In today’s society with the growing demand for privacy protection， federated learning is receiving widespread attention. However， in federated learning， it is difficult for the server to supervise behaviors of clients， so that the existence of lazy clients poses a potential threat to the performance and fairness of federated learning. Aiming at the problem of how to identify lazy clients efficiently and accurately， a dual-task proof-of-work method based on backdoor was proposed， namely FedBD （FedBackDoor）. In FedBD， additional backdoor tasks that are easier to detect were allocated by the server for the clients participating in federated learning， the backdoor tasks were trained by the clients based on the original training tasks， and the clients’ behaviors were supervised by the server indirectly through training status of the backdoor tasks. Experimental results show that FedBD has certain advantages over the classic federated averaging algorithm FedAvg and the advanced algorithm GTG-Shapley （Guided Truncation Gradient Shapley） on datasets such as MNIST and CIFAR10. On CIFAR10 dataset， when the proportion of lazy clients is 15%， FedBD improves the accuracy by more than 10 percentage points compared with FedAvg， and increases the accuracy by 2 percentage points compared with GTG-Shapley. Moreover， the average training time of FedBD is only 11.8% of that of GTG-Shapley， and the accuracy of FedBD in identifying lazy clients can exceed 99% when the proportion of lazy clients is 10%. It can be seen that FedBD can solve the problem of lazy clients being difficult to supervise.

The existing similarity-based moving target trajectory prediction algorithms are generally classified according to the spatial-temporal characteristics of the data， and the characteristics of the algorithms themselves cannot be reflected. Therefore， a classification method based on algorithm characteristics was proposed. The calculation of the distances between two points is required for the trajectory similarity algorithms to carry out the subsequent calculations， however， the commonly used Euclidean Distance （ED） is only applicable to the problem of moving targets in a small region. A method of similarity calculation using geodetic distance instead of ED was proposed for the trajectory prediction of sea targets moving in a large region. Firstly， the trajectory data were preprocessed and segmented. Then， the discrete Fréchet Distance （FD） was adopted as similarity measure. Finally， synthetic and real data were used to test. Experimental results indicate that when sea targets move in a large region， the ED-based algorithm may gain incorrect prediction results， while the geodetic distance-based algorithm can output correct trajectory prediction.

In the method of combining Object detection Network (ObjectNet) and scene recognition network, the object features extracted by the ObjectNet and the scene features extracted by the scene network are inconsistent in dimensionality and property, and there is redundant information in the object features that affects the scene judgment, resulting in low recognition accuracy of scenes. To solve this problem, an improved indoor scene recognition method combined with object detection was proposed. First, the Class Conversion Matrix (CCM) was introduced into the ObjectNet to convert the object features output by ObjectNet, so that the dimension of the object features was consistent with that of the scene features, as a result, the information loss caused by inconsistency of the feature dimensions was reduced. Then, the Context Gating (CG) mechanism was used to suppress the redundant information in the features, reducing the weight of irrelevant information, and increasing the contribution of object features in scene recognition. The recognition accuracy of the proposed method on MIT Indoor67 dataset reaches 90.28%, which is 0.77 percentage points higher than that of Spatial-layout-maintained Object Semantics Features (SOSF) method; and the recognition accuracy of the proposed method on SUN397 dataset is 81.15%, which is 1.49 percentage points higher than that of Hierarchy of Alternating Specialists (HoAS) method. Experimental results show that the proposed method improves the accuracy of indoor scene recognition.

Thread-Level Speculation (TLS) technology can tap the parallel execution potential of programs and improve the utilization of multi-core resources. However, the current TACLeBench kernel benchmarks are not effectively analyzed in TLS parallelization. In response to this problem, the loop-level speculative execution analysis scheme and analysis tool were designed. With 7 representative TACLeBench kernel benchmarks selected, firstly, the initialization analysis was performed to the programs, the program hot fragments were selected to insert the loop identifier. Then, the cross-compilation was performed to these fragments, the program speculative thread and the memory address related data were recorded, and the maximun potential of the loop-level parallelism was analyzed. Finally, the program runtime characteristics (thread granularity, parallelizable coverage, dependency characteristics) and the impacts of the source code on the speedup ratio were comprehensively discussed. Experimental results show that:1) this type of programs is suitable for TLS acceleration, compared with serial execution results, under the loop structure speculative execution, the speedup ratios for most programs are above 2, and the highest speedup ratio in them can reach 20.79; 2) by using TLS to accelerate the TACLeBench kernel programs, most applications can effectively make use of 4-core to 16-core computing resources.

