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.

Aiming at the problem of low detection rate of depth feature based object detection method Faster R-CNN （Faster Region-based Convolutional Neural Network） in flame detection tasks, a color-guided anchoring strategy was proposed. In this strategy, a flame color model was designed to limit the generation of anchors, which means the flame color was used to limit the generation locations of the anchors, thereby reducing the number of initial anchors and improving the computational efficiency. To further improve the computational efficiency of the network, the masked convolution was used to replace the original convolution layer in the region proposal network. Experiments were conducted on BoWFire and Corsician datasets to verify the detection performance of the proposed method. The experimental results show that the proposed method improves detection speed by 10.1% compared to the original Faster R-CNN, has the F-measure of flame detection of 0.87 on BoWFire, and has the accuracy reached 99.33% on Corsician.The proposed method can improve the efficiency of flame detection and can accurately detect flames in images.

Currently, the adaptivity of global training dictionary is not strong for brain medical images, and the "max-L ₁" rule may cause gray inconsistency in the fused image, which cannot get satisfactory image fusion results. A multi-modal brain image fusion method based on adaptive joint dictionary learning was proposed to solve this problem. Firstly, an adaptive joint dictionary was obtained by combining sub-dictionaries which were adaptively learned from registered source images using improved K-means-based Singular Value Decomposition ( K-SVD) algorithm. The sparse representation coefficients were computed by the Coefficient Reuse Orthogonal Matching Pursuit (CoefROMP) algorithm by using the adaptive joint dictionary. Furthermore, the activity level measurement of source image patches was regarded as the "multi-norm" of the sparse representation coefficients, and an unbiased rule combining "adaptive weighed average" and "choose-max" was proposed, to chose fusion rule according to the similarity of "multi-norm" of the sparse representation coefficients. Then, the sparse representation coefficients were fused by the rule of "adaptive weighed average" when the similarity of "multi-norm" was greater than the threshold, otherwise the rule of "choose-max" was used. Finally, the fusion image was reconstructed according to the fusion coefficient and the adaptive joint dictionary. The experimental results show that, compared with the other three methods based on multi-scale transform and five methods based on sparse representation, the fusion images of the proposed method have more image detail information, better image contrast and sharpness, and clearer edge of lesion, the mean values of the objective parameters such as standard deviation, spatial frequency, mutual information, the gradient based index, the universal image quality based index and the mean structural similarity index under three groups of experimental conditions are 71.0783, 21.9708, 3.6790, 0.6603, 0.7352 and 0.7339 respectively. The proposed method can be used for clinical diagnosis and assistant treatment.

The dictionary training process is time-consuming, and it is difficult to obtain accurate sparse representation by using a single dictionary to express brain medical images currently, which leads to the inefficiency of image fusion. In order to solve the problems, a Computed Tomography (CT)/Magnetic Resonance (MR) brain image fusion method via improved coupled dictionary learning was proposed. Firstly, the CT and MR images were regarded as the training set, and the coupled CT and MR dictionary were obtained through joint dictionary training based on improved K-means-based Singular Value Decomposition (K-SVD) algorithm respectively. The atoms in CT and MR dictionary were regarded as the features of training images, and the feature indicators of the dictionary atoms were calculated by the information entropy. Then, the atoms with the smaller difference feature indicators were regarded as the common features, the rest of the atoms were considered as the innovative features. A fusion dictionary was obtained by using the rule of "mean" and "choose-max" to fuse the common features and innovative features of the CT and MR dictionary separately. Further more, the registered source images were compiled into column vectors and subtracted the mean value. The accurate sparse representation coefficients were computed by the Coefficient Reuse Orthogonal Matching Pursuit (CoefROMP) algorithm under the effect of the fusion dictionary, the sparse representation coefficients and mean vector were fused by the rule of "2-norm max" and "weighted average" separately. Finally, the fusion image was obtained via reconstruction. The experimental results show that, compared with three methods based on multi-scale transform and three methods based on sparse representation, the image visual quality fused by the proposed method outperforms on the brightness, sharpness and contrast, the mean value of the objective parameters such as mutual information, the gradient based, the phase congruency based and the universal image quality indexes under three groups of experimental conditions are 4.1133, 0.7131, 0.4636 and 0.7625 respectively, the average time in the dictionary learning phase under 10 experimental conditions is 5.96 min. The proposed method can be used for clinical diagnosis and assistant treatment.

