of ground objects at different pixel points in fuzzy rough set. Then, the band dependence, band information entropy and band correlation were considered comprehensively to define the fitness function of band subset. Finally, with J48 and K-Nearest Neighbor ( KNN) adopted as the classifier algorithms, the proposed algorithm was compared with Band Correlation Analysis (BCA) and Normalized Mutual Information (NMI) algorithms in the classification performance on a common hyperspectral dataset Indiana Pines agricultural area. The experimental results show that the proposed algorithm has the overall average classification accuracy increased by 2.46 and 1.54 percentage points respectively when fewer bands are selected.

In order to reduce the redundancy between hyperspectral band images, decrease the computing time and facilitate the following classification task, a hyperspectral band selection algorithm based on kernelized fuzzy rough set was proposed. Due to strong similarity between adjacent bands of hyperspectral images, the kernelized fuzzy rough set theory was introduced to measure the importance of bands more effectively. Considering the distribution characteristics of categories in the bands, the correlation between bands was defined according to the distribution of the lower approximate set of bands, and then the importance of bands was defined by combining the information entropy of bands. The search strategy of maximum correlation and maximum importance was used to realize the band selection of hyperspectral images. Finally, experiments were conducted on the commonly used hyperspectral dataset Indiana Pines agricultural area by using the J48 and KNN classifiers. Compared with other hyperspectral band selection algorithms, this algorithm has overall average classification accuracy increased by 4.5 and 6.6 percentage points respectively with two classifiers. The experimental results show that the proposed algorithm has some advantages in hyperspectral band selection.

For the problem that the Rivest Cipher 4 (RC4) algorithm has invariant weak key, the randomness of the key stream sequence is not high and the initial state of the algorithm can be cracked, an improved RC4 algorithm based on elliptic curve was proposed. In the algorithm, the initial key was generated by using elliptic curve, Hash function and pseudo-random number generator, and a nonlinear transformation was performed under the action of the S-box and the pointer to finally generate a key stream sequence with high randomness. The randomness test carried out by National Institute of Standards and Technology (NIST) shows that the frequency test, run test and Maurer are 0.13893, 0.13081, and 0.232050 respectively higher than those of the original RC4 algorithm, which can effectively prevent the generation of invariant weak keys and resist the "sentence" attack. The initial key is a uniformly distributed random number without deviation, which can effectively resist the distinguishing attack. The elliptic curve and Hash function have one-way irreversibility, the pseudo-random number generator has high password strength, the initial key guess is difficult to assign and is not easy to crack, which can resist the state guessing attack. Theoretical and experimental results show that the improved RC4 algorithm is more random and safe than the original RC4 algorithm.

The problem of misclassification of minority class samples appears frequently when classifying massive amount of imbalanced data in real life with traditional classification algorithms, because most of these algorithms only suit balanced class distribution or samples with same misclassification cost. To overcome this problem, a classification algorithm for imbalanced dataset based on cost sensitive ensemble learning and oversampling-New Imbalanced Boost (NIBoost) was proposed. Firstly, the oversampling algorithm was used to add a certain number of minority samples to balance the dataset in each iteration, and the classifier was trained on the new dataset. Secondly, the classifier was used to classify the dataset to obtain the predicted class label of each sample and the classification error rate of the classifier. Finally, the weight coefficient of the classifier and new weight of each sample were calculated according to the classification error rate and the predicted class labeles. Experimental results on UCI datasets with decision tree and Naive Bayesian used as weak classifier algorithm show that when decision tree was used as the base classifier of NIBoost, compared with RareBoost algorithm, the F-value is increased up to 5.91 percentage points, the G-mean is increased up to 7.44 percentage points, and the AUC is increased up to 4.38 percentage points. The experimental results show that the proposed algorithm has advantages on imbalanced data classification problem.

For image instance retrieval in the field of computational vision, a semantic region weighted aggregation method based on significance guidance of deep convolution features was proposed. Firstly, a tensor after full convolutional layer of deep convolutional network was extracted as deep feature. A feature saliency map was obtained by using Inverse Document Frequency (IDF) method to weight deep feature, and then it was used as a constraint to guide deep feature channel importance ordering to extract different special semantic region deep feature, which excluded interference from background and noise information. Finally, global average pooling was used to perform feature aggregation, and global feature representation of image was obtained by using Principal Component Analysis (PCA) to reduce the dimension and whitening for distance metric retrieval. The experimental results show that the proposed image retrieval algorithm based on significant semantic region weighting is more accurate and robust than the current mainstream algorithms on four standard databases, because the image feature vector extracted by the proposed algorithm is richer and more discerning.

