Aiming at the problems of premature convergence and low recall caused by using greedy strategy for community expansion in the existing local community detection algorithms， a local community detection algorithm based on Monte-Carlo iterative solving strategy was proposed. Firstly， in the community expansion stage of each iteration， the selection probabilities were given to all adjacent candidate nodes according to the contribution ratio of each node to the community tightness gain， and one node was randomly selected to join the community according to these probabilities. Then， in order to avoid random selection causing the expansion direction to deviate from the target community， it was determined whether the node elimination mechanism was triggered in this round of iteration according to the changes in community quality. If it was triggered， the similarity sum of each node joining the community and other nodes in the community was calculated， the elimination probabilities were assigned according to the reciprocal of the similarity sum， a node was randomly eliminated according to these probabilities. Finally， whether to continue the iteration was judged on the basis of whether the community size increased in a given number of recent iteration rounds. Experimental results show that， on three real network datasets， compared to Local Tightness Expansion （LTE） algorithm， Clauset algorithm， Common Neighbors with Weighted Neighbor Nodes （CNWNN） algorithm and Fuzzy Similarity Relation （FSR） algorithm， the proposed algorithm has the F-score value of local community detection results increased by 32.75 percentage points， 17.31 percentage points， 20.66 percentage points and 25.51 percentage points respectively， and can effectively avoid the influence of the location of the query node in the community on the local community detection results.

Concerning the problems of the lack of a large number of relationships in the current Knowledge Graph (KG), and the lack of full consideration of the hidden information in the multi-hop path between two entities when performing relationship reasoning, a relationship reasoning method combining multi-hop relationship path information was proposed. Firstly, for the given candidate relationships and two entities, the convolution operation was used to encode the multi-hop relationship path connecting the two entities into a low-dimensional space and extract the information. Secondly, the Bidirectional Long Short-Term Memory (BiLSTM) network was used for modeling to generate the relationship path representation vector, and the attention mechanism was used to combine it with the candidate relationship representation vector. Finally, a multi-step reasoning method was used to find the relationship with the highest matching degree as the reasoning result and judge its precision. Compared with the current popular Path Ranking Algorithm (PRA), the neural network model named Path-RNN and reinforcement learning model named MINERVA, the proposed algorithm had the Mean Average Precision (MAP) increased by 1.96,8.6 and 1.6 percentage points respectively when using the large-scale knowledge graph dataset NELL995 for experiments. And when using the small-scale knowledge graph dataset Kinship for experiments, the proposed algorithm had the MAP improved by 21.3,13 and 12.1 percentage points respectively compared to PRA and MINERVA. The experimental results show that the proposed method can infer the relationship links between entities more accurately.

For the problem of time, effort and money consuming to obtain a large number of samples by conventional means faced by Artificial Intelligence (AI) application research in different fields, a variety of sample augmentation methods have been proposed in many AI research fields. Firstly, the research background and significance of data augmentation were introduced. Then, the methods of data augmentation in several common fields (including natural image recognition, character recognition and discourse parsing) were summarized, and on this basis, a detailed overview of sample acquisition or augmentation methods in the field of medical image assisted diagnosis was provided, including X-ray, Computed Tomography (CT), Magnetic Resonance Imaging (MRI) images. Finally, the key issues of data augmentation methods in AI application fields were summarized and the future development trends were prospected. It can be concluded that obtaining a sufficient number of broadly representative training samples is the key to the research and development of all AI fields. Both the common fields and the professional fields have conducted sample augmentation, and different fields or even different research directions in the same field have different sample acquisition or augmentation methods. In addition, sample augmentation is not simply to increase the number of samples, but to reproduce the existence of real samples that cannot be completely covered by small sample size as far as possible, so as to improve sample diversity and enhance AI system performance.

News text usually consists of tens to hundreds of sentences, which has a large number of characters and contains more information that is not relevant to the topic, affecting the classification performance. In view of the problem, a long text classification method combined with attention mechanism was proposed. Firstly, a sentence was represented by a paragraph vector, and then a neural network attention model of paragraph vectors and text categories was constructed to calculate the sentence's attention. Then the sentence was filtered according to its contribution to the category, which value was mean square error of sentence attention vector. Finally, a classifier base on Convolutional Neural Network (CNN) was constructed. The filtered text and the attention matrix were respectively taken as the network input. Max pooling was used for feature filtering. Random dropout was used to reduce over-fitting. Experiments were conducted on data set of Chinese news text classification task, which was one of the shared tasks in Natural Language Processing and Chinese Computing (NLP&CC) 2014. The proposed method achieved 80.39% in terms of accuracy for the filtered text, which length was 82.74% of the text before filtering, yielded an accuracy improvement of considerable 2.1% compared to text before filtering. The emperimental results show that combining with attention mechanism, the proposed method can improve accuracy of long text classification while achieving sentence level information filtering.

