In order to solve the problems that traditional video summarization methods often do not consider temporal information and the extracted video features are too complex and prone to overfitting, a video summarization generation model based on improved Bi-directional Long Short-Term Memory (BiLSTM) network was proposed. Firstly, the deep features of the video frames were extracted by Convolutional Neural Network (CNN), and in order to make the generated video summarization more diverse, the BiLSTM was adopted to convert the deep feature recognition task into the sequence feature annotation task of the video frames, so that the model was able to obtain more context information. Secondly, considering that the generated video summarization should be representative, the fusion of max pooling was adopted to reduce the feature dimension and highlight the key information to weaken the redundant information, so that the model was able to learn the representative features, and the reduction of the feature dimension also reduced the parameters required in the fully connected layer to avoid the overfitting problem. Finally, the importance scores of the video frames were predicted and converted into the shot scores, which was used to select the key shots to generate video summarization. Experimental results show that the improved video summarization model improves the accuracy of video summarization generation on two standard datasets TvSum and SumMe, its F1-score values are improved by 1.4 and 0.3 percentage points respectively compared with the existing Long Short-Term Memory (LSTM) network based video summarization model DPPLSTM (Determinantal Point Process Long Short-Term Memory).

Aiming at the problems of high power consumption and complex scene for scene perception in universal transportation modes, a new transportation mode detection algorithm combining Residual Network (ResNet) and dilated convolution was proposed. Firstly, the 1D sensor data was converted into the 2D spectral image by using Fast Fourier Transform (FFT). Then, the Principal Component Analysis (PCA) algorithm was used to realize the downsampling of the spectral image. Finally, the ResNet was used to mine the local features of transportation modes, and the global features of transportation modes were mined with dilated convolution, so as to detect eight transportation modes. Experimental evaluation results show that, compared with 8 algorithms including decision tree, random forest and AlexNet, the transportation mode recognition algorithm combining ResNet and dilated convolution has the highest accuracy in eight traffic patterns including static, walking and running, and the proposed algorithm has good identification accuracy and robustness.

In the multi-step traffic flow prediction task, the spatial-temporal feature extraction effect is not good and the prediction accuracy of future traffic flow is low. In order to solve these problems, a fusion model combining Long-Short Term Memory (LSTM) network, convolutional residual network and attention mechanism was proposed. Firstly, an encoder-decoder-based architecture was used to mine the temporal domain features of different scales by adding LSTM network into the encoder-decoder. Secondly, a convolutional residual network based on the Squeeze-and-Excitation (SE) block of attention mechanism was constructed and embedded into the LSTM network structure to mine the spatial domain features of traffic flow data. Finally, the implicit state information obtained from the encoder was input into the decoder to realize the prediction of high-precision multi-step traffic flow. The real traffic data was used for the experimental testing and analysis. The results show that, compared with the original graph convolution-based model, the proposed model achieves the decrease of 1.622 and 0.08 on the Root Mean Square Error (RMSE) for Beijing and New York traffic flow public datasets, respectively. The proposed model can predict the traffic flow efficiently and accurately.

Mobile Edge Computing (MEC) provides an effective solution to the conflict between computationally intensive applications and resource constrained mobile devices. However, most studies on the MEC offloading only consider the resource allocation between mobile devices and MEC servers, and ignore the huge computing resources in the cloud computing centers. In order to make full use of cloud and MEC resources, a task offloading strategy of cloud-edge collaboration was proposed. Firstly, the task offloading problem of the cloud-edge servers was transformed into a game problem. Then, the existence and uniqueness of Nash Equilibrium (NE) in this game were proved, and the solution to this game problem was obtained. Finally, a two-stage task offloading algorithm based on game theory was proposed to solve the task offloading scheme, and the performance of this algorithm was evaluated by performance indicators. The simulation results show that the total overhead of using the proposed algorithm is reduced by 72.8%, 47.9%, and 2.65% compared with those of local execution, cloud server execution and MEC server execution, respectively. The numerical results confirm that the proposed strategy can achieve higher energy efficiency and lower task offloading overhead, and extend scale well with the number of mobile devices increases.

