In order to enhance the robustness of the 3D point cloud face recognition system for multiple expressions and multiple poses, a deep learning-based point cloud feature extraction network was proposed, namely ResPoint. The modules such as grouping, sampling and local feature extraction (ResConv) were used in the ResPoint network, and skip connection was used in ResConv module, so that the proposed network had good recognition results for sparse point cloud. Firstly, the nose tip point was located by the geometric feature points of the face, and the face area was cut with this point as the center. The obtained area had noisy points and holes, so Gaussian filtering and 3D cubic interpolation were performed to it. Secondly, the ResPoint network was used to extract features of the preprocessed point cloud data. Finally, the features were combined in the fully connected layer to realize the classification of 3D faces. In the experiments on CASIA 3D face database, the recognition accuracy of the ResPoint network is increased by 5.06% compared with that of the Relation-Shape Convolutional Neural Network (RS-CNN). Experimental results show that the ResPoint network increases the depth of the network while using different convolution kernels to extract features, so that the ResPoint network has better feature extraction capability.

Aiming at the problem that the existing human skeleton-based action recognition algorithms cannot fully explore the temporal and spatial characteristics of motion, a human skeleton-based action recognition algorithm based on Spatiotemporal Attention Graph Convolutional Network (STA-GCN) model was proposed, which consisted of spatial attention mechanism and temporal attention mechanism. The spatial attention mechanism used the instantaneous motion information of the optical flow features to locate the spatial regions with significant motion on the one hand, and introduced the global average pooling and auxiliary classification loss during the training process to enable the model to focus on the non-motion regions with discriminability ability on the other hand. While the temporal attention mechanism automatically extracted the discriminative time-domain segments from the long-term complex video. Both of spatial and temporal attention mechanisms were integrated into a unified Graph Convolution Network (GCN) framework to enable the end-to-end training. Experimental results on Kinetics and NTU RGB+D datasets show that the proposed algorithm based on STA-GCN has strong robustness and stability, and compared with the benchmark algorithm based on Spatial Temporal Graph Convolutional Network (ST-GCN) model, the Top-1 and Top-5 on Kinetics are improved by 5.0 and 4.5 percentage points, respectively, and the Top-1 on CS and CV of NTU RGB+D dataset are also improved by 6.2 and 6.7 percentage points, respectively; it also outperforms the current State-Of-the-Art (SOA) methods in action recognition, such as Res-TCN (Residue Temporal Convolutional Network), STA-LSTM, and AS-GCN (Actional-Structural Graph Convolutional Network). The results indicate that the proposed algorithm can better meet the practical application requirements of human action recognition.

The core of the Blockchain as a Service(BaaS) platform is how to deploy the blockchain network on the cloud computing platform. Fabric deployment can be divided into static components and dynamic chaincodes according to the component startup time, and chaincode deployment is the core and the most complex part of Fabric cloudification. Because the Fabric has no interfaces for Kubernetes,the current solutions in the industry implement chaincode deployment through a series of auxiliary technologies, but these solutions do not incorporate the chaincodes into the Kubernetes management environment along with static components. In response to the existing problems of BaaS scheme, the following works were mainly done:1) a comprehensive study of the underlying infrastructure, especially of the Kubernetes platform with high availability in the production environment; 2) the cloud deployment of Kubernetes on Fabric was designed and implemented, especially in the chaincode part, a brand-new container control plug-in was used to realize the support for Kubernetes at the code level and complete the goal of incorporating chaincodes into Kubernetes environment management; 3) the functional computing service was used to manage the Fabric chaincodes to realize a brand-new chaincode execution mode, which means changing from the "start-wait-call-wait" mode to the "start-call-exit" mode. The above works in Fabric cloud deployment, especially in chaincode deployment management, have certain reference value for the optimization of the BaaS platform based on Fabric and Kubernetes.