As the traditional Scale Invariant Feature Transform (SIFT) image matching algorithm has high false matching rate and eliminating the condition of mismatching points is unitary, an adaptive image matching method based on SIFT operator fused with maximum dissimilarity coefficient was proposed. Firstly, On the basis of Euclidean distance measurement, the optimal maximum dissimilarity coefficients values of the 128-dimensional feature vectors in SIFT algorithm were obtained. Then, the matching points were selected according to the obtained optimal values. Random Sample Consensus (RANSAC) was used to calculate the correct rate of matching. Finally, the stereo matching images of Daniel Scharstein and Richard Szeliski were used to verify the algorithm. The experimental results show that the correct matching rate of the improved algorithm is about 10 percentage points higher than that of the traditional SIFT algorithm. The improved algorithm effectively reduces the mismatches and is more suitable for image matching applications with similar regions. In terms of runtime, the proposed method has an average time of 1.236 s, which can be applied to the systems with low real-time requirements.

Traditional trajectory clustering algorithm is no longer applicable due to the lack of continuous ship navigation data for large-scale sea area extraction. To solve this problem, a technique of vessel traffic pattern extraction using Hough transformation was proposed. Based on Automatic Identification System (AIS) data, the target area was divided into grids so that the ship density distribution was analyzed. Considering the problem of density distribution resolution, median filtering and morphological filtering were used to optimize the density distribution. Thus a method combining Hough transformation and Kernel density estimation was proposed to extract vessel traffic pattern and estimate the width of pattern. The experimental verification of the method with real historical AIS data shows that the trajectory clustering method cannot extract vessel traffic pattern in lower ship-density areas, its extracted number of ship trajectories in trajectory clusters accounts for 29.81% of the total number in the area, compared to 95.89% using the proposed method. The experimental result validates the effectiveness of the proposed method.

Those well-known types of intrusions can be detected with higher detection rate in the network at present, but it is very difficult to detect those new unknown types of network intrusions. In order to solve the problem, a network intrusion detection method of Deep Belief Network (DBN) model based on optimization of data processing was proposed. The data processing and method model were improved respectively without destroying the existing knowledge and increasing detection time seriously to solve the above problem. Firstly, the data processed by Probability Mass Function (PMF) encoding and MaxMin normalization was applied to the DBN model. Then, the relatively optimal DBN structure was selected through fixing other parameters, changing a parameter and the cross validation. Finally, the proposed method was tested on the benchmark NSL-KDD dataset. The experimental results show that, the optimization of data processing can improve the classification accuracy of the DBN model, the proposed intrusion detection method based on DBN has good adaptability and higher recognition ability of unknown samples. The detection time of DBN algorithm is similar to that of Support Vector Machine (SVM) algorithm and Back Propagation (BP) neural network model.

A spatial co-location pattern represents a subset of spatial features whose instances are frequently located together in spatial neighborhoods. The existing interesting metrics for spatial co-location pattern mining do not take account of the difference between features and the diversity between instances belonging to the same feature. In addition, using the traditional data-driven spatial co-location pattern mining method, the mining results often contain a lot of useless or uninteresting patterns. In view of the above problems, firstly, a more general study object-spatial instance with utility value was proposed, and the Utility Participation Index (UPI) was defined as the new interesting metric of the spatial high utility co-location patterns. Secondly, the domain knowledge was formalized into three kinds of semantic rules and applied to the mining process, and a new domain-driven iterative mining framework was put forward. Finally, by the extensive experiments, the differences between mined results with different interesting metrics were compared in two aspects of utility ratio and frequency, as well as the changes of the mining results after taking the domain knowledge into account. Experimental results show that the proposed UPI metric is a more reasonable measure in consideration of both frequency and utility, and the domain-driven mining method can effectively find the co-location patterns that users are really interested in.