Multi-label feature selection has been widely used in many domains, such as image classification and disease diagnosis. However, there usually exist missing labels in the label space of data in practice, which destroys the structure and correlation between labels, so that the learning algorithms are difficult to exactly select important features. To address this problem, a Multi-label Feature Selection based on Label-specific feature with Missing Labels (MFSLML) algorithm was proposed. Firstly, the label-specific feature for each class label was obtained via sparse learning method. At the same time, the mapping relations between labels and label-specific features were constructed based on linear regression model, and were used to recover the missing labels. Finally, experiments were performed on 7 datasets with using 4 evaluation metrics. Experimental results show that compared to some state-of-the-art multi-label feature selection algorithms, such as multi-label feature selection algorithm based Max-Dependency and Min-Redundancy (MDMR) and the Multi-label Feature selection with Missing Labels via considering feature interaction (MFML), MFSLML can increase the average precision by 4.61-5.5 percentage points. It can be seen that MFSLML achieves better classification performance.

In order to solve the problems of insufficient consideration of the traffic flow characteristics and the low accuracy of the prediction, a short-term prediction method of urban road traffic flow based on spatio-temporal node selection and deep learning was proposed. Firstly, the characteristics of traffic flow were analyzed in theory and data representation to obtain its spatial characteristics, and temporal characteristics and candidate spatio-temporal nodes set. Secondly, the set of candidate spatio-temporal nodes was determined according to the reachable range of traffic flow, and the fitness was calculated by taking the inverse of the sum of squares of errors as the objective function. In the historical training set, genetic algorithm and Back Propagation Neural Network (BPNN) were used to select spatio-temporal nodes, and the final spatio-temporal nodes and BPNN structure were obtained. Finally, the measured values of the selected spatio-temporal nodes were taken as the input of BPNN in the working set to obtain the predicted values. The experimental results show that compared with only using data of adjacent spatio-temporal nodes, using other time node ranges, Support Vector Machine (SVM) and Gradient Boosting Decision Tree (GBDT), the proposed model has a slight reduction in Mean Absolute Error （MAE） and Mean Absolute Percentage Error (MAPE), which are 10.631 6 and 14.275 8%, respectively; and 0.257 3和0.999 1 percentage points lower than those by using adjacent spatio-temporal nodes.

In order to improve the total utility of task allocation in spatiotemporal crowdsourcing dynamic reality, a Dynamic Threshold algorithm based on online Random Forest (DTRF) was proposed. Firstly, the online random forest was initialized based on the historical matching data of workers and tasks on the crowdsourcing platform. Then, the online random forest was used to predict the expected task return rate of each worker as the threshold, and the candidate matching set was selected for each worker according to the threshold. Finally, the matching with the highest sum of current utility was selected from the candidate match set, and the online random forest was updated based on the allocation result. The experiments show that the algorithm can improve the average income of workers while increasing the total utility. Compared with the greedy algorithm, the proposed algorithm has the task assignment rate increased by 4.1%, the total utility increased by 18.2%, and the average worker income increased by 11.2%. Compared with the random threshold algorithm, this algorithm has a better improvement in task allocation rate, total utility, average income of workers with better stability.

With the popularity of mobile smart devices, Mobile Crowd Sensing (MCS) has been widely used while facing serious privacy leaks. Focusing on the issue that the existing original data privacy protection scheme is unable to resist collusion attacks and reduce the perception data availability, a Data Privacy Protection algorithm based on Mobile Node (DPPMN) was proposed. Firstly, the node manager in DPPMN was used to establish an online node list and send it to the source node. An anonymous path for data transmission was built by the source node through the list. Then, the data was encrypted by using paillier encryption scheme, and the ciphertext was uploaded to the application server along the path. Finally, the required perception data was obtained by the server using ciphertext decryption. The data was encrypted and decrypted during transmission, making sure that the attacker was not able to wiretap the content of the perception data and trace the source of the data along the path. The DPPMN ensures that the application server can access the original data without the privacy invasion of the nodes. Theoretical analysis and experimental results show that DPPMN has higher data security with increasing appropriate communication, and can resist collusion attacks without affecting the availability of data.

