The traditional Deep Learning （DL）-based multi-objective solvers have the problems of low model utilization and being easy to fall into the local optimum. Aiming at these problems， a Multi-objective Optimization model for Unmanned aerial vehicles Trajectory based on Decomposition and Trajectory search （DTMO-UT） was proposed. The proposed model consists of the encoding and decoding parts. First， a Device encoder （Dencoder） and a Weight encoder （Wencoder） were contained in the encoding part， which were used to extract the state information of the Internet of Things （IoT） devices and the features of the weight vectors. And the scalar optimization sub-problems that were decomposed from the Multi-objective Optimization Problem （MOP） were represented by the weight vectors. Hence， the MOP was able to be solved by solving all the sub-problems. The Wencoder was able to encode all sub-problems， which improved the utilization of the model. Then， the decoding part containing the Trajectory decoder （Tdecoder） was used to decode the encoding features to generate the Pareto optimal solutions. Finally， to alleviate the phenomenon of greedy strategy falling into the local optimum， the trajectory search technology was added in trajectory decoder， that was generating multiple candidate trajectories and selecting the one with the best scalar value as the Pareto optimal solution. In this way， the exploration ability of the trajectory decoder was enhanced during trajectory planning， and a better-quality Pareto set was found. The results of simulation experiments show that compared with the mainstream DL MOP solvers， under the condition of 98.93% model parameter quantities decreasing， the proposed model reduces the distribution of MOP solutions by 0.076%， improves the ductility of the solutions by 0.014% and increases the overall performance by 1.23%， showing strong ability of practical trajectory planning of DTMO-UT model.

For the problem that the principal component vector obtained by Principal Component Analysis (PCA) algorithm is not sparse enough and has many non-zero elements, PCA algorithm was optimized by the reweighting method, and a new method for extracting high-dimensional data features was proposed, namely Reweighted Sparse Principal Component Analysis (RSPCA) algorithm. Firstly, the reweighted l ₁ optimization framework and LASSO (Least Absolute Shrinkage and Selection Operator) regression model were introduced into PCA algorithm to establish a new dimensionality reduction model. Then, the model was solved by using alternat minimization algorithm, singular value decomposition algorithm and minimum angle regression algorithm. Finally, the face recognition experiment was carried out to verify the effectiveness of the algorithm. In the experiment, the K-fold cross-validation method was used to realize the recognition experiment on the ORL face dataset by using PCA algorithm and RSPCA algorithm. The experimental results show that RSCPA algorithm can obtain sparser vector while performs as good as PCA algorithm, has the average recognition accuracy reached 95.1%, which is increased by 6.2 percentage points compared with that of the best performing algorithm sPCA-rSVD (sparse PCA via regularized SVD). And in the experiment of the real-world specific practical application handwritten digit recognition, RSPCA algorithm has the average recognition accuracy of 96.4%, The superiority of the proposed algorithm in face recognition and handwritten digit recognition was proved.

Neuromarketing, by which consumer responses to advertisements and products are analyzed through research on human brain activity, is receiving new attention. Aiming at neuromarketing based on ElectroEncephaloGraphy (EEG), a method of consumer preference prediction based on deep learning neural network was proposed. Firstly, in order to extract features of consumer's EEG, five different frequency bands of EEG topographic videos were obtained from multi-channel EEG signals by using Short Time Fourier Transform (STFT) and biharmonic spline interpolation. Then, a prediction model combining five three-Dimensional Convolutional Neural Networks (3D CNNs) and multi-layer Long Short-Term Memory (LSTM) neural networks was proposed for predicting consumer preference from EEG topographic videos. Compared with the Convolutional Neural Network (CNN) model and LSTM neural network model, the average accuracy of consumer-dependence model was increased by 15.05 percentage points and 19.44 percentage points respectively, and the average accuracy of consumer-independence model was increased by 16.34 percentage points and 17.88 percentage points respectively. Theoretical analysis and experimental results show that the proposed consumer preference prediction system can provide effective marketing strategy development and marketing management at low cost.

In order to reduce the harm caused by bursty events in social media, a multi-feature based microblog bursty events detection algorithm was proposed. The algorithm combines text emotion filtering and user influence calculation methods. Firstly, the microblog text with negative emotion was obtained through noise filtering and emotion filtering. Then the proposed user influence calculation method was combined with the burst word extraction algorithm to extract the characteristics of burst words. Finally, a cohesive hierarchical clustering algorithm was introduced to cluster bursty word sets, and extract bursty events from them. In the experimental test, the accuracy is 66.84%, which proves that the proposed method can effectively detect bursty events.