In order to make full use of appearance and motion information in video abnormal behavior detection, a Siamese network model that can capture appearance and motion information at the same time was proposed. The two branches of the network were composed of the same autoencoder structure. Several consecutive frames of RGB images were used as the input of the appearance sub-network to predict the next frame, while RGB frame difference image was used as the input of the motion sub-network to predict the future frame difference. In addition, considering one of the reasons that affected the detection effect of the prediction-based method, that is the diversity of normal samples, and the powerful "generation" ability of the autoencoder network, that is it has a good prediction effect on some abnormal samples. Therefore, a memory enhancement module that learns and stores the "prototype" features of normal samples was added between the encoder and the decoder, so that the abnormal samples were able to obtain greater prediction error. Extensive experiments were conducted on three public anomaly detection datasets Avenue, UCSD-ped2 and ShanghaiTech. Experimental results show that, compared with other video abnormal behavior detection methods based on reconstruction or prediction, the proposed method achieves better performance. Specifically, the average Area Under Curve (AUC) of the proposed method on Avenue, UCSD-ped2 and ShanghaiTech datasets reach 88.2%, 97.5% and 73.0% respectively.

Sentiment analysis of massive text information on social networks can better mine the behavior rules of Internet users,helping decision-making institutions understand the public opinion tendencies and helping businesses improve the quality of service. The task of Chinese implicit sentiment classification is more difficult than those of other languages due to the absence of key emotional features,expression vector forms and cultural customs. The existing Chinese implicit sentiment classification methods are mainly based on Convolutional Neural Network(CNN),and have some defects, such as the inability to obtain the sequence of words and not using contextual emotional features reasonably in implicit emotion discrimination. A Chinese implicit sentiment classification model combining sequence and contextual features named GGBA (GCNN-GRU-BiGRU-Attention) was proposed to solve the above problems. In the model, Gated Convolutional Neural Network (GCNN) was used to extract the local important information of sentences with implicit sentiments,and Gated Recurrent Unit(GRU)network was used to enhance the temporal information of features. In the context feature processing of sentences with implicit sentiments,the combination of Bidirectional Gated Recurrent Unit (BiGRU)and attention was used to extract the important emotional features. After obtaining the two types of features,the contextual important features were integrated into the implicit emotion discrimination through the fusion layer. Experimental results on the implicit sentiment analysis evaluation dataset showed that the macro average precision of GGBA model was 3. 72% higher than that of normal text CNN named TextCNN,2. 57% higher than that of GRU,and 1. 90% higher than that of Disconnected Recurrent Neural Network(DRNN). Therefore,GGBA model achieves better classification performance than the basic models in implicit sentiment analysis tasks.

For the monitoring dead angles of fixed surveillance cameras installed in large quantities and low hardware performance of mobile devices, an urban management case target recognition algorithm that can run on IOS mobile devices with low performance was proposed. Firstly, the number of channels of input and output images and the number of feature maps generated by each channel were optimized by adding new hyperparameters to MobileNet. Secondly, a new recognition algorithm was formed by combining the improved MobileNet with the SSD recognition framework and was transplanted to the IOS mobile devices. Finally, the accurate detection of the common 8 specific urban management case targets was achieved by the proposed algorithm, in which the camera provided by the mobile device was used to capture the scene video. The mean Average Precision (mAP) of the proposed algorithm was 15.5 percentage points and 10.4 percentage points higher than that of the prototype YOLO and the prototype SSD, respectively. Experimental results show that the proposed algorithm can run smoothly on low-performance IOS mobile devices, reduce the dead angles of monitoring, and provide technical support for urban management team to speed up the classification and processing of cases.