A new method of large-scale optimization based on divide-and-conquer strategy was proposed. Firstly, based on the principle of additive separability, an improved variable grouping method was proposed. The randomly accessing point method was used to check the correlation between all variables in pairs. At the same time, by making full use of the interdependency information of learning, the large groups of separable variables were re-grouped. Secondly, a new subcomponent optimizer was designed based on an improved differential evolution algorithm to enhance the subspace optimization performance. Finally, this two kinds of improvements were introduced to co-evolutionary framework to construct a DECC-NDG-CUDE (Cooperative differential evolution with New Different Grouping and enhancing Differential Evolution with Commensal learning and Uniform local search) algorithm. Two experiments of grouping and optimization were made on 10 large-scale optimization problems. The experimental results show the interdependency between variables can be effectively identified by the new method of grouping, and the performance of DECC-NDG-CUDE is better than two state-of-the-art algorithms DECC-D (Differential Evolution with Cooperative Co-evolution and differential Grouping) and DECCG (Differential Evolution with Cooperative Co-evolution and Random Grouping).

In order to solve the problem of redundancy and transmission energy consumption in the process of data acquisition in wireless sensor networks, a method for designing the measurement matrix of compressive sensing was proposed in this paper. The method is based on the linear representation theory of diagonal matrix orthogonal basis and the process of constructing the matrix is simple with short time, high sparsity and low redundancy, which is very suitable for the nodes with limited hardware resources. The simulation results show the measurement method based on the linear representation theory of diagonal matrix gains higher signal recovery rate compared with Gauss random matrix and part Hadamard matrix under the same signal reconstruction accuracy. This method in the paper greatly reduces the traffic of networks, saves the network energy consumption and prolongs the network life cycle.

Algebraic immunity is one of the most significant cryptographic properties for Boolean functions. In order to resist algebraic attack, high algebraic immunity is necessary for those Boolean functions used in stream ciphers. This paper constructed more than one even-variable rotation symmetric Boolean functions with optimum algebraic immunity by giving an even n. Based on majority function, some orbits of different hamming weights were chosen, then the values of functions on these orbits were changed. Given a sufficient condition of Boolean functions with optimum algebraic immunity, the new constructed Boolean functions were proved to satisfy the condition. Therefore, it shows the algebraic immunity of the functions is optimum. Thus, algebraic attacks can be resisted effectively.

The arithmetic correlation function is a new method for studying the cryptographic properties of Boolean functions. Based on the basic definitions of addition and multiplication of multi-2-adic integer, the study constructed a new algebraic ring and realized the arithmetic or “with-carry” analogs of classic correlation functions. In this paper the definition of arithmetic autocorrelation function was introduced. The arithmetic correlation value of symmetric Boolean functions was studied. The results show that the arithmetic autocorrelation function of symmetric Boolean functions is a real symmetric function with at most n1 values.

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.

Concerning the problem that the linear eigentransformation method cannot capture the statistical properties of the nonlinear facial image, a Data-driven Local Eigentransformation (DLE) method for face hallucination was proposed. Firstly, some samples most similar to the input image patch were searched. Secondly, a patch-based eigentransformation method was used for modeling the relationship between the Low-Resolution (LR) and High-Resolution (HR) training samples. Finally, a post-processing approach refined the hallucinated results. The experimental results show the proposed method has better visual performance as well as 1.81dB promotion over method of locality-constrained representation in objective evaluation criterion for face image especially with noise. This method can effectively hallucinate surveillant facial images.

To distribute the score of composite service obtained from customer to each component service based on actual and historical performance of component services, Analytic Hierarchy Process (AHP) was used to calculate the distribution weight of each component service, in which a method was presented to convert Web service process into structure tree process, and the weight matrix was used to calculate the weight of each node in the tree structure. The relationship between actual Quality of Service (QoS) of component services and its advertised utility interval of QoS were taken into consideration, and through deviation function, the deviation proportion between actual QoS utility value of component service and actual QoS average utility value of all component services was calculated, meanwhile the influence on score distribution by history performance of each component service was considered. The experimental results show that actual QoS and history performance of component services have some influence on score which was distributed, and demonstrate that the proposed approach can achieve a reasonable and fair score distribution.