To solve the problems of long time consuming and large space occupation in the route planning process of Unmanned Underwater Vehicle (UUV) under complex environments and multi-constraint conditions, a new route planning method based on Charged System Search (CSS) was proposed. Firstly, the UUV route planning problem model was established, and the cost function was designed. Then, a route planning method based on CSS was presented, and the charged particle was affected by the electric field force of other charged particles in the search space to achieve the purpose of iterative optimization. In addition, a method of nonlinear adjustment of speed and acceleration parameters was proposed, which could effectively balance global search and local search processes, and avoid premature convergence of algorithm. Finally, the proposed method was compared with the route planning methods based on A ^* algorithm, ant colony optimization algorithm and particle swarm optimization from two respects of the quality of planning route and the time consuming of algorithm. The experimental results show that, the proposed method has faster convergence speed and lower time complexity while ensuring the quality of planned route.

In the industrial control network, there are some known anomaly behaviors and some unknown anomaly behaviors in network communication. The white list method can effectively detect the known abnormal behaviors in the rule library, but the detection rate of unknown anomaly behaviors is low. In order to improve the detection rate on the basis of full mining of valid information, an intrusion detection method combining white list filtering and neural network unsupervised learning algorithm named AMPSO-BP was proposed to apply on routers between the servers of manage network and industrial network. Firstly, the white list technology was used to filter the communication behaviors that could not match with the white list rules base at first time; then the results of sample training by offline unsupervised learning in neural network system were used to filter the abnormal communication behaviors that trusted with the white list at second time. The neural network was used to improve the detection rate under incomplete information, and according to the neural network detection results, the white list rule library was improved constantly to promote the detection rate of abnormal communication over network. The Particle Swarm Optimization algorithm with Adaptive Mutation (AMPSO) was used as training function for the BP (Back Propagation) neural network, and the adaptive mutation process was added to the Particle Swarm Optimization (PSO) algorithm to avoid falling into the local optimal solution prematurely during the training process. Two groups of training and testing data sets were used in experiment. The experimental results show that the detection accuracy of AMPSO-BP combined with white list is higher than that of PSO-BP combined with white list.

Focusing on the electricity user group feature, a residential electricity consumption analysis method based on geographic regularized matrix factorization in smart grid was proposed to explore the characteristics of electricity users and provide decision support for personalized better power dispatching. In the proposed algorithm, customers were firstly mapped into a hidden feature space, which could represent the characteristics of users' electricity behavior, and then k-means clustering algorithm was employed to segment customers in the hidden feature space. In particular, geographic information was innovatively introduced as a regularization factor of matrix factorization, which made the hidden feature space not only meet the orthogonal characteristics of user groups, but also make the geographically close users mapping close in hidden feature space, consistent with the real physical space. In order to verify the effectiveness of the proposed algorithm, it was applied to the real residential data analysis and mining task of smart grid application in Sino-Singapore Tianjin Eco-City (SSTEC). The experimental results show that compared to the baseline algorithms including Vector Space Model (VSM) and Nonnegative Matrix Factorization (NMF) algorithm, the proposed algorithm can obtain better clustering results of user segmentation and dig out certain power modes of different user groups, and also help to improve the level of management and service of smart grid.

Aiming at the problem of information fusion between Light Detection And Ranging (LiDAR) data and camera images in the detection process of Unmanned Ground Vehicle (UGV) following the target vehicle, a method of joint calibration of LiDAR and camera based on a trapezoidal checkerboard was proposed. Firstly, by using the scanning information of the LiDAR in the trapezoidal calibration plate, the LiDAR installation angle and installation height were accessed. Secondly, the external parameters of the camera relative to the body were calibrated through the black and white checkerboard on the trapezoidal calibration plate. Then combining with the correspondence between the LiDAR data points and the pixel coordinates of the image, two sensors were jointly calibrated. Finally, integrating the LiDAR and the camera calibration results, the pixel data fusion of the LiDAR data and the camera image was carried out. As long as the trapezoidal calibration plate was placed in front of the vehicle body, the image and the LiDAR data were collected only once in the entire calibration process of two types of sensors. The experimental results show that the proposed method has high calibration accuracy with average position deviation of 3.5691 pixels (13 μm), and good fusion effect of LiDAR data and the visual image. It can effectively complete the spatial alignment of LiDAR and the camera, and is strongly robust to moving objects.