To efficiently detect outliers in massive hydrologic sensor data, an anomaly detection method for hydrological time series based on SparkR was proposed. Firstly, a sliding window and Autoregressive Integrated Moving Average (ARIMA) model were used to forecast the cleaned data on SparkR platform. Then, the confidence interval was calculated for the prediction results, and the results outside the interval range were judged as anomaly data. Finally, based on the detection results, K-Means algorithm was used to cluster the original data, the state transition probability was calculated, and the anomaly data were evaluated in quality. Taking the data of hydrologic sensor obtained from the Chu River as experimental data, experiments on the detection time and outlier detection performance were carried out respectively. The results show that the millions of data calculation by two slaves costs more time than that by one slave, but when calculating the tens of milllions of data, the time costed by two slaves is less than that by one slave, and the maximum reduction is 16.21%. The sensitivity of the evaluation is increased from 5.24% to 92.98%. It shows that under big data platform, the proposed algorithm which is based on the characteristics of hydrological data and combines forecast test and cluster test can effectively improve the computational efficiency of hydrologic time series detection for tens of millions data and has a significant improvement in sensitivity.

In recent years, the Generative Adversarial Network (GAN) has achieved remarkable success in text-to-image synthesis, but there are still problems such as edge blurring of images, unclear local textures, small sample variance. In view of the above shortcomings, based on Stack Generative Adversarial Network model (StackGAN++), a Multi-Level structure Generative Adversarial Networks (MLGAN) model was proposed, which is composed of multiple generators and discriminators in a hierarchical structure. Firstly, hierarchical structure coding method and word vector constraint were introduced to change the condition vector of generator of each level in the network, so that the edge details and local textures of the image were clearer and more vivid. Then, the generator and the discriminator were jointed by trained to approximate the real image distribution by using the generated image distribution of multiple levels, so that the variance of the generated sample became larger, and the diversity of the generated sample was increased. Finally, different scale images of the corresponding text were generated by generators of different levels. The experimental results show that the Inception scores of the MLGAN model reached 4.22 and 3.88 respectively on CUB and Oxford-102 datasets, which were respectively 4.45% and 3.74% higher than that of StackGAN++. The MLGAN model has improvement in solving edge blurring and unclear local textures of the generated image, and the image generated by the model is closer to the real image.

Interpolation algorithms are usually used to estimate the channel frequency response value at the data location in Long Term Evolution-Advanced (LTE-A) system. Concerning the problem that traditional Linear Minimum Mean Square Error (LMMSE) algorithm needs to obtain channel statistical properties in advance and it suffers from a high computational complexity due to an inversion matrix operation, an improved LMMSE channel estimation interpolation algorithm was proposed. Firstly, the pilots were interpolated to add virtual pilots, which improved the performance of the algorithm. Secondly, an approximate estimation method of autocorrelation matrix and Signal-to-Noise Ratio (SNR) was given by using the fact that channel energy in the time domain is more concentrated. Finally, a sliding window method was adopted to further simplify the algorithm complexity to complete the LMMSE interpolation in frequency domain. The simulation results show that the overall performance of the proposed algorithm is better than that of linear interpolation method and Discrete Fourier Transform (DFT) interpolation method, and it has similar Bit Error Rate (BER) and Mean Squared Error (MSE) performance with the traditional LMMSE interpolation algorithm. Furthermore, it reduces the complexity by 98.67% compared with traditional LMMSE estimator without degrading the overall BER and MSE performance, so it is suitable for practical engineering applications.

For complex battlefield environment is lack of stable end-to-end communication path between user terminals, a Delay Tolerant Network (DTN) clustering routing algorithm based on Vehicular Ad Hoc NETwork (VANET) communication terminals and motion information named CVCTM (Cluster based on VANET Communication Terminals and Motion information) was proposed. Firstly, the clustering algorithm based on cluster head election was studied; secondly, the routing algorithm for intra-cluster source vehicle was studied based on hop count, relay mode and geographical location information. Then, the routing algorithm for inter-cluster source vehicle was realized by introducing waiting time, threshold of retransmission times and downstream cluster heads. Finally, the optimal way of communicating with upper headquarter was chosen by VANET communication terminals. The ONE simulation results show that the message delivery ratio of CVCTM increased nearly 5%, the network overhead of it decreased nearly 10%, the recombination times of cluster structure decreased nearly 25% in the comparison with AODV (Ad Hoc On-demand Distance Vector routing); the message delivery ratio of CVCTM increased nearly 10%, the network overhead of it decreased nearly 25%, the recombination times of cluster structure decreased nearly 40% in the comparison with CBRP (Cluster Based Routing Protocal) algorithm and DSR (Dynamic Source Routing) protocal. CVCTM can effectively reduce network overhead and recombination times of cluster structure and increase message delivery ratio.