Focusing on the problem that the existing car high-beam requires the driver to manually change the headlamp through his own judgment of the road condition, which may results in a traffic accident due to the illegal use of the high-beam, a matrix LED high-beam intelligent assistant control system which can automatically adjust the radiation way of high-beam according to the road condition and environment was designed and implemented. Firstly, according to the driving characteristics of vehicles and related traffic regulations, the intelligent control strategy of matrix LED high-beam assistant system was proposed for different road conditions. Then the hardware and software of the system were designed and implemented. In the hardware part, the device selection and circuit design of the modules like main controller, LED power driver and matrix switch controller were given, and the software part was composed of function modules like driving circuit control, matrix switch control and intelligent control strategy. Finally, a complete experiment system under laboratory conditions was built for functional test. The experiment test results indicate that the proposed method has accurate results and is steady, reliable, better in real-time and easy to realize, which achieves the expected goal.

To meet the requirements for adaptivity, and robustness of the algorithm to optimize urban traffic signal control, a traffic signal control algorithm based on Deep Reinforcement Learning (DRL) was proposed to control the whole regional traffic with a control Agent contructed by a deep learning network. Firstly, the Agent predicted the best possible traffic control strategy for the current state by observing continously the state of the traffic environment with an abstract representation of a location matrix and a speed matrix, because the matrix representation method can effectively abstract vital information and reduce redundant information about the traffic environment. Then, based on the impact of the strategy selected on the traffic environment, a reinforcement learning algorithm was employed to correct the intrinsic parameters of the Agent constantly in order to maximize the global speed in a period of time. Finally, after several iterations, the Agent learned how to effectively control the traffic.The experiments in the traffic simulation software Vissim show that compared with other algorithms based on DRL, the proposed algorithm is superior in average global speed, average queue length and stability; the average global speed increases 9% and the average queue length decreases 13.4% compared to the baseline. The experimental results verify that the proposed algorithm can adapt to complex and dynamically changing traffic environment.

Focusing on the poor overall allocation effect due to the total utility of task allocation or task waiting time being considered respectively in the study of task allocation under spatial crowdsourcing environment, a dynamic threshold algorithm based on allocation time factor was proposed. Firstly, the allocation time factor of task was calculated based on the estimated waiting time and the already waiting time. Secondly, the task allocation order was obtained by comprehensively considering the return value of task and the allocation time factor. Thirdly, the dynamic adjustment item was added based on the initial value to set the threshold for each task. Finally, candidate matching set was set for each task according to the threshold condition, and the candidate matching pair with the largest matching coefficient was selected from the candidate matching set to join the result set, and the task allocation was completed. When the task allocation rate was 95.8%, compared with greedy algorithm, the proposed algorithm increased total allocation utility by 20.4%; compared with random threshold algorithm, it increased total allocation utility by 17.8% and decreased task average waiting time by 13.2%; compared with Two phase based Global Online Allocation-Greedy (TGOA-Greedy) algorithm, it increased total allocation utility by 13.9%. The experimental results show that proposed algorithm can shorten the average waiting time of task while improving the total utility of task allocation, to achieve the balance between the total allocation utility and the task waiting time.

Real-time urban traffic monitoring has become an important part of modern urban management, and traffic big data collected by video monitoring is wildly applied to urban management and traffic control. However, such huge citywide monitoring traffic big data is rarely used for urban traffic and urban computing research. The vehicle type mining and application analysis were implemented on the citywide monitoring traffic big data of a provincial capital city. Firstly, three types of vehicles with important influence on urban traffic:periodic private car, taxi and public commuter bus were defined. And the corresponding mining method for each type of vehicles was proposed. Experiments on 120 million vehicle records collected from 1704 video monitoring points in Jinan demonstrated the effectiveness of the proposed definitions and mining methods. Secondly, with four communities as examples, the residents' traffic modes and the relationships between the modes and the distribution of surrounding Points of Interest (POI) were mined and analyzed. Moreover, the potential applications of the urban traffic big data incorporated with POI in urban planning, demand forecasting and preference recommendation were explored.