Currently, two vectorization methods which exploit Simple Instruction Multiple Data (SIMD) parallelism are loop-based method and Superword Level Parallel (SLP) method. Focusing on the problem that the current compiler cannot realize function level vectorization, a method of function level vectorization based on static single assignment was proposed. Firstly, the variable properties of program were analysed, and then a set of compiling directives including SIMD function annotations, uniform clauses, linear clauses were used to realize function level vectorization. Finally, the vectorized code was optimized by using the variable attribute result. Some test cases from the field of multimedia and image processing were selected to test the function and performance of the proposed function level vectorization on Sunway platform. Compared with the scalar program execution results, the execution of the program after the function level vectorization is more efficient. The experimental results show that the function level vectorization can achieve the same effect of task level parallelism, which is instructive to realize the automatic function level vectorization.

To solve the problem that Circulant Structure of tracking-by-detection with Kernels (CSK) is difficult to adapt to the target scale change, a multi-scale kernel correlation filter classifier was proposed to realize the scale adaptive target tracking. Firstly, the multi-scale image was used to construct the sample set, the multi-scale kernel correlation filtering classifier was trained by the sample set, for target size estimation to achieve the goal of the optimal scale detection, and then the samples collected on the optimal target scale were used to update the classifier on-line to achieve the scale-adaptive target tracking. The comparative experiments and analysis illustrate that the proposed algorithm can adapt to the scale change of the target in the tracking process, the error of the eccentricity is reduced to 1/5 to 1/3 that of CSK algorithm, which can meet the needs of long time tracking in complex scenes.

Concerning the influence of various environmental ambiances on sound event recognition, a sound event recognition method based on Optimized Orthogonal Matching Pursuit (OOMP) and Deep Belief Network (DBN) was proposed. Firstly, Particle Swarm Optimization (PSO) algorithm was used to optimize Orthogonal Matching Pursuit (OMP) sparse decomposition of sound signal, which realized fast sparse decomposition of OMP and reserved the main body of sound signal and reduced the influence of noise. Then, an optimized composited feature was composed by Mel-Frequency Cepstral Coefficient (MFCC), time-frequency OMP feature and Pitch feature extracted from the reconstructed sound signal, which was called OOMP feature. Finally, the DBN was employed to learn the OOMP feature and recognize 40 classes of sound events in different environments and Signal-to-Noise Ratio (SNR). The experimental results show that the proposed method which combined OOMP and BDN is suitable for sound event recognition in various environments, and can effectively recognize sound events in various environments; it can still maitain an average accuracy rate of 60% even when the SNR is 0 dB.

When calculating curved Ricci Flow, non-convergence emerges due to the existence of undersized angles in triangular meshes. Concerning the problem of non-convergence, a Delaunay triangulation subdivision algorithm of spherical convex graph of enhancing the minimum angle was proposed. First of all, the Delaunay triangulation subdivision algorithm of spherical convex graph was given. The proposed algorithm had two key operations:1) if a Delaunay minor arc was "encroached upon", a midpoint of the Delaunay minor arc was added to segment the Delaunay minor arc; 2) if there was a "skinny" spherical triangle, it was disassembled by adding the center of minor circle of its circumscribed sphere. Then, the convergence criteria of the proposed algorithm was explored on local feature scale and an upper-bound formula of the output vertex was given. The grids based on the output of experiment show that the spherical triangle generated by the grids of the proposed algorithm has no narrow angle, so it is suitable for calculating Ricci Flow.

In consideration of intelligence algorithms' extensive applicability to various types of feed formulation optimization models, the Artificial Fish Swarm Algorithm (AFSA) was firstly applied in feed formulation optimization. For meeting the required precision of feed formulation optimization, a symbiotic system-based AFSA was employed. which significantly improved the convergence accuracy and speed compared with the original AFSA. In the process of optimization, the positions of Artificial Fish (AF) individuals in solution space were directly coded as the form of solution vector to the problem via the feed ratio, a penalty-based objective function was employed to evaluate AF individuals' fitness. AF individuals performed several behavior operators to explore the solution space according to a predefined behavioral strategy. The validity of the proposed algorithm was verified on three practical instances. The verification results show that, the proposed algorithm has worked out the optimal feed formulation, which can not only remarkably reduce the fodder cost, but also satisfy various nutrition constraints. The optimal performance of the proposed algorithm is superior to the other existing algorithms.