Concerning unsatisfactory classification accuracy and low efficiency of the existing pattern-based classification methods for model training, a concept of density-aware pattern and an algorithm for biological sequence classification based on density-aware patterns, namely BSC (Biological Sequence Classifier), were proposed. Firstly, frequent sequence patterns based on density-aware concept were mined. Then, the mined frequent sequence patterns were filtered and sorted for designing the classification rules. Finally, the sequences without classification were classified by classification rules. According to a number of experiments conducted on four real biological sequence datasets, the influence of BSC algorithm parameters on the results were analyzed and the recommended parameter settings were provided. Meanwhile, the experimental results showed that the accuracies of BSC algorithm were improved by at least 2.03 percentage points compared with other four pattern-based baseline algorithms. The results indicate that BSC algorithm has high biological sequence classification accuracy and execution efficiency.

Aiming at the uniqueness of head feature and high detection error rate extracted by traditional zenithal pedestrian detection method, a multi-feature fusion zenithal pedestrian detection algorithm based on improved Aggregate Channel Feature (ACF) and Gray-Level Co-occurrence Matrix (GLCM) was proposed. Firstly, the extracted Hue, Sturation, Value (HSV) color features, gradient magnitude and improved Histogram of Oriented Gradients (HOG) feature were combined into ACF descriptor. Then, the improved GLCM parameter descriptor was calculated by the window method, and the texture features were extracted. The co-occurrence matrix feature descriptor was obtained by concatenating the feature vectors of each window. Finally, the aggregate channel and co-occurrence matrix features were input into Adaboost for training to get the classifier, and the final results were obtained by detection. The experimental results show that, the proposed algorithm can effectively detect targets in the presence of interference background, and improve the detection precision and recall.

To solve the problems existing in the off-line construction method of location fingerprint database for location fingerprint positioning based on Received Signal Strength Indication (RSSI), including large workload of collecting all the fingerprint information in the location, low construction efficiency of the location fingerprint database, and the limited precision of interpolation, a high efficient off-line construction method of the location fingerprint database based on the Singular Value Thresholding (SVT) Matrix Completion (MC) algorithm improved by the Backtracking Search optimization Algorithm (BSA) was proposed. Firstly, using the collected location fingerprint data of some reference nodes, a low-rank matrix completion model was established. Then the model was solved by the low rank MC algorithm based on the SVT. Finally, the complete location fingerprint database could be reconstructed in the location area. At the same time, the BSA was introduced to improve the optimization process of MC algorithm with the minimum kernel norm as the fitness function to solve the problem of the fuzzy optimal solution and the poor smoothness of the traditional MC theory, which could further improve the accuracy of the solution. The experimental results show that the average error between the location fingerprint database constructed by the proposed method and the actual collected location fingerprint database is only 2.7054 dB, and the average positioning error is only 0.0863 m, but nearly 50% of the off-line collection workload can be saved. The above results show that the proposed off-line construction method of the location fingerprint database can effectively reduce the workload of off-line collection stage while ensuring the accuracy, significantly improve the construction efficiency of location fingerprint database, and improve the practicability of fingerprint positioning method to a certain extent.

Focusing on the data sparsity and low accuracy of recommendation existed in traditional Collaborative Filtering (CF) recommendation algorithm, a micro blog User Recommendation algorithm based on the Similarity of Multi-source Information, named MISUR, was proposed. Firstly, the micro blog users were classified by K-Nearest Neighbor ( KNN) algorithm according to their tag information. Secondly, the similarity of the multi-source information, such as micro blog content, interactive relationship and social information, was calculated for each user in each class. Thirdly, the time weight and the richness weight were introduced to calculate the total similarity of multi-source information, and the TOP- N recommendation was used in a descending order. Finally, the experiment was carried out on the parallel computing framework Spark. The experimental results show that, compared with CF recommendation algorithm and micro blog Friend Recommendation algorithm based on Multi-social Behavior (MBFR), the superiority of the MISUR algorithm is validated in terms of accuracy, recall and efficiency.