Aiming at the problem that "with the increase of the prediction distance, the prediction of travel time becomes more and more difficult", a comprehensive prediction model of Long Short Term Memory (LSTM) and Artificial Neural Network (ANN) based on spatiotemporal eigenvectors was proposed. Firstly, 24 hours were segmented into 288 time slices to generate time eigenvectors. Secondly, the LSTM time window model was established based on the time slices. This model was able to solve the window movement problem of long-time prediction. Thirdly, the bus line was divided into multiple space slices and the average velocity of the current space slice was used as the instantaneous velocity. At the same time, the predicted time of each space slice would be used as the spatial eigenvector and sent to the new hybrid neural network model named LSTM-A (Long Short Term Memory Artificial neural network). This model combined with the advantages of the two prediction models and solved the problem of bus travel time prediction. Finally, based on the experimental dataset, experiments and tests were carried out:the prediction problem between bus stations was divided into sub-problems of line slice prediction, and the concept of real-time calculation was introduced to each related sub-problem, so as to avoid the prediction error caused by complex road conditions. Experimental results show that the proposed algorithm is superior to single neural network models in both accuracy and applicability. In conclusion, the proposed new hybrid neural network model LSTM-A can realize the long-distance arrival time prediction from the dimension of time feature and the short-distance arrival time prediction from the dimension of spatial feature, thus effectively solving the problem of urban bus travel time prediction and avoiding the remote dependency and error accumulation of buses.

In order to solve the problems of the traditional Hand-Foot-Mouth Disease (HFMD) incidence trend prediction algorithm, such as low prediction accuracy, lack of the combination of other influencing factors and short prediction time, a method of long-term prediction using meteorological factors and Long Short-Term Memory (LSTM) network was proposed. First, the sliding window was used to convert the incidence sequence into the input and output of the network. Then, the LSTM network was used for data modeling and prediction, and the iterative prediction was used to obtain the long-term prediction results. Finally, the temperature and humidity variables were added to the network to compare the impact of these variables on the prediction results. Experimental results show that adding meteorological factors can improve the prediction accuracy of the model. The proposed model has the Mean Absolute Error (MAE) on the Jinan dataset of 74.9, and the MAE on the Guangzhou dataset of 427.7. Compared with the commonly used Seasonal Autoregressive Integrated Moving Average (SARIMA) model and Support Vector Regression (SVR) model, the proposed model has the prediction accuracy higher, which proves that the model is an effective experimental method for the prediction of the incidence trend of HFMD.

Lane line detection is an important part of intelligent driving system. The traditional lane line detection method relies heavily on manual selection of features, which requires a large amount of work and has low accuracy when it is interfered by complex scenes such as object occlusion, illumination change and road abrasion. Therefore, designing a robust detection algorithm faces a lot of challenges. In order to overcome these shortcomings, a lane line detection model based on deep learning instance segmentation method was proposed. This model is based on the improved Mask R-CNN model. Firstly, the instance segmentation model was used to segment the lane line image, so as to improve the detection ability of lane line feature information. Then, the cluster model was used to extract the discrete feature information points of lane lines. Finally, an adaptive fitting method was proposed, and two fitting methods, linear and polynomial, were used to fit the feature points in different fields of view, and the optimal lane line parameter equation was generated. The experimental results show that the method improves the detection speed, has better detection accuracy in different scenes, and can achieve robust extraction of lane line information in various complex practical conditions.

Aiming at the shortage of the airport gate resources and the disturbance caused by the actual flight arrival and departure time deviation from the planned time, a gate assignment scheduling method was proposed by adding buffer time between the adjacent flights in the same gate. Firstly, a robust gate assignment model with a goal to achieve minimum gate idle time and apron occupancy time was established. Then, a Lagrangian relaxation optimization algorithm based on double targets was designed, and the dual problem in the Lagrangian algorithm was solved by using the subgradient algorithm. Based on the operation data of a hub airport in China, the simulation results show that, compared with those of the original gate assignment scheme, the gate usage amount and the gate idle time of the proposed method is respectively reduced by 15.89% and 7.56%, the gate occupancy rate of the optimization scheme of proposed method is increased by 18.72% and the conflict rate is reduced to 3.57%, proving that the proposed method achieves the purpose of effectively improving the utilization and robustness of airport gates.

