Concerning the problems of slow convergence and low accuracy of software quality prediction model based on Back Propagation Neural Network （BPNN）， a Software Quality Prediction method based on BPNN optimized by Ant Colony Optimization algorithm （SQP-ACO-BPNN） was proposed. Firstly， the software quality evaluation factors were selected and a software quality evaluation system was determined. Secondly， BPNN was adopted to build initial software quality prediction model and ACO algorithm was used to determine network structures， initial connection weights and thresholds of network. Then， an evaluation function was given to select the best structure， initial connection weights and thresholds of the network. Finally， the network was trained by BP algorithm， and the final software quality prediction model was obtained. Experimental results of predicting the quality of airborne embedded software show that the accuracy， precision， recall and F1 value of the optimized BPNN model are all improved with faster convergence， which indicates the validity of SQP-ACO-BPNN.

Concerning the problems of large number of parameters, poor real-time performance and low accuracy of traffic sign recognition algorithms based on deep learning, an improved traffic sign recognition algorithm based on YOLO v3 was proposed. First, the depthwise separable convolution was introduced into the feature extraction layer of YOLO v3, as a result, the convolution process was decomposed into depthwise convolution and pointwise convolution to separate intra-channel convolution and inter-channel convolution, thus greatly reducing the number of parameters and the calculation of the algorithm while ensuring a high accuracy. Second, the Mean Square Error (MSE) loss was replaced by the GIoU (Generalized Intersection over Union) loss, which quantified the evaluation criteria as a loss. As a result, the problems of MSE loss such as optimization inconsistency and scale sensitivity were solved. At the same time, the Focal loss was also added to the loss function to solve the problem of severe imbalance between positive and negative samples. By reducing the weight of simple background classes, the new algorithm was more likely to focus on detecting foreground classes. The results of applying the new algorithm to the traffic sign recognition task show that, on the TT100K (Tsinghua-Tencent 100K) dataset, the mean Average Precision (mAP) of the algorithm reaches 89%, which is 6.6 percentage points higher than that of the YOLO v3 algorithm; the number of parameters is only about 1/5 of the original YOLO v3 algorithm, and the Frames Per Second (FPS) is 60% higher than YOLO v3 algorithm. The proposed algorithm improves detection speed and accuracy while reducing the number of model parameters and calculation.

Focused on the problems of low accuracy and long time cost of manual detection of breast cancer lymph node metastasis, a neural network detection model based on residual network structure and with Octave convolution method to design convolution layers was proposed. Firstly, based on the convolution layer of residual network, the input and output eigenvectors in the convolution layer were divided into high frequencies and low frequencies, and the channel width and height of the low-frequencies were reduced to half of those of the high frequencies. Then, the convolution operation between the low-frequency vector and the high-frequency vector was realized by up-sampling the low-frequency vector with the reduction by half, and the convolution operation between the high-frequency vector and the low-frequency vector was realized by average pooling of the high-frequency vector. Finally, the convolutions between high-frequency vectors and between high-frequency vector and low-frequency vector were added to obtain the high-frequency output, and the convolutions between low-frequency vectors and between low-frequency vector and high-frequency vector were added to obtain the low-frequency output. In this way, Octave convolution layer was constructed, and all convolution layers in residual network were replaced by Octave convolution layers to construct the detection model. In theory, the amount of computation of convolution in Octave convolution layer was reduced by 75%, effectively speeding up the training of the model. On the cloud server with maximum memory of 13 GB and free disk size of 4.9 GB, the PCam (PatchCamelyon) dataset was used for testing. The results show that the model has the recognition accuracy of 95.1%, the memory occupied of 8.7 GB, the disk occupied of 356.4 MB, and the average single training time of 4 minutes 42 seconds. Compared with the ResNet50, this model has the accuracy reduced by 0.6%, the memory saved by 0.6 GB, the disk saved by 105.9 MB, and the single training time shortened by 1 minute. The experimental results demonstrate that the proposed model has high recognition accuracy, short training time and small memory consumption, which reduces the requirement of computing resources under the background of big data era, making the model have application value.

To improve the face detection accuracy of mobile devices, a new real-time face detection algorithm for mobile devices was proposed. The improved Viola-Jones was used for a quick region segmentation to improve segmentation precision without decreasing segmentation speed. At the same time, the optical flow estimation method was used to propagate the features of discrete keyframes extracted by the sub-network of a convolution neural network to other non-keyframes, which increased the efficiency of convolution neural network. Experiments were conducted on YouTube video face database, a self-built one-minute face video database of 20 people and the real test items at different resolutions. The results show that the running speed is between 2.35 frames per second and 22.25 frames per second, reaching the average face detection level; the recall rate of face detection is increased from 65.93% to 82.5%-90.8% at rate of 10% false alarm, approaching the detection accuracy of convolution neural network, which satisfies the speed and accuracy requirements for real-time face detection of mobile devices.