In view of that the traditional Slope One algorithm does not consider the influence of project attribute information and time factor on project similarity calculation, and there exists high computational complexity and slow processing in current large data background, a weighted Slope One algorithm based on clustering and Spark framework was put forward. Firstly, the time weight was added to the traditional item score similarity calculation, and comprehensive similarity was computed with the similarities of the item attributes. And then the set of nearest neighbors was generated through combining with the Canopy- K-means algorithm. Finally, the data was partitioned and iterated to realize parallelization by Spark framework. The experimental results show that the improved algorithm based on the Spark framework is more accurate than the traditional Slope One algorithm and the Slope One algorithm based on user similarity, which can improve the operating efficiency by 3.5-5 times compared with the Hadoop platform, and is more suitable for large-scale dataset recommendation.

Chinese addresses on the Internet are always non-normalized, which cannot be used directly in location-based services. To solve the problem, an algorithm to extract administrative divisions from non-normalized Chinese addresses was proposed. Firstly, preprocessing "road" feature word grouping for original data; using administrative division dictionary and moving window maximum matching algorithm, extract all possible administrative region data sets from Chinese address. Then, using the Chinese administrative divisions between the elements of the hierarchical relationship between the characteristics, the administrative set conditional set operation rule was established and the acquired data set was aggregated. using the administrative division of matching, a set of administrative division set rules were established to calculate the credibility of the administrative division. Finally, the credibility of the maximum amount of information the most complete Chinese address of the administrative divisions were obtained. By using the extracted from the Internet about 250000 Chinese address data whether the use of "road" feature word packet processing and whether to carry on the credibility calculation process was verified for the availability of the algorithm, and with the current address matching technology for comparison, the accuracy rate of 93.51%.

As a widely used recommendation algorithm in the industry, collaborative filtering algorithm can predict the likely favorite items based on the user's historical behavior records. However, the traditional collaborative filtering algorithms do not take into account the drifting of user interests, and there are also some deficiencies when the recommendation's timeliness is considered. To solve these problems, the measure method of similarity was improved by combining with the characteristics of user interests change with time. At the same time, an enhanced time attenuation model was introduced to measure the predictive value. By combining these two ways together, the concept drifting problem of user interests was solved and the timeliness of the recommendation algorithm was also considered. In the simulation experiment, predictive scoring accuracy and Top N recommendation accuracy were compared among the proposed algorithm, UserCF, TCNCF, PTCF and TimesSVD++ algorithm in different data sets. The experimental results show that the improved algorithm can reduce the Root Mean Square Error (RMSE) of the prediction score, and it is better than all the compared algorithms on the accuracy of Top N recommendation.

The original K-means clustering algorithm is seriously affected by initial centroids of clustering and easy to fall into local optima. To solve this problem, an improved K-means clustering algorithm based on Adaptive Cuckoo Search (ACS), namely ACS-K-means, was proposed, in which the search step of cuckoo was adjusted adaptively so as to improve the quality of solution and boost speed of convergence. The performance of ACS-K-means clustering was firstly evaluated on UCI dataset, and the results demonstrated that it surpassed K-means, GA-K-means (K-means based on Genetic Algorithm), CS-K-means (K-means based on Cuckoo Search) and PSO-K-means (K-means based on Particle Swarm Optimization) in clustering quality and convergence rate. Finally, the ACS-K-means clustering algorithm was applied to the development of heat map of urban management cases of Qingxiu district of Nanning city, the results also showed that the proposed method had better quality of clustering and faster speed of convergence.

A hashtag based method was proposed to solve the problem how to accurately extract keywords from microblog. Hashtag, the social feature of a microblog was used to extract keywords from microblog content. A word-post weighted graph was built firstly, then a random walker was used on the graph by jumping to any hashtag node repeatedly. At last, every word rank was determined by its probability which would not change after walker iteration. The experiments were conducted on real microblogs from Sina platform. The results show that, compared to word-word graph method, the proposed hashtag-based approach gets higher accuracy of keyword extraction by 50%.