Focusing on the traffic congestion problem in inland ports with vehicle transportation and large throughput, a collaborative routing method for operation vehicles in inland port based on game theory was proposed. Firstly, the interaction between the operation vehicles that simultaneously request route planning was modeled as a game with incomplete information and the idea of Satisfaction Equilibrium (SE) was applied to analyze the proposed game. It was assumed that every vehicle has an expected utility for routing result, when all vehicles were satisfied, the game achieved an equilibrium. Then, a collaborative routing algorithm was proposed. In this algorithm, firstly every vehicle selected the route according to greedy strategy, then all vehicles were divided into groups by the rule and vehicles in the group performed adaptive learning based on historical routing results to complete the game. The experimental results show that the collaborative routing algorithm reduces the average driving time of vehicles up to 50.8% and 16.3% respectively and improves the system profit up to 51.7% and 24.5% respectively compared with Dijkstra algorithm and Self-Adaptive Learning Algorithm (SALA) when the number of simultaneously working vehicles in port is 286. The proposed algorithm can effectively reduce the average driving time of vehicles, improve system profit, and is more suitable for the routing problem of vehicles in inland port.

Information Retrieval (IR) technology is widely used in automatic tracing from software documents to source codes, but Chinese document and source code are written in different languages, which leads to low accuracy of automatic tracing by using IR. In view of the above problems, an automatic tracing method of Chinese document to source code based on version control was proposed. Firstly, the similarity score between the documents and the source code was calculated by information retrieval method combined with text-to-source heuristic rules. Then the score was modified by the version update information which was submitted to the version control software during software development and maintenance. Finally, the tracing relationship between the Chinese document and source code was determined according to the set threshold. The experimental results show that the precision and recall of the proposed method have a certain improvement compared with the traditional IR method, and the tracing relationship missed in the traditional IR method can be extracted.

For the traditional 3D robot scanners, the measuring precision is dependent on the positioning precision of the robot, and it is difficult to achieve high measuring precision. A photogrammetric method was proposed to track and position the 3D scanning probe accurately. First, a probe tracking system consisting of multiple industrial cameras was set up, and coded markers were pasted on the probe. Then, the camera was calibrated with high precision and interior and exterior parameters of the camera were obtained. Second, all cameras were synchronized, the markers in the image were matched according to the coding principle, and the projection matrix was obtained. Finally, the 3D coordinates of the markers in space were computed to track and position the probe. The experimental results show that the mean error of the marker position is 0.293 mm, the average angle error is 0.136°, and the accuracy of the algorithm is within reasonable range. The photogrammetric method can improve the positioning precision of the probe, so as to achieve high precision measurement.

Pedestrian crowd needs to be divided into groups by using clustering algorithms before using the Social Force Model (SFM) to simulate crowd evacuation. Nevertheless, k-medoids and STatistical INformation Grid (STING) are two traditional clustering algorithms, cannot meet the requirements in the aspect of efficiency and accuracy. To solve the above problem, a new method named Binary Clustering Algorithm (BCA) was proposed in this paper. BCA was composed of two kinds of algorithms:center point clustering and grid clustering. Moreover, the dichotomy was used to divide the grid without repeated clustering. First of all, the data was divided into grids, through the use of dichotomy. Next, the core grid would be selected, according to the data density in a grid. Then, the core grid was used as the center, and the neighbors were clustered. Finally, the residual grids were was merged according to the nearest principle. The experimental results show that, in the clustering time, BCA is only 48.3% of the STING algorithm, less than 14% of the k-medoids algorithm; and in the clustering accuracy, k-medoids is only 50% of BCA, STING doesn't reach to 90% of BCA. Therefore, BCA is better than k-medoids and STING algorithm in both efficiency and accuracy.

Since the existing block-sparse system identification algorithm based on Mean Square Error (MSE) shows poor performance under impulsive interference, an Improved Block Sparse-Normalization Least Mean Square (IBS-NLMS) algorithm was proposed by introducing the inverse hyperbolic sine cost function instead of MSE. A new cost function was constructed and the additive value was obtained by steepest-descent method. Furthermore, a new vector updating equation for filter coefficients was deduced. The adaptive update of the weight vector was close to zero in the presence of impulsive interference, which eliminated the estimation error of adaptive updating based on the wrong information. Meanwhile, mean convergence behavior was analyzed theoretically and then the simulation results demonstrate that in comparison with the Block Sparse-Normalization Least Mean Square (BS-NLMS) algorithm, the proposed algorithm has higher convergence rate and less steady-state error under non-Gaussion noise impulsive interference and abrupt change.

