In order to improve the accuracy of image classification with convolution network model without increasing more computation, a new image deep convolution classification method based on complex network description was proposed. Firstly, the complex network model degree matrices under different thresholds were obtained by using complex network description of image. Then, the feature vector was obtained by deep convolution neural networks based on degree matrix description of image. Finally, the obtained feature vectors were used for image K-Nearest Neighbors ( KNN) classification. The verification experiments were carried out on the ImageNet Large Scale Visual Recognition Challenge 2014 (ILSVRC2014) database. The experimental results show that the proposed model has higher accuracy and fewer iterations.

As the low ability of triggering the data racing fault of the existing mutation operators for concurrent program in mutation testing, some new mutation strategies based on data racing fault were proposed. From the viewpoint of mutation operator designing, Lock-oriented Mutation Strategy (LMS) and Shared-variable-oriented Mutation Strategy (SMS) were introduced, and two new mutation operators that named Synchronized Lock Resting Operator (SLRO) and Move Shared Variable Operator (MSVO) were designed. From the viewpoint of mutation point selection, also a new mutation point selection strategy named Synchronized relationship pair Mutation Point Selection Strategy (SMPSS) was proposed. SLRO and MSVO mutation operators were used to inject the faults which generated by SMPSS strategy on 12 Java current libraries, and then the ability of mutants to trigger the data racing fault was checked by using Java Path Finder (JPF). The results show that the SLRO and MSVO for 12 Java libs can generate 121 and 122 effective mutants respectively, and effectiveness rates are 95.28% and 99.19% respectively. In summary, the new current mutation operators and mutation strategies can effectively trigger the data racing fault.

In view of significance of color image recognition, the method of color image recognition based on data of image features and Deep Belief Network (DBN) was presented. Firstly, data field of color image was constructed in accord with human visual characteristics; secondly, wavelet transforms was applied to describe multi-scale feature of image; finally, image recognition could be made by training unsupervised DBN.The experimental results show that compared with the methods of Adaboost and Support Vector Machine(SVM),classification accuracy is improved by 3.7% and 2.8% respectively and better image recognition is achieved by the proposed method.

Aiming at the problem of low efficiency on generating Thread Scheduling Sequence (TSS) that cover synchronization statements in multi-thread concurrent program, a TSS Generation Based on All synchronization pairs coverage criteria (TGBA) method was proposed. First, according to the synchronization statements in concurrent program, the synchronization pair and All Synchronization Pairs Coverage Criteria (APSC) were defined. Second, a construction method of Synchronization Pair Thread Graph (SPTG) was given. On that basis, TSSs that satisfied APSC were generated. Finally, by using JPF (Java PathFinder) detection tool, TSS generation experiments were conducted on four Java Library concurrent programs, and the comparison analysis of generation efficiency was conducted with general sequence generation methods of Default Scheduling (DS), Preemptive Scheduling (PS) and Cross Scheduling (CS). The experimental results illustrate that TSSs generated by TGBA method can cover all synchronization pairs compared to the DS and CS method. Moreover, when satisfying APSC, TGBA method decreases at least 19889 states and 44352 transitions compared to the PS method, and the average generation efficiency increases by 1.95 times. So TGBA method can reduce cost of state space and improve the efficiency of TSS generation.

The null models of complex networks generated by random scrambling algorithm often can't tell when null models can be stable because of the difference of successful scrambling probabilities of different order null models. Focusing on the issue, the concept of "successful scrambling times" was defined and used to replace the usual "try scrambling times" to set the algorithm. The index of the proposed successful scrambling times could be added only when the randomly selected edges could meet the scrambling conditions of corresponding null models, and thus be successfully scrambled. The generation experiments of null models of every order show that every index can be stable in a small scale of successful scrambling times. Further quantitative analyses show that, according to the corresponding orders, 0-order, 1-order and 2-order null models with good quality can be got by setting successfully scrambling times to be 2 times, 1 times and 1 times of actual networks' edge number respectively.

CT image denoising restoration is a basic procedure in medical image processing. Cartoon-texture decomposition method was extended in order to resolve the problems of computational difficulty and low precision while applying cartoon-texture models in medical image denoising. First, the structure of cartoon-texture model was described by curvelet transform. Second, the texture of cartoon-texture decomposition was described using more sparse dual-tree complex wavelet transform. Third, an image cartoon images-texture model was established by combining curvelet transform and sparse representation. The algorithms of cartoon-texture model were discussed at last. The simulation experimental results show that the new method can effectively resolve the problem of large amount of iterative calculation using medical image denoising algorithm, and the image quality after processing can be improved as well.

With the continuing increase of users and kinds of resources, the problem of rating matrix's sparsity is becoming more and more prominent, which seriously affects the quality of the recommendation system. Singular Value Decomposition (SVD) is a dimension reduction method, and Symbolic Data Analysis (SDA) is a new analytical approach to processing mass data. This paper proposed a new collaborative filtering recommendation algorithm which combines SVD with SDA. The experimental results based on EachMovie database set indicate that the proposed method is significantly better than traditional general recommendation algorithm when the data is particularly sparse.

Feature selection is a valid method to reduce the dimension of text vector in automatic text categorization system. After analyzing some normal evaluation functions for feature selection, a new evaluation function named the ratio of mutual information in feature selection was presented. Experiments show that the method is simple and feasible. It is advantageous in improving the efficiency of the selected feature subset.