Focused on the problem that the huge number of parameters and calculations of Convolutional Neural Network (CNN) limit the application of CNN on resource-constrained devices such as embedded systems, a neural network compression method of statistics based network pruning and tensor decomposition was proposed. The core idea was to use the mean and variance as the basis for evaluating the weight contribution. Firstly, Lenet5 was used as a pruning model, the mean and variance distribution of each convolutional layer of the network were clustered to separate filters with weaker extracted features, and the retained filters were used to reconstruct the next convolutional layer. Secondly, the pruning method was combined with tensor decomposition to compress the Faster Region with Convolutional Neural Network (Faster RCNN). The pruning method was adopted for the low-dimensional convolution layers, and the high-dimensional convolutional layers were decomposed into three cascaded convolutional layers. Finally, the compressed model was fine-tuned, making the model be at the convergence state once again on the training set. Experimental results on the PASCAL VOC test set show that the proposed method reduces the storage space of the Faster RCNN model by 54% while the decrease of the accuracy is only 0.58%, at the same time, the method can reach 1.4 times acceleration of forward computing on the Raspberry Pi 4B system, which helpful for the deployment of deep CNN models on resource-constrained embedded devices.

To solve the problems of updating, modifying, upgrading and maintaining the function of robot by offline and static method, SoftMan was introduced for robot platform, and the architecture of robot system, whose managing center is host SoftMan, was built. The host SoftMan was mainly researched. Firstly, the architecture of host SoftMan was constructed. Then the descriptive unification model of knowledge and behavior of host SoftMan was put forward, the knowledge model was constructed and implemented based on data structure, and the design specifications and reference realization of the algorithm were given for its main service behaviors. Finally, the robot system was unified with the SoftMan system. Through the test, the function of robot was successfully replaced online and dynamically, verifying the correctness and feasibility of the method of designing and implementing the host SoftMan.

Greatest Common Divisor (GCD) is one of the basic subjects in computational number theory. It has a wide application in encryption and analysis of cryptography. For inputing B and C, an algorithm based on right-shift k-ary reduction proposed by Sorenson was presented for finding the integers x and y which satisfy the least significant bits of Bx-Cy were 0,i.e., Bx-Cy=0(mod2^e) where positive integer e was a constant. It could do a lot of right shifts and reduce a large number of cycles with taking advantage of the algorithm for finding the integers x and y. A fast GCD algorithm was proposed combined with modulus algorithm. When the size of the input was n bits, the worst complexity of the fast GCD algorithm was still O(n²).In the best case, the complexity of the proposed algorithm could achieve O(nlog² nlog logn). The experimental data show that actual implementations given input about more than 200000 bits, the fast GCD algorithm is faster than the Binary GCD algorithm, and the fast GCD algorithm is twice as fast as the Binary GCD algorithm for 1 million bits of input.

In view of the problem that verifying the conformance of e-government network structure, a conformance verification method for e-government network based on graph approximate matching was proposed. The method firstly abstracted the graph model of e-government network, then used the modular characteristic of network structure and k-hop neighboring relationship of vertices to realize extendible approximate graph matching which got all the similar structures between the two graphs. And then it proposed an improved graph similarity measure function by introducing the node importance factor and path distance attenuation factor so as to make the conformity assessment results more accurate. The experimental result shows that the method can accurately evaluate the conformance degree of e-government network structure, and fine-grainedly reflect the similarities or differences between the network structures which include all kinds of violations in the network topology and system deployment.

Basic Input Output System (BIOS) trapdoor has huge impact on computer system, and it is difficult to detect the existence of BIOS trapdoor effectively with the existing tools. After researching BIOS structure and BIOS code obfuscation technique based on reverse analysis, BIOS trapdoors were divided into module-level BIOS trapdoor and instruction-level BIOS trapdoor according to implementation granularity, followed by analyzing the implementation principle and characteristics of these two BIOS trapdoors in detail. Finally the detection method of module-level trapdoor based on analyzing module structure and the detection method of instruction-level trapdoor based on integrity measurement were presented. The experimental results show that these two methods can detect the existence of their corresponding BIOS trapdoors effectively.