In order to solve the problem that the existing repair methods for missing bus arrival information have little factors considered, low accuracy and poor robustness, a method to repair missing bus arrival data based on DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm and multi-source data was proposed. Bus GPS (Global Positioning System) data, IC (Integrated Circuit) card data and other source data were used to repair the missing arrival information. For the name, longitude and latitude data of the missing arrival station, the association analysis of complete arrival data and static line information were carried out to repair. For the missing arrival time data, the following steps were taken to repair. Firstly, for every missing data station and its nearest non-missing data station, the travel time and schedule in the historical complete arrival data between the two stations were clustered based on DBSCAN algorithm. Secondly, whether the two adjacent runs of the studied bus with complete data belonged to the same cluster was judged, and if they belonged to the same cluster, th cluster would not change, otherwise the two clusters would be merged. Finally, the maximum travel time corresponding to the cluster midpoint was used as the missing travel time to determine whether there was a passenger swiping his card to board the bus at this station or not, if so, the arrival time was calculated from the time of swiping cards, and if not, the mean of the maximum and minimum travel time corresponding to the cluster midpoint was used as the missing travel time to calculate the arrival time. Taking Xia'men bus arrival data as examples, in the repair of name, longitude and latitude of the missing arrival station, the clustering method based on GPS data, the maximum probability estimation method and the proposed method can repair the data by 100.00%. In the repair of missing arrival time, the mean relative error of the proposed method is 0.0301% and 0.0004% lower than that of two comparison methods respectively, and the correlation coefficient of the proposed method is 0.005 and 0.0075 higher than that of two comparison methods respectively. The simulation results show that the proposed method can effectively improve the accuracy of repair of missing bus arrival data, and reduce the impact of the number of missing stations on accuracy.

Aiming at the problem that the traditional Laplacian score for feature selection cannot be directly applied to multi-label tasks, a multi-label feature selection algorithm based on Laplacian score was proposed. Firstly, the sample similarity matrix was reconstructed by the correlation of the common and non-correlated correlations of the samples in the overall label space. Then, the correlation and redundancy between features were introduced into Laplacian score, and a forward greedy search strategy was designed to evaluate the co-operation ability between candidate features and selected features, which was used to evaluate the importance of candidate features. Finally, extensive experiments were conducted on six multi-label data sets with five different evaluation criteria. The experimental results show that compared with Multi-label Dimensionality reduction via Dependence Maximization (MDDM), Feature selection for Multi-Label Naive Bayes classification (MLNB) and feature selection for multi-label classification using multivariate mutual information (PMU), the proposed algorithm not only has the best classification performance, but also has a remarkable performance of up to 65%.

To solve the problem of properly mapping the computational task onto a three-dimensional Network-on-Chip (NoC), an improved algorithm based on Genetic Algorithm (GA) was proposed. GA has the fast random searching ability and Prim algorithm can get the minimal spanning tree of a weighted connected graph. By combining the two algorithms' advantages, the improved algorithm could properly assign computational tasks onto each network node, achieving a high efficiency on solving network power consumption and heat problems. The simulation experiments were carried out to compare the proposed improved GA based on Prim algorithm with GA based 3D NoC mapping algorithm. The simulation results indicate that the average power consumption of the improved GA based on Prim algorithm is lower:from the overall trend, the reduction on power consumption is positive correlated to the increase of the number of processing units, and when there are 101 processing units, the average power consumption is 32% lower than that of the traditional GA.

Concerning the problem that rule-based inference engine in smart environment needs to centralize data to sink node resulting in excessive data transmission in sensor network, a Minimum Transmission Cost of Rete Distribution Scheme algorithm (MCoRDS) based on Rete network cost model was proposed. Through the dependence statistics of sub-rule patterns in Rete Network on fact data, it was found that many sub-rule patterns could be reasoned nearby the source data collected sensor. Data transmission to sink node could be cut down and the data transmission of whole sensor network was decreased by distributing sub-rule patterns of Rete network into the sensor which firstly collected all the source data for it. Compared to centralized inference which places the Rete network in the sink node, 4 experiments were conducted. In the 4th experiment, total sensor network hops was reduced from 85000 to 8036, about 90.5% reduction, the other experiments had some reduction too. The experimental results show that MCoRDS has lower data transmission, especially in the case of large-scale rules and low frequency rule trigger.