To solve the problem of information loss and obvious block effect in multi-focus image fusion, a novel multi-focus image fusion method based on image matting technique was proposed. Firstly, the rough focus information of each source image was obtained by focusing measure, all of which was used to generate the trimap of the fusion image, namely, foreground, background and unknown region. Secondly, according to the trimap, the precise focus region of each source image could be gotten by using the image matting. Finally, the obtained focus regions were combined to consist of new foreground and background. And the optimal fusion of the unknown region was conducted on the basis of the foreground and background, which enhanced the correlations between the nearby pixels of the three focusing regions. The experimental results show that compared with the traditional algorithms, the proposed algorithm can acquire higher Mutual Information (MI) and edge preservation on the objective evaluation. For the subjective evaluation, it can be seen that the block effect is obviously suppressed and the visual effect is more excellent. The proposed algorithm can be applied to the object identification and computer vision to obtain optimal fusion result.

Since the lackness of diversity may lead to bad performance in ensemble learning, a new two-phase ensemble learning method based on probability calibration was proposed, as well as two methods to reduce the impact of multiple collinearity. In the first phase, the probabilities given by the original classifiers were calibrated using different calibration methods. In the second phase, another classifier was trained using the calibrated probabilities and the final result was predicted. The different calibration methods used in the first phase provided diversity for the second phase, which has been shown to be an important factor to enhance ensemble learning. In order to address the limited improvement due to the correlation between base classifiers, two methods to reduce the multiple collinearity were also proposed, that is, choose-best and bootstrap sampling method. The choose-best method just selected the best base classifier among original and calibrated classifiers; the bootstrap method combined a set of classifiers, which were chosen from the base classifiers with replacement. The experimental results showed that the use of different calibrated probabilities indeed improved the effectiveness of the ensemble; after using the choose-best and bootstrap sampling methods, further improvement was also achieved. It means that probability calibration provides a new way to produce diversity, and the multiple collinearity caused by it can be solved by sampling method.

In order to solve the poor robust problem of Automatic Speech Recognition (ASR) system in noisy environment, a new estimation algorithm of switching speech power spectrum was proposed. Firstly, based on the assumption of the speech spectral amplitude was better modelled for a Chi distribution, a modified estimation algorithm of speech power spectrum based on Minimum Mean Square Error (MMSE) was proposed. Then incorporating the Speech Presence Probability (SPP), a new MMSE estimator based on SPP was obtained. Next, the new approach and the conventional Wiener filter were combined to develop a switch algorithm. With the heavy noise environment, the modified MMSE estimator was used to estimate the clean speech power spectrum; otherwise, the Wiener filter was employed to reduce calculating amount. The final estimation algorithm of switching speech power spectrum for ASR system was obtained. The experimental results show that,compared with the traditional MMSE estimator with Rayleigh prior, the recognition accurate of the proposed algorithm was averagely improved by 8 percentage points in various noise environments. The proposed algorithm can improve the robustness of the ASR system by removing the noise, and reduce the computational cost.

In general blind restoration algorithms, only the gray information of a color image is utilized to estimate the blurring kernel, and thus a restored image may be unsatisfactory if its size is too small or the salient edge in it is too little. Focused on the above mentioned problem, a new blind image restoration algorithm was proposed under a new tensorial framework, in which a color image was regarded as a third-order tensor. First, the blurring kernel was estimated utilizing the multi-scale edge information of blurred color image which could be obtained by adjusting the regularization parameter in tensorial total variation model. Then a deblurring algorithm based on tensorial total variation was adopted to recover the latent image. The experimental results show that the proposed algorithm can achieve obvious improvement on Peak Signal-to-Noise Ratio (PSNR) and subjective vision.

When drift-scan CCD shooting the sky where bright stars are in the filed of view, because of the frame transfer feature, the trailing smear will appear throughout the star image. A fast smear trailing elimination algorithm was proposed by analyzing the imaging mechanism. The method firstly decreased the background non-uniformity by fitting the background, then located smear trailing by calculating the mean gray value of every column in star image and comparing the mean gray values before and after fitting, finally eliminated smear trailing by setting the trailing pixel with the mean gray value after fitting. The experimental results show that the smear trailing is removed completely and the mean deviation of background is apparently reduced, moreover the consuming time of this method is only 20% of that of traditional smear elimination method, which proves the validity of the method.