To deal with the problem of the locomotive's dynamically wireless access to transmission network in the mine locomotive unmanned system, the Successive Interference Cancellation (SIC) region partition strategy was proposed. Firstly, the nonlinear region partition model was constructed in the Access Point (AP) communication coverage. Secondly, based on the theoretical derivation, the relationship between the number of region partitions and the AP communication coverage, and the relationship between the locomotive position and the transmission power were found. Finally, SIC region partition strategy was designed. The simulation results show that SIC region partition strategy can make one AP access three locomotives at the same time, and the overall optimization effect of locomotive's total passing time and AP coverage utilization increases at least 50%.

Aiming at the shortcomings of Wolf Pack Algorithm (WPA), such as slow convergence, being easy to fall into local optimum and unsatisfactory artificial wolf interactivity, a wolf pack algorithm based on modified search strategy was proposed, which named Modified Wolf Pack Algorithm (MWPA). In order to promote the exchange of information between the artificial wolves, improve the wolves' grasp of the global information and enhance the exploring ability of wolves, the interactive strategy was introduced into scouting behaviors and summoning behaviors. An adaptive beleaguering strategy was proposed for beleaguering behaviors, which made the algorithm have a regulatory role. With the constant evolution of algorithm, the beleaguered range of wolves decreased constantly and the exploitation ability of algorithm strengthened constantly. Thus the convergence rate of algorithm was enhanced. The simulation results of six typical complex functions of optimization problems show that compared to the Wolf Colony search Algorithm based on the strategy of the Leader (LWCA), the proposed method obtains higher solving accuracy, faster convergence speed and is especially suitable for function optimization problems.

Aiming at the shortcomings of Artificial Bee Colony (ABC) algorithm and its improved algorithms in solving high-dimensional complex function optimization problems, such as low solution precision, slow convergence, being easy to fall in local optimum and too many control parameters of improved algorithms, an improved artificial bee colony algorithm using phased search was proposed. In this algorithm, to reduce the probability of being falling into local extremum, the segmental-search strategy was used to make the employed bees have different characteristics in different stages of search. The escape radius was defined to guide the precocity individual to jump out of the local extremum and avert the blindness of escape operation. Meanwhile, to improve the quality of initialization food sources, the uniform distribution method and opposition-based learning theory were used. The simulation results of eight typical high-dimensional complex functions of optimization problems show that the proposed method not only obtains higher solving accuracy, but also has faster convergence speed. It is especially suitable for solving high-dimensional optimization problems.

In order to mitigate the network congestion by link failures, a new survivability evaluation method named SASFL 〖BP(〗(Survivability Algorithm based on Shuffled Frog Leaping, )〖BP)〗 was proposed by shuffled frog leaping algorithm and wavelet technology. In this method, the evaluation index of survivability was presented at first, and wavelet transform was used to decompose the arrivel flow in failures state. Then, the optimization wavelet coefficients with shuffled frog leaping was reconstructed to network remained traffic. Finally, simulation was conducted to study the relationship between network survivability and failures link, as well as weight factor with OPNET and Matlab. Compared with the other methods, SASFL algorithm has better adaptability.

Feature extraction and similarity measurement for graphs are important issues in computer vision and pattern recognition. However, traditional methods could not describe the graphs under some non-rigid transformation adequately, so a new graph feature descriptor and its similarity measurement method were proposed based on Graph Context (GC) descriptor. Firstly, a sample point set was obtained by discretely sampling. Secondly, graph context descriptor was presented based on the sample point set. At last, improved Earth Mover's Distance (EMD) was used to measure the similarity for graph context descriptor. Different from the graph edit distance methods, the proposed method did not need to define cost function which was difficult to set in those methods. The experimental results demonstrate that the proposed method performs better for the graphs under some non-rigid transformation.

Considering the problem that the abnormal features have great similarity so that simple structure and high precision modeling cannot be achieved, a control chart pattern recognition method based on Kernel Principal Component Analysis (KPCA) and neural network was proposed. Firstly, the kernel method was used to translate the nonlinear feature into a higher dimensional linear feature space. Secondly this feature was projected to lower dimensional feature space. Finally the BP neural network classifier was introduced to identify the control chart pattern. This method was verified through stochastic simulation. The result demonstrates that the model can cluster each control chart pattern effectively and improve recognition accuracy.

An algorithm with multi-features for curves interpolation on adjustable surface was proposed. Similarity parameter of the result surfaces was introduced and the control mesh was modified topologically so as to meet the requirement of curves interpolation and maintain the different similarities between result surface and initial control networks at the same time. Two-side adjustable parameters were introduced. It used the predefined curvature to adjust the bended degree of the curves and created the curvature control formula. The result shows it is available to the curve interpolating adjustable surfaces and the extremely surfaces are diverse.