Aiming at the management requirement of real-time personnel statistics in office scenes with relatively fixed personnel positions, a laboratory personnel statistics and management system based on Faster Region-based Convolutional Neural Network (Faster R-CNN) and Intersection over Union (IoU) optimization was designed and implemented with an ordinary university laboratory as the example. Firstly, Faster R-CNN model was used to detect the heads of the people in the laboratory. Then, according to the output results of the model detection, the repeatedly detected targets were filtered by using IoU algorithm. Finally, a coordinate-based method was used to determine whether there were people at each workbench in the laboratory and store the corresponding data in the database. The main functions of the system are as follows:① real-time video surveillance and remote management of the laboratory; ② timed automatic photo, detection and acquisition of data to provide data support for the quantitative management of the laboratory; ③ laboratory personnel change data query and visualization. The experimental results show that the proposed laboratory personnel statistics and management system based on Faster R-CNN and IoU optimization can be used for real-time personnel statistics and remote management in office scenes.

Most of the current image classification methods use supervised learning or semi-supervised learning to reduce image dimension. However, supervised learning and semi-supervised learning require image carrying label information. Aiming at the dimensionality reduction and classification of unlabeled images, a mixed-order feature stacked sparse autoencoder was proposed to realize the unsupervised dimensionality reduction and classification learning of the images. Firstly, a serial stacked sparse autoencoder network with three hidden layers was constructed. Each hidden layer of the stacked sparse autoencoder was trained separately, and the output of the former hidden layer was used as the input of the latter hidden layer to realize the feature extraction of image data and the dimensionality reduction of the data. Secondly, the features of the first hidden layer and the second hidden layer of the trained stacked autoencoder were spliced and fused to form a matrix containing mixed-order features. Finally, the support vector machine was used to classify the image features after dimensionality reduction, and the accuracy was evaluated. The proposed method was compared with seven comparison algorithms on four open image datasets. The experimental results show that the proposed method can extract features from unlabeled images, realize image classification learning, reduce classification time and improve image classification accuracy.

Spiking neuron is a novel artificial neuron model. The purpose of its supervised learning is to stimulate the neuron by learning to generate a series of spike sequences for expressing specific information through precise time coding, so it is called spike sequence learning. Because the spike sequence learning for single neuron has the characteristics of significant application value, various theoretical foundations and many influential factors, the existing spike sequence learning methods were reviewed and contrasted. Firstly, the basic concepts of spiking neuron models and spike sequence learning were introduced. Then, the typical learning methods of spike sequence learning were introduced in detail, the theoretical basis and synaptic weight adjustment way of each method were pointed out. Finally, the performance of these learning methods was compared through experiments, the characteristics of each method was systematically summarized, the current research situation of spike sequence learning was discussed, and the future direction of development was pointed out. The research results are helpful for the comprehensive application of spike sequence learning methods.

It is difficult for existing rumor detection methods to overcome the disadvantage of data collection and detection delay. To resolve this problem, a rumor detection method based on burst topic detection inspired by the momentum model and domain expert discovery was proposed. The dynamics theory in physics was introduced to model the topic features spreading among the Weibo platform, and dynamic physical quantities of the topic features were used to describe the burst characteristics and tendency of topic development. Then, emergent topics were extracted after feature clustering. Next, according to the domain relativity between the topic and the expert, domain experts for each emergent topic were selected within experts pool, which is responsible for identifying the credibility of the emergent topic. The experimental results show that the proposed method gets 13 percentage points improvement on accuracy comparing with the Weibo rumor identification method based merely on supervised machine learning, and the detection time is reduced to 20 hours compared with dominating manual methods, which means that the proposed method is applicable for real rumor detection situation.

To solve the virtual machine deployment problem in Infrastructure as a Service (IaaS) of cloud computing, a virtual machine deployment strategy based on Particle Swarm Optimization (PSO) algorithm was proposed. Since the PSO algorithm has weaknesses of having a slow convergence speed and falling into local optimum easily when dealing with large-scale and complex problems like virtual machine deployment, firstly, a Multiple-population Gaussian Learning Particle Swarm Optimization (MGL-PSO) algorithm was proposed, with using the model of multiple population evolution to accelerate the algorithm convergence, as well as adding Gaussian learning strategy to avoid local optimum. Then according to the deployment model, with using Round Robin (RR) algorithm to initialize the MGL-PSO, a virtual machine deployment strategy aiming to load balancing was proposed. Through the simulation experiment in CloudSim, it validates that MGL-PSO has a higher convergence speed and load imbalance degree is reduced by 13% compared with PSO algorithm. In the two experimental situations, compared with the Opportunistic Load Balancing (OLB) algorithm, the load imbalance degrees of the proposed algorithm decrease by 25% and 15% respectively, and compared with the Greedy Algorithm (GA) the load imbalance degrees decrease by 19% and 7% respectively.

In the smart grid, the development of electric power Demand Response (DR) brings great change to the traditional power utilization mode. Combined with real-time electricity price, consumers can adjust their power utilization mode by their energy demand. This makes load forecasting more complicated. The multi-input and two-output Least Squares Support Vector Machine (LS-SVM) was proposed to preliminarily predict the load and price at the same time. Considering the interaction between the real-time electricity price and load, the fuzzy recursive inference system based on data mining technology was adopted to simulate the game process of the forecasting of the price and load, and then the preliminary forecast results of multi-variable LS-SVM prediction algorithm were recursively corrected until the forecasting results were tending towards stability. Multi-variable LS-SVM can avoid running into local optima and has an excellent capacity of generalization, the improved association rules mining algorithm and loop predictive control algorithm have good completeness and robustness, and can correct the forecasting result approximately in every real situation. Simulation results of the actual power system show that the proposed method has better application effects.