The time complexity of K-Nearest Neighbor( KNN) classification algorithm is proportional to the number of training samples, which needs a large number of computation, and the bottleneck of slow processing exists in traditional architecture under the big data background. In order to solve the problems, a highly efficient algorithm of KNN based on Spark framework and clustering was proposed. Firstly, the training set was cut twice by the optimized K-medoids algorithm through introducing constriction factor. Then the K was iterated constantly in the process of classification and the classification result was obtained. And the data was partitioned and iterated to realize parallelization combining the Spark framework in the calculation. The experimental results show that, the classification time of the traditional KNN algorithm and the KNN algorithm based on K-medoids is 3.92-31.90 times of the proposed algorithm in different datasets. The proposed algorithm has high computational efficiency and better speedup ratio than KNN based on Hadoop platform, and it can effectively classify the big data.

In modeling and simulation of meteor burst communication system, concerning the problem of network delay caused by Hybrid Automatic Repeat Request (HARQ), an estimation model of transmission delay based on HARQ was proposed. Firstly, in consideration of the network structure and channel characters in meteor burst communication, a network delay model was constructed by analyzing the theory of HARQ. Then, based on queuing theories, the improvement mechanism of HARQ was introduced to establish an estimation model of transmission delay of Type-Ⅰ HARQ and one of Type-Ⅱ HARQ. Finally, the simulation was realized to compare and analyze the transmission delay performance of two kinds of HARQ. When packet transmission accuracy or packet transmission time changes independently, the transmission delay of Type-Ⅱ HARQ is less than that of Type-Ⅰ HARQ. The experimential results show that Type-Ⅱ HARQ has advantages of network delay performance in meteor burst communication compared to Type-Ⅰ HARQ.

In order to improve the locating accuracy in Wireless Senor Network (WSN) node localization, an algorithm based on Particle Swarm Optimization (PSO) and Shuffled Frog Leaping Algorithm (SFLA) was proposed, namely Bi-system Cooperative Optimization (BCO) algorithm. With the advantages of fast convergence in PSO and high optimization precision in SFLA, the proposed algorithm was easier to converge through less iterations and achieve higher accuracy of depth search. The simulation experiments indicate that the BCO algorithm is effective. First, the BCO algorithm can be very close to the optimal solution when it is used for solving the test target functions with better locating accuracy and higher convergence speed. Meanwhile, when the proposed algorithm is used for node localization based on Received Signal Strength Indicator (RSSI), the absolute distance error of the prediction location and the actual location is less than 0.05 meters. Compared with the Improved Particle Swarm Optimization algorithm based on RSSI (IPSO-RSSI), the locating accuracy of the proposed algorithm can be increased 10 times at least.

Focusing on the safety analysis of the 3D block cipher, a new method on this algorithm against the meet-in-the-middle attack was proposed. Based on the structure of the 3D algorithm and the differential properties of the S-box, the research reduced the number of required bytes during structuring the multiple sets in this attack and constructed a new 6-round meet-in-the-middle distinguisher. According to extending the distinguisher 2-round forward and 3-round backward, an 11-round meet-in-the-middle attack of the 3D algorithm was finally achieved. The experimental results show that:the number of required bytes on constructed the distinguisher is 42, the attack requires a data complexity of about 2⁴⁹⁷ chosen plaintexts, a time complexity of about 2^325.3 11-round 3D algorithm encryption and a memory complexity of about 2³⁴² bytes. The new attack shows that the 11-round of the 3D algorithm is not immune to the meet-in-the-middle attack.

The spatial index structure and the query technology plays an important role in the spatial database. According to the disadvantages in the approximation and organization of the complex spatial objects of the existing methods, a new index structure based on Minimum Bounding Rectangle (MBR), trapezoid and circle (RTC (Rectangle Trapezoid Circle) tree) was proposed. To deal with the Nearest Neighbor (NN) query of the complex spatial data objects effectively, the NN query based on RTC (NNRTC) algorithm was given. The NNRTC algorithm could reduce the nodes traversal and the distance calculation by using the pruning rules. According to the influence of the barriers on the spatial data set, the barrier-NN query based on RTC tree (BNNRTC) algorithm was proposed. The BNNRTC algorithm first queried in an idea space and then judged the query result. To deal with the dynamic simple continuous NN chain query, the Simple Continues NN chain query based on RTC tree (SCNNCRTC) algorithm was given. The experimental results show that the proposed methods can improve the efficiency of 60%-80% in dealing with large complex spatial object data set with respect to the query method based on R tree.