With the emergence of various social networks, the social media network data is analyzed from the perspective of variety by more and more researchers. The data fusion of multiple social networks relies on user identification across social networks. Concerning the low utilization problem of heterogeneous relation between social networks of the traditional Friend Relationship-based User Identification (FRUI) algorithm, a new Weighted Hypergraph based User Identification (WHUI) algorithm across social networks was proposed. Firstly, the weighted hypergraph was accurately constructed on the friend relation networks to describe the friend relation and the heterogeneous relation in the same network, which improved the accuracy of presenting topological environment of nodes. Then, on the basis of the constructed weighted hypergraph, the cross network similarity between nodes was defined according to the consistency of nodes' topological environment in different networks. Finally, the user pair with the highest cross network similarity was chosen to match each time by combining with the iterative matching algorithm, while two-way authentication and result pruning were added to ensure the recognition accuracy. The experiments were carried out in the DBLP cooperation networks and real social networks. The experimental results show that, compared with the existing FRUI algorithm, the average precision, recall, F of the proposed algorithm is respectively improved by 5.5 percentage points, 3.4 percentage points, 4.6 percentage points in the real social networks. The WHUI algorithm can effectively improve the precision and recall of user identification in practical applications by utilizing only network topology information.

The "consumption first and replenishment afterward" operation model of the power supply companies may cause the risk of arrears due to poor credit of some power consumers. Therefore, it is necessary to analyze of the tremendous user data in real-time and quickly before the arrears' happening and provide a list of the potential customers in arrear. In order to solve the problem, a method for arrears alert of power consumers based on the parallel classification algorithm was proposed. Firstly, the arrear behaviors were modeled by the parallel Random Forest (RF) classification algorithm based on the Spark framework. Secondly, based on previous consumption behaviors and payment records, the future characteristics of consumption and payment behavior were predicted by time series. Finally, the list of the potential hig-risk customers in arrear was obtained by using the obtained model for classifying users. The proposed algorithm was compared with the parallel Support Vector Machine (SVM) algorithm and Online Sequential Extreme Learning Machine (OSELM) algorithm. The experimental results demonstrate that, the prediction accuracy of the proposed algorithm performs better than the other algorithms in comparison. Therefore, the proposed method is a convenient way for electricity recycling management to remind the customers of paying the electricity bills ahead of time, which can ensure timeliness electricity recovery. Moreover, the proposed method is also beneficial for consumer arrear risk management of the power supply companies.

The traditional video super-resolution reconstruction algorithm cannot preserve the details of the image edge effectively while removing the noise. In order to solve this problem, a video super-resolution reconstruction algorithm combining adaptive regularization term with multi-order derivative data item was put forward. Based on the regularization reconstruction model, the multi-order derivative of the noise, which described the statistical characteristics of the noise well, was introduced into the improved data item; meanwhile, Total Variation (TV) and Non-Local Mean (NLM) which has better denoising effect were used as the regularization items to constrain the reconstruction process. In addition, to preserve the details better, the coefficient of regularization was weighted adaptively according to the structural information, which was extracted by the regional spatially adaptive curvature difference algorithm. In the comparison experiments with the kernel-regression algorithm and the clustering algorithm when the noise variance is 3, the video reconstructed by the proposed algorithm has sharper edge, the structure is more accurate and clear; and the average Mean Squared Error (MSE) is decreased by 25.75% and 22.50% respectively; the Peak Signal-to-Noise Ratio (PSNR) is increased by 1.35 dB and 1.14 dB respectively. The results indicate that the proposed algorithm can effectively preserve the details of the image while removing the noise.

To solve the time-consuming problem of group-by aggregation operation in case of data-intense computation, a cache-friendly group-by aggregation method was proposed. In this paper, the group-by aggregation operation was optimized in two aspects. Firstly, designing cache-friendly group-by aggregation algorithm on Supersonic, an open-source and column-oriented query execution engine, to take the full advantage of column-storage on in-memory computation. Secondly, rewriting the algorithm with multi-threads to speed up the query. In this paper, four different parallel aggregation algorithms were put forward, respectively named Shared-Nothing Parallel Group-by Aggregation (NSHPGA) algorithm, Table-Lock Shared-Hash Parallel Group-by Aggregation (TLSHPGA) algorithm, Bucket-Lock Shared-Hash Parallel Group-by Aggregation (BLSHPGA) algorithm and Node-Lock Shared-Hash Parallel Group-by Aggregation (NLSHPGA) algorithm. Through a series of comparison experiment on different group power set and different number of worker threads, NLSHPGA algorithm was proved to have the best performance both on speed-up ratio and concurrency, which achieved 10x speedups on part of queries. Besides, considering Cache miss and memory utilization, the results shows that NSHPGA algorithm is suitable for smaller group power set, which was 8 in the experiment, and when getting larger, NLSHPGA algorithm performs better than NSHPGA algorithm.