In order to improve the processing ability for uncertainty data using the traditional Fuzzy Support Vector Machine (FSVM), FSVM with fuzzy similarity measure and high dimensional space fuzzy mapping was proposed. Firstly, by using Gregson similarity measure, the fuzzy similarity measure function was established, which was effective to explain the uncertainty information. And then, using the theory of mapping and Mercer, fuzzy similarity kernel learning was formulated and used in the algorithm of the FSVM. Finally, this algorithm was used to the modeling of the material removal rate in the rotary ultrasonic machining with uncertainty data. Compared to the results using traditional FSVM methods, the current approach can better process uncertainty data with less operation steps. And the proposed method has higher accuracy in processing uncertainty data with lower computational complexity.

The UMHexagonS motion estimation algorithm in H.264 was studied, and an improved fast motion estimation algorithm was proposed. First, the fixed search range, the unsymmetrical cross search, the 5×5 small rectangular spiral search, the uneven multi-hexagon-grid search and the extended hexagon-based search were analyzed. Then the optimized search modes were given respectively, which called dynamic search window, adaptive rood pattern search, the directional 3×3 small rectangular search pattern, the predictive intensive direction search and the modified extended hexagon-based search. Thus Adaptive Pattern Direction Search (APDS) algorithm was formed by these optimized search modes. The experimental results conducted on different test sequences show that, compared to UMHexagonS algorithm, the APDS algorithm can save about 29.64% Motion Estimation (ME) time and reduce the average number of checking points per Motion Vector (MV) generation about 21.64, while incurring nothing obvious loss in the reconstructed picture quality and less increment in the bit rate. With the efficiency improvement of ME, the real-time performance of the encoder is further enhanced.

In the field of cloud computing, it is a key problem that how task schedules. This paper presented an artificial immune algorithm for dynamic task scheduling on cloud computing platform. Firstly, the algorithm used the queuing theory to determine the conditions of cloud computing platform to maintain steady-state, and provided the basic data for the following algorithm. Then, this paper used the clone selection algorithm to search out the approximate optimal configuration which calculated resources for different virtual machines of different nodes in the cluster. Finally, proper load balancing processing algorithm joined with immune theory improved the antibody genes. The results of simulation experiment show that, this algorithm can effectively improve the convergence speed and accuracy, search reasonable allocation quickly and improve the cluster resource utilization.

Current ultrasound image denoising algorithms cannot maintain edge well while denoising. An improved anisotropic diffusion denoising method called anisotropic diffusion based on Logarithmic Compression (LCAD) was proposed to reduce ultrasound speckle noise after the study of anisotropic diffusion model. The proposed method estimated noise distribution model after logarithmic compression of the image and then generated a diffusion coefficient based on generalized Gamma distribution to achieve denoising purpose while diffusing.

An adaptive inter-layer rate control scheme was proposed for H.264/AVC scalable extension. A switched model was put forward to predict the number of bits used for encoding inter frame either from the previous frame of the current layer or from the current frame of the previous layer. First, a Rate-Complexity-Quantization (R-C-Q) model was extended in scalable video coding. Second, the Proportional+Integral+Derivative (PID) buffer controller was adopted to provide the inter frame bit estimation according to the buffer state. Third, to achieve more accurate prediction when an abrupt change happens, the bit estimation was predicted from the actual bit of the current frame of the previous layer. Finally, a switched model was used to decide the bit estimation, and then the Quantization Parameter (QP) could be calculated according to the R-C-Q model. The simulation results demonstrate that the proposed algorithm outperforms JVT-W043 rate control algorithm by providing more accurate output bit rate for each layer, maintaining stable buffer fullness, reducing frame skipping and quality fluctuation, and improves the overall coding quality.

The decline of recommending quantity resulted in the extreme sparsity of user data was dealt with,based on nearest neighbor algorithm and the combination of selective sampling and lookahead framework for recommendation system,with the purpose of improving the accuracy of recommendation.The algorithm was put into experiment,with much lower average error rate and higher stability.

A human face recognition technique based on modular 2DPCA was presented. First, the original images were divided into modular images in proposed approach. Then the 2DPCA method could be directly used to the sub-images obtained from the previous step. There are three advantages for this way: 1)dimension reduction of discriminant features can be done conveniently; 2)singular value decomposition of matrix is fully avoided in the process of feature extraction, so the features for recognition can be gained easily; 3)as opposed to 2DPCA, the feature matrix of lower dimension can be employed, and higher (not less at least) correct recognition rate can be reached. Moreover, 2DPCA is the special case of modular 2DPCA. To test modular 2DPCA and evaluate its performance, a series of experiments were performed on three human face image databases: ORL and NJUST603 human face databases. The experimental results indicated that the performance of modular 2DPCA is superior to that of 2DPCA.

In DHT algorithm of structured P2P architecture, to increase the routing performance and decrease the maintenance cost when peers join/leave the network, a recommendatory policy-based search algorithm(RPSA) was put forward. RPSA improved the existed DHT algorithm,and the performance of it was validated on the P-Grid prototype system.

A remote real-time management system for FC-RAID 3000 based on embedded Web server GoAhead was designed and implemented. The system communicated with remote user through GoAhead, which accepted the user’s requests and responded with corresponding Web pages stored in the minimal file system. When the user logged in, the remote management module would authenticates the user’s authority; and when the user sent configuring and monitoring command, the remote management module would parse the command and execute it by calling the functions provided by RAID controlling module. The system can efficiently improve the management performance of FC-RAID 3000.