Concerning the problem that jobs of high proportion of resources execute inefficiently in job scheduling algorithms of the present hierarchical queues structure, the resource matching maximum set algorithm was proposed. The proposed algorithm analysed job characteristics, introduced the percentage of completion, waiting time, priority and rescheduling times as urgent value factors. Jobs with high proportion of resources or long waiting time were preferentially considered to improve jobs fairness. Under the condition of limited amount of available resources, the double queues was applied to preferentially select jobs with high urgent values, select the maximum job set from job sets with different proportion of resources in order to achieve scheduling balance. Compared with the Max-min fairness algorithm, it is shown that the proposed algorithm can decrease average waiting time and improve resource utilization. The experimental results show that by using the proposed algorithm, the running time of the same type job set which consisted of jobs of different proportion of resources is reduced by 18.73%, and the running time of jobs of high proportion of resources is reduced by 27.26%; the corresponding percentages of reduction of the running time of the mixed-type job set are 22.36% and 30.28%. The results indicate that the proposed algorithm can effectively reduce the waiting time of jobs of high proportion of resources and improve the overall jobs execution efficiency.

Concerning the problem that the efficiency of serial PageRank algorithm is low in dealing with mass Web data, a PageRank parallel algorithm based on Web link classification was proposed. Firstly, the Web was classified according to its Web link, and the weights of different Web which was from diverse websites were set variously. Secondly, with the Hadoop parallel computation platform and MapReduce which has the characteristics of dividing and conquering, the Webpage ranks were computed parallel. At last, a data compression method of three layers including data layer, pretreatment layer and computation layer was adopted to optimize the parallel algorithm. The experimental results show that, compared with the serial PageRank algorithm, the accuracy of the proposed algorithm is improved by 12% and the efficiency is improved by 33% in the best case.

To improve the accuracy of bird sounds recognition in low Signal-to-Noise Ratio (SNR) environment, a new bird sounds recognition technology based on Radon Transform (RT) and Translation Invariant Discrete Wavelet Transform (TIDWT) from spectrogram after the noise reduction was proposed. First, an improved multi-band spectral subtraction method was presented to reduce the background noise. Second, short-time energy was used to detect silence of clean bird sound, and the silence was removed. Then, the bird sound was translated into spectrogram, RT and TIDWT were used to extract features. Finally, classification was achieved by Support Vector Machine (SVM) classifier. The experimental results show that the method can achieve better recognition effect even the SNR belows 10dB.

As regards the poor local optimization ability of Particle Swarm Optimization (PSO), a nonlinear dynamic adjusting inertia weight was put forward to improve the particle swarm path planning algorithm. This algorithm combined the grid method and particle swarm algorithm, introduced the two concepts of safety and smoothness based on path length, and established dynamic adjustment path length of the fitness function. Compared with the traditional PSO. The experimental results show that the improved algorithm has stronger security, real-time and optimization ability.

For the purpose of improving the recognition accuracy of bird sounds in various kinds of noisy environments in real world, a new bird sounds recognition approach based on energy detection was proposed. First of all, the useful bird sound signals were detected and selected by the method of energy detection from the bird sounds with noises. Secondly, according to the distribution of Mel scale, the feature of Wavelet Packet decomposition Subband Cepstral Coefficient (WPSCC) was extracted from the useful signals. Finally, the classifier of Support Vector Machine (SVM) was applied to model on the WPSCC and Mel-Frequency Cepstral Coefficient (MFCC) respectively for classification and identification. Meanwhile, the comparisons of recognition performance difference were implemented on 15 kinds of bird sounds at different Signal-to-Noise Ratio (SNR) in different noises, before or after energy detection. The experimental results show that WPSCC has better noise immunity function, and the recognition performance after energy detection can be greatly improved, which means it is more suitable for the bird sounds recognition in complex environments.

This paper proposed a motion-adaptive de-interlacing algorithm with high performance based on the analysis of the advantages and disadvantages of traditional de-interlacing algorithm for video surveillance systems. The algorithm divided the picture into static region and motion region on the basis of the motion state of interpolation points through 4-field motion detection which could detect the spatial-periodic pattern moving. Field insertion algorithm was exploited for interpolation of the static region. A modified edge-adaptive interpolation algorithm was used for the interpolation of the motion region which could increase the function of horizontal edge detection and enhance the level of consistency edge direction estimation. The proposed interpolation algorithm was implemented on DSP for experimental verification. The results show that the algorithm improves Peak Signal-to-Noise Ratio (PSNR) and Structural SIMilarity (SSIM) and restrains saw-tooth, interline flicker, motion virtual image and other adverse effects and gets bettter visual effects.