For the difficulty of complex non-linear system modeling, a new system modeling algorithm based on the Takagi-Sugeno (T-S) Fuzzy Radial Basis Function (RBF) neural network optimized by improved Particle Swarm Optimization (PSO) algorithm was proposed. In this algorithm, the good interpretability of T-S fuzzy model and the self-learning ability of RBF neural network were combined together to form a T-S fuzzy RBF neural network for system modeling, and the network parameters were optimized by the improved PSO algorithm with dynamic adjustment of the inertia weight combined with recursive least square method. Firstly, the proposed algorithm was used to do the approximation simulation of a non-linear multi-dimensional function, the Mean Square Error (MSE) of the approximation model was 0.00017, the absolute error was not greater than 0.04, which shows higher approximation precision; the proposed algorithm was also used to build a dynamic flow soft measurement model and to finish related experimental study, the average absolute error of the dynamic flow measurement results was less than 0.15L/min, the relative error is 1.97%, these results meet measurement requirements well and are better than the results of the existing algorithms. The above simulation results and experimental results show that the proposed algorithm is of high modeling precision and good adaptability for complex non-linear system.

There are some deficiencies in traditional two-step algorithm for under-determined blind source separation, such as the value of K is difficult to be determined, the algorithm is sensitive to the initial value, noises and singular points are difficult to be excluded, the algorithm is lacking theory basis, etcetera. In order to solve these problems, a new two-step algorithm based on the potential function algorithm and compressive sensing theory was proposed. Firstly, the mixing matrix was estimated by improved potential function algorithm based on multi-peak value particle swarm optimization algorithm, after the sensing matrix was constructed by the estimated mixing matrix, the sensing compressive algorithm based on orthogonal matching pursuit was introduced in the process of under-determined blind source separation to realize the signal reconstruction. The simulation results show that the highest estimation precision of the mixing matrix can reach 99.13%, and all the signal reconstruction interference ratios can be higher than 10dB, which meets the reconstruction accuracy requirements well and confirms the effectiveness of the proposed algorithm. This algorithm is of good universality and high accuracy for under-determined blind source separation of one-dimensional mixing signals.

To meet the application demand of high speed scanning and massive data transmission in industrial Computed Tomography (CT) of low-energy X-ray, a system of high-speed data acquisition and transmission for low-energy X-ray industrial CT was designed. X-CARD 0.2-256G of DT company was selected as the detector. In order to accommodate the needs of high-speed analog to digital conversion, high-speed time division multiplexing circuit and ping-pong operation for the data cache were combined; a gigabit Ethernet design was conducted with Field Programmable Gate Array (FPGA) selected as the master chip,so as to meet the requirements of high-speed transmission of multi-channel data. The experimental result shows that the speed of data acquisition system reaches 1MHz, the transmission speed reaches 926Mb/s and the dynamic range is greater than 5000. The system can effectively shorten the scanning time of low energy X-ray detection, which can meet the requirements of data transmission of more channels.

Aiming at the problem of sample labeling in network traffic feature selection, and the deficiency of traditional semi-supervised methods which can not select a strong correlation feature set, a Semi-supervised Feature Selection based on Extension of Label (SFSEL) algorithm was proposed. The model started from a small number of labeled samples, and the labels of unlabeled samples were extended by K-means algorithm, then MDrSVM (Multi-class Doubly regularized Support Vector Machine) algorithm was combined to achieve feature selection of multi-class network data. Comparison experiments with other semi-supervised algorithms including Spectral, PCFRSC and SEFR on Moore network data set were given, where SFSEL got higher classification accuracy and recall with fewer selection features. The experimental results show that the proposed algorithm has a better classification performance with selecting a strong correlation feature set of network traffic.

In the recent years, the computer image understanding has wide and profound applications in intelligence traffic, satellite remote sensing, machine vision, image analysis of medical treatment, Internet image search and etc. As its extension, the image holistic scene understanding is more complex and integrated than basic image scene understanding task. In this paper, the basic framework for image understanding, the researching implication and value, typical models for image holistic scene understanding were summarized. The four typical holistic scene understanding models were introduced, and the model frameworks were thoroughly compared. At last, some research insufficiency and future direction in image holistic scene understanding were presented, which pointed out some new insights for the further research in this area.