Due to the diversity of Web users and their complex personal demands, Quality of Service (QoS) information released by some users is not completely reliable, which affects the accuracy of the evaluation on service quality. To address this problem, a service quality evaluation model based on Credible Recommendation (TR-SQE) was presented. In TR-SQE, recommendation trust for the user was defined as the degree of similarity between user's recommendation data and user group's accumulated recommendation data. QoS data released by the user whose recommendations trust was lower than threshold were shielded. By using such correctional QoS information as recommendation data of service quality, then the user, according to the degree of similarity with recommended preference, evaluated service quality. Analysis and simulation results demonstrate that evaluation results from TR-SQE are basically consistent with the real quality of service, which has smaller MAE compared with the contrast methods, and it is helpful to the user's service selection.

Since the gesture signals in gesture interaction are similar and instable, an acceleration gesture recognition method based on Random Projection (RP) was designed and implemented. The system incorporated two parts, one was the training stage and the other was the testing stage. In the training stage, the system employed Dynamic Time Warping (DTW) and Affinity Propagation (AP) algorithms to create exemplars for each gesture; in the testing stage, the method firstly calculated the distance between the unknown trace and all exemplars to find the candidate traces, then used the RP algorithm to translate all the candidate traces and the unknown trace onto the same lower dimensional subspace, and by formulating the whole recognition problem as an l₁-minimization problem, the unknown trace was recognized. The experimental results on 2400 gesture traces show that the proposed algorithm achieves an accuracy rate of 98.41% for specific individuals and 96.67% for unspecific individuals, and it can effectively identify acceleration gestures.

To solve the problem of existing random number generator in high computational and storage cost, a new random number generator was proposed. It generated new any random sequences with longer length by introducing random variable into the process. It has four advantages: simple structure, low computational cost, low storage cost and excellent chaotic property. Besides, it solves the problem of the decimal sequence, limited length of random sequence. The auto-correlation, correlation and probability distribution analysis demonstrates that new Decimal sequence outperforms existing one in random property. These properties make the new random number generator more suitable than the existing complex random number generators for applications in Wireless Sensor Network (WSN), such as chaotic-based and hardware-based random number generators, considering limited computational ability, storage and energy.

A classification method based on trinocular stereovision, which consisted of geometrical classifier and color classifier, was proposed to autonomously guide vehicles on unstructured terrain. In this method, rich 3D data which were taken by stereovision system included range and color information of the surrounding environment. Then the geometrical classifier was used to detect the broad class of ground according to the collected data, and the color classifier was adopted to label ground subclasses with different colors. During the classifying stage, the new classification data needed to be updated continuously to make the vehicle adapt to variable surrounding environment. Two broad categories of terrain what vehicles can drive and can not drive were marked with different colors by using the classification method. The experimental results show that the classification method can make an accurate classification of the terrain taken by trinocular stereovision system.

To solve the problem of gateway copyright protection, a new fast approach was put forward for embedding and extracting gateway watermark in digital video. In intra frame watermarking, the method elegantly detected and selected affine covariant regions in nearly linear complexity. Then, the overlapped affine covariant regions were eliminated based on minimal spanning tree. Last, watermark bits were embedded in Discrete Wavelet Transform (DWT) coefficients in linear time. In inter frame watermarking, the method effectively utilized the continuity inside video scenes to predict the affine covariant regions in no-boundary frames based on boundary frame. The attacking experimental results show that under geometric attacks and format conversion attacks, the accuracies of watermark detection are above 93% and 83% respectively. The simulation results show that in local network with 400 online hosts, the proposed method can block gateway watermark video transmission within 10 frames.

In order to effectively improve the recognition accuracy of the speech emotion recognition system, an improved speech emotion recognition algorithm based on Support Vector Machine (SVM) was proposed. In the proposed algorithm, the SVM parameters, penalty factor and nuclear function parameter, were optimized with genetic algorithm. Furthermore, an emotion recognition model was established with SVM method. The performance of this algorithm was assessed by computer simulations, and 91.03% and 96.59% recognition rates were achieved respectively in seven-emotion recognition experiments and common five-emotion recognition experiments on the Berlin database. When the Chinese emotional database was used, the rate increased to 97.67%. The obtained results of the simulations demonstrate the validity of the proposed algorithm.