Node deployment is a basic problem in sensor networks, which directly relates to the performance of the entire network. Most existing researches on sensor network node deployment are for the case of twodimensional planar and three dimensions space, but very few researches for threedimensional surface deployment scenario. This paper proposed an algorithm of optimal surface deployment in wireless sensor networks. First by mathematical or differential geometry method for threedimensional surface it constructed mathematical model, and then through the centroid of the threedimensional surface Voronoi subdivision partitions, an error function was proposed to evaluate the superiority of deployment method. Finally compared with other surface deployment methods, the performance of the proposed algorithm in this paper is superior.

To improve the calculation speed and robustness of the Speaker Recognition (SR) system, the authors proposed a speaker recognition algorithm method based on utterance level Principal Component Analysis (PCA), which was derived from the frame level features. Instead of frame level features, this algorithm used the utterance level features in both training and recognition. What's more, the PCA method was also used for dimension reduction and redundancy removing. The experimental results show that this algorithm not only gets a little higher recognition rate, but also suppresses the effect of the noise on speaker recognition system. It verifies that the algorithm based on utterance level features PCA can get faster recognition speed and higher system recognition rate, and it enhances system recognition rate in different noise environments under different Signal-to-Noise Ratio (SNR) conditions.

To meet the physiological perception of human eyes, a color image enhancement algorithm based on improved Intersecting Cortical Model (ICM) was proposed. The internal activities and dynamic threshold were improved to nonlinear attenuation, which satisfied the nonlinear perception of human eyes. And the decay factor was replaced by the step factor, while maintaining some of the significant features of the original model. It applied the Threshold Versus Intensity (TVI) function of the human visual system on the intensity component of the input image to adjust the dynamic range compression. At the same time, it also adjusted the saturation component of the input image by nonlinearity. Compared to the original ICM, this algorithm reduced the complexity and improved the adaptability. The experimental results confirm that the method can obtain clear and bright results.

An unsupervised extraction method is proposed in order to extract bead-like particles regions of fly ash from scanning electron microscope image, which is based on region growing with gray similarity bounded by gradient and shape. The process is automatic, including seeds selecting , regions growing and shape distinguishing. The experimental error is measured by the acreage probability of missing segmentation and false segmentation. The minimum error rate of the experimental results is 6.8%, and the average error rate is 8%. The time of extraction from 60 SEM images is within 10 minutes. The method is effective for the content estimate of fly ash in the material.

Concerning the disadvantage of funneling-MAC protocol, this paper gave an improved proposal of DRAND algorithm. Based on the centralized Time Division Multiple Access (TDMA) scheduling algorithm of funneling-MAC protocol, it introduced the DRAND scheme, which guaranteed nodes did not overlap within the time slots in two-hop range, so it could greatly avoid interference and collision. The NS-2 simulation results show that the improved protocol can effectively reduce system power consumption, and maintain higher channel utilization.

To improve the efficiency of neighbor discovery in wireless networks with directional antennas, a busy-tone aided algorithm was proposed. With the help of omni-directional busy-tones, the problems of collision and idleness in wireless communication were effectively resolved; as a result, the channel utilization ratio was increased. The direction of antenna beam was adjusted according to the Direction of Arrival (DOA) of busy-tones. Through this strategy, communication efficiency was improved. The experimental results show that, compared with the conventional ALOHA-like algorithm and the neighbor discovery algorithm based on the feedback mechanism, the proposed algorithm has a better performance.

Traditional network management can not meet the needs of the Next Generation Network (NGN), so service-oriented network management is the inevitable development trend. Based on the research of active network technology and using advanced ideas of active network technology and telecommunication management network, we proposed the organization model, the function model, the communication model and the information model. To propose a new network management model: service-oriented network management model based on active network technology. The network management system based on this model can preferably and effectively manage network services.

To solve the problem of query-by-example in multimedia audio data, the characteristics of wavelet multiresolution, wavelet packet transform and its best base were analyzed. A method for audio retrieval was proposed using wavelet frame coefficients of packet best base and wavelet multiresolution pyramidal algorithm. First, audio data files were prepocessed by transforming them into frame coefficients of best base and wavelet coefficient files with audio data. And then elementary classification for these files was carried out using frame coefficients of best base, and after that these files were searched using the different hierarchy pyramidal algorithms. By comparing our method with the method using different level wavelet approximate coefficient algorithm, it is found that our method is highly efficient and reduces the searching time without influencing the retrieval precision.

Some problems of the current measures for association rules were analyzed. A new measure named consult was defined and added to the mining algorithm for association rules. According to the value of consult, association rules were classified into positive, negative and invalid association rules. The new algorithm could find out the negative-item-contained rules. Finally, the algorithm was evaluated and analyzed through experiments and practices.