Since the Bit Error Rate (BER) of the Blind Source Separation (BSS) of mixed digital modulation signals under the noisy environment is excessively high, a two-stage blind source separation algorithm named R-TSBS was proposed based on RobustICA (Robust Independent Component Analysis). Firstly, the algorithm used RobustICA to estimate the mixing matrix consisting of array response vector. In the second phase, each symbol sequence transmitted by digital modulation source signal was estimated by Maximum Likelihood Estimation (MLE) method using the finite symbol values character. Finally, R-TSBS achieved the purpose of blind source separation. The simulation results show that, when the Signal to Noise Ratio (SNR) is 10 dB, the BER of traditional Independent Component Analysis (ICA) algorithm such as FastICA (Fast Independent Component Analysis) and RobustICA reached 3.5×10^-2, which is exactly high. However, the BER of the two-stage blind source separation on the basis of FastICA algorithm which named F-TSBS and the proposed R-TSBS algorithm dropped to 10^-3, the separation performance has been significantly improved. At the same time, R-TSBS algorithm can obtain about 2 dB performance increase in low SNR (0~4 dB) compared to F-TSBS algorithm.

Finite element analysis for elastic-plastic problem is very time-consuming. A parallel substructure Preconditioned Conjugate Gradient (PCG) algorithm combined with Minimal Residual (MR) smoothing was proposed under the environment of Message Passing Interface (MPI) cluster. The proposed method was based on domain decomposition, and substructure was treated as isolated finite element model via the interface conditions. Throughout the analysis, each processor stored only the information relevant to its substructure and generated the local stiffness matrix. A parallel substructure oriented preconditioned conjugate gradient method was developed, which combined with MR smoothing and diagonal storage scheme. Load balance was discussed and interprocessor communication was optimized in the parallel algorithm. A substepping scheme to integrate elastic-plastic stress-strain relations was used. The errors in the integration process were controlled by adjusting the substep size automatically according to a prescribed tolerance. Numerical example was implemented to validate the performance of the proposed PCG algorithm on workstation cluster. The performance of the proposed PCG algorithm was analyzed and the performance was compared with conventional PCG algorithm. The example results indicate that the proposed algorithm has good speedup and efficiency and is superior in performance to the conventional PCG algorithm. The proposed algorithm is efficient for parallel computing of 3D elastic-plastic problems.

The research of dissemination effect of micro-blog message has an important role in improving marketing, strengthening public opinion monitoring and discovering hotspots accurately. Focused on difference between individuals which was not considered previously, this paper proposed a method of predicting scale and depth of retweeting based on behavior analysis. This paper presented a predictive model of retweet behavior with Logistic Regression (LR) algorithm and extracted nine relative features from users, relationship and content. Based on this model, this paper proposed the above predicting method which considered the character of information disseminating along users and iterative statistical analysis of adjacent users step by step. The experimental results on Sina micro-blog dataset show that the accuracy rate of scale and depth prediction approximates 87.1% and 81.6 respectively, which can predict the dissemination effect well.

The traditional graph-based recommendation algorithm neglects the combined time factor which results in the poor recommendation quality. In order to solve this problem, a personalized recommendation algorithm integrating roulette walk and combined time effect was proposed. Based on the user-item bipartite graph, the algorithm introduced attenuation function to quantize combined time factor as association probability of the nodes; Then roulette selection model was utilized to select the next target node according to those associated probability of the nodes skillfully; Finally, the top-N recommendation for each user was provided. The experimental results show that the improved algorithm is better in terms of precision, recall and coverage index, compared with the conventional PersonalRank random-walk algorithm.

According to the shape features of wind shear images extracted by wavelet invariant moment based on cubic B-spline wavelet basis, an improved Genetic Algorithm (GA) was proposed to apply to the type recognition of microburst, low-level jet stream, side wind shear and tailwind-or-headwind shear. In the improved algorithm, the adaptive crossover probability only considered the number of generation and mutation probability just emphasized the fitness valve of individuals and group, so that it could control the evolution direction uniformly, and greatly maintain the population diversity simultaneously. Lastly, the best feature subset chosen by the improved genetic algorithm was fed into 3-nearest neighbor classifier to classify. The experimental results show that it has a good direction and be able to rapidly converge to the global optimal solution, and then steadily chooses the critical feature subset in order to obtain a better performance of wind shear recognition that the mean recognition rate can reach more than 97% at last.