Aiming at the problems of scalability, algorithm convergence performance and operational efficiency in building large-scale machine learning systems, the challenges of the large-scale sample, model and network communication to the machine learning system were analyzed and the solutions of the existing systems were also presented. Taking Latent Dirichlet Allocation (LDA) model as an example, by comparing three open source distributed LDA systems-Spark LDA, PLDA+ and LightLDA, the differences in system design, implementation and performance were analyzed in terms of system resource consumption, algorithm convergence performance and scalability. The experimental results show that the memory usage of LightLDA and PLDA+ is about half of Spark LDA, and the convergence speed is 4 to 5 times of Spark LDA in the face of small sample sets and models. In the case of large-scale sample sets and models, the network communication volume and system convergence time of LightLDA is much smaller than PLDA+ and SparkLDA, showing a good scalability. The model of "data parallelism+model parallelism" can effectively meet the challenge of large-scale sample and model. The mechanism of Stale Synchronous Parallel (SSP) model for parameters, local caching mechanism of model and sparse storage of parameter can reduce the network cost effectively and improve the system operation efficiency.

Focusing on the issue that the HBase storage without spatio-temporal index degrades the traffic data query performance, some HBase spatio-temporal indexes based on row keys were proposed for massive traffic data. Firstly, the dimensionality reduction method based on Geohash was used to convert two-dimensional spatial position data into a one-dimensional code. Then the code was combined with the temporal dimension. Secondly, four index models were put forward based on combination order, and the structures of the models and their adaption conditions for traffic data query were discussed. Finally, the algorithm of index creation as well as traffic data query algorithm was proposed. Experimental results show that the proposed HBase spatio-temporal index structure can effectively enhance the traffic data query performance. In addition, the query performance of four different spatio-temporal index structures in different data size, different query radius and different query time range were compared, which verified the different adaption scenes of different index structures in traffic data query.

To improve the availability of dynamic trust model in trusted network, a trusted network management model based on clustering analysis was built. Behavior expectations were used to describe the trust of user behavior by introducing clustering analysis to the traditional trust model. Clustering analysis of the user's historical data was used to build behavior expection model, which was used to evaluate user's behaviors. Finally the trust evaluation results were utilized to realize the network user management. The experimental results show that the proposed model can generate trust evaluation results firmly, detect and isolate the malicious users rapidly, it has better accuracy and efficiency than traditional model, basically improving the network reliability.

In view of the neural network's initial structure set depends on the workers experience and its adaptive ability is poor, a dynamic neural network structure design method based on Semi-Supervised Learning (SSL) algorithm was proposed. In order to get a more perfect performance of the initial network structure, the authors trained neural network based on semi-supervised learning method of using both tagged sample and unmarked sample, and judged the impact of the hidden layer neurons on the network output by using Global Sensitivity Analysis method (GSA). The optimal design of dynamic neural network structure was accomplished by cutting or increasing hidden layer neurons based on sensitivity size timely, and the convergence of the dynamic process was investigated. Theoretical analysis and Matlab simulation experiments show that the neural network hidden layer neurons based on Semi-Supervised Learning algorithm will change with training time, and the structure design of the dynamic network is accomplished. The application of hydraulic Automatic Gauge Control (AGC) system, about 160 s later, the system output is becoming stable, and the output error is as small as about 0.03 mm, and compared with Supervised Learning (SL) method and UnSupervised Learning (USL) method, the output error reduces by 0.03 mm and 0.02 mm respectively, which indicate that dynamic network based on SSL algorithm effectively improve the precision of the system output in actual applications.

In order to improve the robustness and accuracy of relative orientation, an approach combining direct resolving and iterative refinement for relative orientation was proposed. Firstly, the essential matrix was estimated from some corresponding points. Afterwards the initial relative position and posture of two cameras were obtained by decomposing the essential matrix. The process for determining the only position and posture parameters were introduced in detail. Finally, by constructing the horizontal epipolar coordinate system, the constraint equation group was built up from the corresponding points based on the coplanar constraint, and the initial position and posture parameters were refined iteratively. The algorithm was resistant to the outliers by applying the RANdom Sample Consensus (RANSAC) strategy and dynamically removing outliers during iterative refinement. The simulation experiments illustrate the resolving efficiency and accuracy of the proposed algorithm outperforms that of the traditional algorithm under the circumstance of importing varies of random errors. And the experiment with real data demonstrates the algorithm can be effectively applied to relative position and posture estimation in 3D reconstruction.