Communication of most smart meters is based on MODBUS protocol or 645 protocol, which is insufficient for the transmission of large amounts of data from smart meter. A new reading method and its system based on PROFIBUS were proposed. The system included PROFIBUS based master (Programmable Logic Controller, PLC) and slave stations connected by ROFIBUS bus, and the master station consisted of the Network Interface Card (NIC) of Siemens PLC/CP5611, STEP 7 programming software and WINCC monitoring software; the slave station was a smart meter which contained a PROFIBUS module wired with the Micro Control Unit (MCU) in the smart meter. This PROFIBUS-based reading method and its system of smart meter's data is easy to implement, fast in data transmission, and with communication rate of up to 12Mb/s.

To solve the problems of single and low rate of convolutional codes and large loss of data rate in the G3 standard, the low voltage power line carrier communication system model based on Orthogonal Frequency Division Multiplexing (OFDM) in the G3 standard was analyzed, and a designing scheme of forward error correction encoder was presented based on RS encoding, convolutional encoding, puncturing and depuncturing, repetition encoding and two dimensional time and frequency interleaving algorithm. Moreover, a method for raising the code rate by puncturing and depuncturing was mainly introduced. The simulation results show that the rate of convolutional codes is raised from 1/2 to 2/3, the data rate is improved without increasing the complexity of decoding, and the effective and reliable communication can be realized, which means the scheme can be widely used in low voltage Power Line Communication (PLC).

The checking of bullet's primer is the most important step in controlling the quality of bullet products. In order to segment the image of bullet's primer surface defect accurately, a new method of image segmentation was proposed. According to the checking requirement and the properties of cartridge's bottom, firstly, the image of bullet's primer was ascertained approximately to be detected, and Log operator was applied to extract the circle edge of primer. After analyzing both advantages and disadvantages of the Hough transform and the least square method, a new algorithm of circle detection combined improved Hough transform and the least square method was proposed, by which the center of circle and radius were acquired accurately. Finally, the image of primer circle was extracted by the parameters of circle, the primer surface defect was segmented by threshold, and the results of segmentation were optimized by mathematical morphology. The experimental results show that the proposed method is of accuracy and robustness in the application of bullet's primer surface defect segmentation. The average wrong segmentation rate is below 10%, and the average deviation is less than 17 pixels.

Concerning the problems that traditional path planning of group animation needs long time for searching and is of poor optimization, the authors proposed a multi-threaded path planning algorithm based on group search optimization. Firstly, to solve the problem that the algorithm easily gets trapped in local optimum, metroplis rule was introduced in this search mode. Secondly, by using random path through the multi-threading and stitching techniques, the algorithm was applied to path planning. The simulation results show that the algorithm has better global convergence both in high-dimensional and low-dimensional cases, and the method is good enough to meet the requirements of path planning in complex animation environment.

The major disadvantages of the current software reliability models were discussed. And then based on analyzing classic PSO-SVM model and the characteristics of software reliability prediction, some measures of the improved PSO-SVM model were proposed and an improved model was established. Lastly, the simulation results show that compared with classic models，the improved model has better prediction precision，better generalization ability and lower dependence on the number of sample, which is more applicable for software reliability prediction.

To improve the efficiency of Advanced Encryption Standard （AES） algorithm on ARM processor, aiming at AES algorithm with 128-bit block length and key length, an optimization method was proposed. The method can speed up execution efficiently on ARM processor while consuming less ROM memory. A theoretical analysis of the Rijndael algorithm and of the proposed optimization was discussed. S box was generated by real-time calculation. The MixColumns and InvMixColumns transformations were amended to execute efficiently on 32-bit processor. On-the-fly key expansion was adapted. Simulation results of the optimized algorithm on S3C2440 processor were presented. The experimental results show that the optimization of AES algorithm can execute efficiently on S3C2440 and consume less ROM memory. The method can be applied to embedded systems with memory constraints.

To support Quality of Service (QoS) requirements for different kinds of network in medium and high rate wireless sensor networks, a priority-based service differentiation and rate control strategy was proposed. The strategy gives high priority to the traffic with strong real-time and reliability requirement for service differentiation, calculates the rate difference using the factor ε, and adjusts the sending rate of its upstream node by using hop-by-hop method. The simulation results show the proposed strategy can guarantee high-priority real-time traffic with high throughput and low latency, and maintain the stability of the network throughput.