In order to solve the problems, such as facial change and uneven gray, caused by the variations of expression and illumination in face recognition, a novel feature extraction method based on Sub-pattern Row-Column Two-Dimensional Linear Discriminant Analysis (Sp-RC2DLDA) was proposed. In the proposed method, by dividing the original images into smaller sub-images, the local features could be extracted effectively, and the impact of variations in facial expression and illumination was reduced. Also, by combining the sub-images at the same position as a subset, the recognition performance could be improved for making full use of the spatial relationship among sub-images. At the same time, two classes of features which complemented each other can be obtained by synthesizing the local sub-features which were achieved by performing 2DLDA (Two-Dimensional Linear Discriminant Analysis) and Extend 2DLDA (E2DLDA) on a set of partitioned sub-patterns in the row and column directions, respectively. Then, the recognition performance was expected to be improved by employing a fusion method to effectively fuse these two classes of complementary features. Finally, nearest neighbor classifier was applied for classification. The experimental results on Yale and ORL face databases show that the proposed Sp-RC2DLDA method reduces the influence of variations in illumination and facial expression effectively, and has better robustness and classification performance than the other related methods.

In recommendation system, it is difficult to predict the behavior of users on items and give the accurate recommendation. In order to improve the accuracy of recommendation system, the recommendation potential was introduced and a novel personalized hybrid recommendation model fused with recommendation potential was proposed. Firstly, the trend momentum was calculated according to the visits of items in recent short time and long time; then, the current recommendation potential was calculated utilizing trend momentum; finally, the hybrid recommendation model was achieved according to the fusion of recommendation potential and personalized recommendation model. The experimental results show that the personalized trend prediction fused with recommendation potential can improve the accuracy of recommendation system in a large scale.

As to the low clustering quality and high communication cost of the existed distributed clustering algorithm, a distributed data stream clustering algorithm (DAPDC) which combined the density with the idea of representative points clustering was proposed. The concept of the class cluster representative point to describe the local distribution of data flows was introduced in the local sites using affinity propagation clustering, while the global site got the global model by merging the summary data structure that was uploaded from the local site by the improved density clustering algorithm. The simulation results show that DAPDC can improve the clustering quality of data streams in distributed environment significantly. Simultaneously, the algorithm can find the clusters of different shapes and reduce the amount of data transferred significantly by using class cluster representative points.

Concerning the safe problem of digital image information hiding, a scrambling algorithm based on bitwise layered Arnold transform was proposed. The secret image was stratified by bit-plane, taking into account the location and pixel gray transform, each bit-plane was scrambled for different times with Arnold transforma, and the pixel was cross transposed, and adjacent pixels were bitwise XOR to get a scrambling image. The experimental results show that the secret image histogram is more evenly distributed after stratification scrambling, its similarity with the white noise is around 0.962, and the scrambling image can be restored and extracted almost lossless, which improves the robustness. Compared with other scrambling algorithms, the proposed algorithm is more robust to resist attack, and improves the spatial information hiding security.

In order to improve the operation efficiency of big module RSA (Rivest-Shamir-Adleman) signature algorithm, four prime Chinese Remainder Theorem (CRT)-RSA digital signature was suggested in this paper. The Hash function SHA512 was used to produce message digest, and CRT combining with Montgomery algorithm was applied to optimize large number modular exponentiation. The security analysis and experiment show that the new algorithm can resist some common attacks, and it has some advantages in signature efficiency.

Aiming at terminal performance bottleneck among current data transfer process, a UDP-based terminal adaptive protocol was proposed. After the analysis and the comparison of many factors which affected terminal performance, this protocol viewed both the previous packet loss ratio and the current one as congestion detection parameters. It employed various rate adaption methods such as finite loop counter and process scheduling function in order to balance performance differences in real-time and ensured reliable and fast data transfer. Compared with traditional idle Automatic Repeat reQuest (ARQ) method, the average delay is reduced by more than 25%. The experimental results show that the proposed algorithm has the features of strong real-time, quick response, and it is compatible with large amount of data transmission, especially suitable for small amount of data transmission in engineering applications.