Focusing on the issues of failing to pinpoint the edge information accurately and lacking texture detail of image by using integer order differential or 0-1-order fractional differential mask operators in digital image processing, a new 1-2-order edge detection operator based on Laplacian operator was proposed. Deduced from the definition of Riemann-Liouville (R-L),the 1-2-order fractional differential had the advantage in enhancing high-frequency signal and reinforcing medium frequency signal. The simulation results demonstrate that the proposed operator can take an higher recognition rate on the subjective recognition, and it's better at extracting the edge information, especially for the image with rich texture detail in the smooth region with little change of gray scale. Objectively, the integrated location error rate is 7.41% which is less than that of integer order differential operators (a minimum of 10.36%) and 0-1-order differential operator (a minimum of 9.97%). Quantitative indicators show the new fractional operator can effectively improve the positioning accuracy of the edge, and the proposed operator is particularly suitable for edge detection with high frequency information.

To resolve the problems on description and proof of semantic property preservation in Model-Driven Software Development (MDSD), a formal approach was proposed for software architecture model on the basis of type category theory and process algebra. The semantic constraints of component specifications which should be kept through model transformation, were deeply analyzed and discussed. From the view of diagram structure, port and configuration constraints, external behavior and component substitutability, the problem of property preservation was described, and the corresponding criteria was built at the same time. The framework provides a guidance for the definition of model transformation rules, and provides the basis to verify the correctness of model transformation as well as to analyze the effect of model transformation. The application research shows that, the approach enhances semantic description capabilities of component model, and can be used as an effective supplement for existing software modeling method.

For improving recovery performance of signals sampled by sub-Nyquist sampling system with Compressed Sensing (CS), the block Simultaneous Iterative Hard Thresholding (SIHT) recovery algorithm for joint sparse model based on ε-closure was proposed. Firstly, The CS synthesis model for Multiple Measurement Vector (MMV) of sampling system was analyzed and the concepts of ε-coherence and Restricted Isometry Property (RIP) were proposed. Then, according to the block coherence of redundant dictionaries, the SIHT algorithm was improved by optimizing the support sets in iterations. In addition, the iterative convergence constant was given and the algorithm convergence property was analyzed. At last, the simulation experiments show that, compared with traditional method, the new algorithm can achieve recovery success rate of 100% with enough sampling channels, while the noise suppressing ability was increased by 7 dB to 9 dB and the total execution time was brought down by at least 37.9%, with higher convergence speed.

The characteristic that mobile opportunistic network exploits node contacts to forward packets is very suitable for the Ad Hoc networking requirements in actual environment, thus a large number of applications are produced. Considering the smart devices are generally carried by people or integrated in vehicles, human involvement is one of the most important factors for the success of these applications, this paper explored the influence of human mobility on data communication. The authors observed that people always visited hot regions, while other regions were visited less frequently. Motivated by this observation, the GS (Gathering Spray), a human gathering point assisted spraying scheme for mobile opportunistic scenarios was proposed. GS assumed each hot region configured a Access Point (AP), which had a higher priority to cache and spray messages than other mobile nodes. Theoretical analysis verifies that GS achieves a lower mean delivery delay than the Spray-Wait, and the simulation results show that GS improves the packet delivery ratio simultaneously.

As many methods of brain tumor Magnetic Resonance Imaging (MRI) segmentation are usually driven by such conditions as noise, intensity inhomogeneity within tumor, fuzzy and discontinuous boundaries, it is difficult to segment tumor accurately. To improve the segmentation results, morphological multiscale modification of controlled marker was proposed. Firstly, this method was based on morphological gradient images because the adaptive structure elements were utilized on different pixels in different areas. In addition, modifying gradient image was key to avoid a larger misregistration of target boundaries. Finally, marker-controlled watershed was applied to segment brain tumor. The experimental results show that the method of brain tumors has more accurate segmentation results. Key words：brain tumor; morphological multi-scale modification; watershed transform

Concerning the problem of the background interference during the salient object detection, a key salient object detection algorithm was proposed based on filtering integration in this paper. The proposed algorithm integrated the locally guided filtering with the improved DoG (Difference of Gaussia) filtering, and made the salient object more highlighted. Then, the key points set was determined by using the saliency map, and the result of saliency detection was got by adjustment factor, which was more suitable for human visual system. The experimental results show that the proposed algorithm is superior to existing significant detection methods. And it can restrain the background interference effectively, and have higher precision and better recall rate compared with other methods, such as Local Contrast (LC), Spectral Residual (SR), Histogram-based Contrast (HC), Region Contrast (RC) and Frequency-Tuned (FT).