A low power branch encoding method was presented for decreasing the SoC bus power dissipation. This method's basic principle is: for the address bus, when the address bus is sequential, the address bus is frozen, and when the address bus is non-sequential, the window size is adjusted dynamically to apply the Bus-Invert (BI) method on the address bus. For the data bus, two threshold values are figured out for different data size respectively. If the Hamming distance locates between these two threshold values, the valid-data-channel switching dense area is found and inverted, otherwise applies the BI encoding. This method's encoding and decoding circuits are realized in the Advanced High Performance Bus (AHB) system. The experimental result demonstrates that compared with uncoded situation, this method decreases the address/data bus toggle rate by 51.2%/22.4%, and the system power is reduced by 28.9%. Compared with T0,BI and other encoding methods realized in the same system, the branch encoding is more superior in the toggle rate and power dissipation.

Image layer visual representation has been currently used in computer vision field, but it is difficult for feed-forward image multilayer visual representation methods to deal with local ambiguities. An image multilayer visual representation method based on Latent Dirichlet Allocation (LDA) named LDA-IMVR was proposed. It derived a recursive generative model of LDA by implementing recursive probabilistic decomposition process. Meanwhile, it learned and deduced all layers of the hierarchy together, and improved classification and learning performance by using feed-back style. The approach was tested on Caltech 101 dataset. The experimental results show that the proposed method improves classification performance of objects compared with related hierarchical approaches, and it achieves better results in learned components and image patches visualization.

Lossless video coding is increasingly used because of the need of high quality videos in digital video areas. For this reason, a lossless video compression algorithm based on fuzzy logic was designed in this paper. It utilized fuzzy-logic-based method to calculate the correlation between two subblocks from neighbor frames and the interior correlation in the subblock, which can be used to decide the selection between temporal prediction and spatial prediction. A new matching rule of motion estimation was defined in temporal prediction. At last, the correlation can be adopted to estimate the parameter of Golomb coding and realize fast and efficient Golomb coding without complex calculation of estimation. The experimental results show that the proposed method has significant improvement in coding efficiency compared with the JPEG-LS.

The parallel runway system is the most popular configuration in airport multi-runway construction. Four types of representative parallel runways (space between 400m, 760m, 920m and 1525m) and corresponding operation modes were simulated by using Simmod. The Ultimate Capacity (UC) and Factual Operation Capacity (FOC) of parallel runways were analyzed taking consideration of runway crossing and non-runway crossing. The simulation result shows that in instrument flight rule, the FOC of widely spaced parallel runways was about 74 sorties per hour, the FOC of medially spaced parallel runways was about 65 sorties per hour, and that of closely spaced parallel runways was only more than 40 sorties per hour. In condition of non-runway crossing, the capacity of parallel runway system would be increased by 13%.

The existing dimensionality reduction methods based on manifold learning are sensitive to the selection of local neighbors, and the reduced data do not have good separability. This paper proposed a dimensionality reduction method with data separability based on adaptive neighborhood selection, which adaptively selected the neighborhood at each sample point based on estimated intrinsic dimensionality of data and local tangent orientation. Meanwhile, it clustered the similar sample points by using clustering information when mapping data, which guaranteed good separability for the reduced data and achieved better dimensionality reduction results. The experimental results show that the new method derives a better embedding result on the artificially generated data sets. In addition, it can get expected result on face visualization classification and image retrieval.

To prolong the lifetime in wireless sensor networks, the maximum lifetime routing based on ant colony algorithm (MLRAC) was proposed. This paper used the link energy consumption model and node transmission data probability to calculate the node total energy consumption in a data gathering cycle. Considering the node initial energy, MLRAC formulized the maximum lifetime routing into an optimal model. To solve the optimal model, the revised ant colony algorithm based on classical algorithm was proposed. It used new formulas for forwarding the probability of neighbor nodes and pheromone update, and a new method for packet detection. After some iterations of calculation, it could obtain the optimal network lifetime value and each node transmission data probability. Finally, sink node informed any other nodes in network with flooding. According to the received optimal probability, all nodes selected the neighbor node where the data packets did not pass to transmit data. The simulation results show that after a certain time of iteration, network lifetime of MLRAC reaches certain convergence. It can prolong network lifetime. Under certain conditions, MLRAC outperforms PEDAP-PA, LET, Sum-w and Ratio-w algorithms.

The available tools for detecting network information can hardly meet the demands of acquiring the completeness and precision of network information for the researchers.The information fusion technology was applied to collect the network information using several detecting tools.The information from different detecting tools was fused in different layers.In data layer,the fuzzy logical statistic method was adopted to identify system type and network device,and in logic layer,the most credible information was obtained with the support of system knowledge database.