Firstly, the paper introduced the researches and basic methods of spatial data visualization technology, and analyzed two common kinds of methods, namely entity-based and region-based. A clustering-based spatial data visualization method was proposed, which firstly made a cluster analysis of spatial data and got the description parameters of the result through the use of spatial clustering algorithms represented by algorithm ASCDT (Adaptive Spatial Clustering algorithm based on Delaunay Triangulation). Secondly, it designed visual objects aimed at the cluster result by combining the basic visualization methods and the characteristics of the parameters. As a result, the mapping relationship was established. Finally, some issues that needed to be further studied and improved were discussed.

Accurate shadow boundary detecting and texture-preserving are two critical difficulties in shadow removal. To solve these problems, a new shadow removal method based on gradient field was proposed. Firstly, shadow boundary was detected approximately. Then, the gradients in internal shadow region and shadow boundary were modified respectively to obtain the non-shadowed gradient field. Based on the gradient field, the information in shadow regions was recovered with Poisson equation. The experimental results with several images indicate that the method can remove shadow from images easily while preserving the textures in the shadow regions, and it is not sensitive to the accuracy of shadow boundary.

To solve the redundancy and high-dimensional problem of the electronic system condition monitoring data, a monitoring data optimization algorithm that combined the sample optimization and features optimization was put forward. Firstly, monitoring data samples were optimized by feature space sample selection algorithm, and the most representative samples were found; then monitoring data characteristics were optimized by KPCA-EDA algorithm after the sample optimization. More recognition information was retained on guarantee that the feature information was enough. Finally, a filter circuit was taken as an example to simulate, and the result shows that this method is effective.

Concerning the contradiction between convergence speed and convergence precision when the traditional fixed pace Least Mean Square (LMS) algorithm was used to radar clutter adaptive filter system, the paper put forward a new kind of variable-pace adaptive filter algorithm. Through combining the relevant error and the former pace to real-time update next iteration of the pace in its basic pace iterative formula, which could reach with higher convergence speed and smaller disorder, and it also could prevent the bad effect from the existing related noise. The simulation results show that, compared with the traditional fixed-pace LMS algorithm and context improved algorithm, the convergence rate, convergence accuracy and noise prevention have been greatly improved. It proves that the proposed algorithm is effective, feasible, and consistent with the theoretical analysis.

First,DC map of I frame was extracted for rough detection of the shots.And then the forward motion compensation vector of P frame was extracted, and Extended Vector Median (EVM) filtering was used for the preprocessing of motion vector.Finally the characteristics of three sports features,exercise intensity value, exercise intensity difference and the direction of motion vector histogram, absolute difference were calculated. Fuzzy inference was introduced to synthesize these three characteristics and classified shots to abrupt-change,gradual-change and no-change ones.The MPEG video shot boundary detection method do not need to decompress the video fully,and extracts information directly from MPEG compressed bit stream,so it is of low computation complexity and high extration speed,which is verified by the experimental results.

False corners will emerge in the corner detection of the image with high texture complexity by Harris algorithm, and when fractional differential is applied to image processing, the order needs to be specified by human. This paper analyzed the reason that caused the false corners and suggested to replace the integral order in the algorithm with fractional order to operate differential coefficient so as to improve the algorithm. The paper also brought forward an approach regarding fractal dimension as a parameter which came from the order in choosing differential coefficient. So the marginal information of image can be saved when operating the differential coefficient of the image, and this approach makes the fractional differential be applied in occasions with high real-time requirements such as video target tracing and video image stabilization. The tests show that the modified algorithm has higher precision in the corner detection.

For further investigate the essential element of information expressed of large-scale irrigation area and constructs the overall model of irrigation area system,the system engineering method was used to analyses the information source and information flow of irrigated area.By this method,the connection and state between each element transmit the relation in the system was described systematically and a model of Information transmission and controlling of the large-scale irrigated area was constructed.The application of this model to the irrigation are informationization shows that this model has offered the feasible quantization model for realization and accurate control of the information system of irrigated area.

After investigating two kinds of already existed Information Unit model,a new information unit model named Categorized Information Unit was proposed as searching object on which to conduct the second time Web search.Experimental results show that the model can enlarge and refine the original searching results,promots the original searching pricision.It suggests a new way for Web search.