Aiming at the problems of variable morphology， low resolution and limited information of small-target water-floating garbage， which lead to unsatisfactory detection results， an improved Faster-RCNN （Faster Regions with Convolutional Neural Network） water-floating garbage detection algorithm was proposed， namely MP-Faster-RCNN （Faster-RCNN with Multi-scale feature and Polarized self-attention）. Firstly， a small-target water-floating garbage dataset in Lanzhou part of the Yellow River was established， the combination of atrous convolution and ResNet-50 was used as the backbone feature extraction network instead of the original VGG-16 （Visual Geometry Group 16） to expand the perception field for extracting more small-target features. Secondly， two layers of convolutions of 3×3 and 1×1 were set in the Region Proposal Network （RPN） by using multi-scale features to compensate for the feature loss caused by a single sliding window. Finally， polarized self-attention was added before RPN to further utilize multi-scale and channel features to extract finer-grained multi-scale spatial information and inter-channel dependencies to generate a feature map with global features， achieving more accurate target box localization. Experimental results show that compared with the original Faster-RCNN， MP-Faster-RCNN can effectively improve the detection accuracy of water-floating garbage with a mean Average Precision （mAP） improvement of 6.37 percentage points， the model size is reduced from 521 MB to 108 MB， and the convergence speed is faster under the same training epoch.

Nowadays， deep learning models have been widely applied in dialogue act recognition， which can improve classification performance by mining various features of dialogue acts. However， the existing methods neglect the latent association and interaction between different features of dialogue acts and also seldom consider the semantic relevance between labels of dialogue act in the classification process， which hinders from improving the performance of dialogue act recognition. To solve these problems， an MFA-HC （Multi-feature Fusion Attention-based Hierarchical Classification） method for recognizing dialogue act was proposed. Firstly， a hierarchical dialogue act classification framework based on learning without forgetting was proposed， which combined various fine-grained features such as words， parts of speech and relevant linguistic statistics to learn and train the dialogue act classification model. Secondly， a universality-individuality model based on attention mechanism was proposed to capture the universality and individuality features among different features. Experimental results on two benchmark datasets SwDA （Switchboard Dialogue Act corpus） and MRDA （ICSI Meeting Recorder Dialogue Act corpus） show that， compared with DARER （Dual-tAsk temporal Relational rEcurrent Reasoning network）， which has the current overall superior performance in existing methods， MFA-HC method improves the classification accuracy by 0.6% and 0.1% by capturing the universality and individuality features hidden in the utterance.

The existing distributed triangle counting algorithms assume that all computing nodes are in the same location， while in reality， the nodes may be located in multiple data centers across continents. Geo-distributed data centers which are connected with wide area networks have characteristics of heterogeneous network bandwidth， high communication cost and uneven distribution， and the existing distributed algorithms cannot be applied to geo-distributed environment. At the same time， the existing research which ignores the temporal locality of the formation of triangles mostly adopts strategies such as random sampling and elimination of edges. Therefore， the triangle counting problem of real graph streams in geo-distributed environment was studied， and a Geo-distributed Triangle Counting （GTC） algorithm was proposed. Firstly， aiming at the problem of too high data transmission caused by the existing edge distribution strategy， a geo-distributed edge distribution strategy was proposed to build a benefit formula combining the time benefit and data benefit of communication and use point-to-point communication to replace broadcast edges. Then， for the triangle repeated counting problem caused by point-to-point communication in geo-distributed environment， a final edge calculation rule was proposed to ensure no counting repetition. Finally， based on the time weighted sampling algorithm， a time-weighted triangle counting algorithm was proposed to use the time locality of the triangle to sample. The GTC was compared with Conditional Counting and Sampling （CoCoS） and Tri-Fly on five real graph streams. The results show that GTC has the communication data size decreased by 17% compared to CoCoS and decreased by 44% compared to Tri-Fly， GTC has the error rate decreased by 53% compared to Tri-Fly and slightly less than CoCoS， and GTC has the running time decreased by 34% compared to Tri-Fly and slightly more than CoCoS. It can be seen that the GTC can reduce the size of communication data effectively while ensuring high accuracy and short algorithm running time.

Refined short-term traffic flow prediction is the premise to ensure the rational decision making in Intelligent Transportation System （ITS）. In order to establish the lane-changing model of self-driving car， predict vehicle trajectories， and guide vehicle routes， the timely traffic flow prediction for each lane has become an urgent problem to solve. However， refined short-term traffic flow prediction faces the following challenges： first， with the increasing diversity of traffic flow data， the traditional prediction methods cannot meet the requirements of ITS for high precision and short time delay； second， training prediction model for each lane make a huge waste of resources. To solve the above problems， a refined short-term traffic flow prediction model combined Convolutional-Gated Recurrent Unit （Conv-GRU） with Grey Relational Analysis （GRA） was proposed to predict lane flow. Considering the characteristics of long training time and relatively short reasoning time of deep learning， a cloud-fog deployment scheme was designed. Meanwhile， to avoid training prediction models for each lane， a model migration deployment scheme was proposed， which only needs to train the prediction model of some lanes， and then the trained prediction models were migrated to the associated lane for prediction through GRA. Experimental results of extensive comparisons on a real-world dataset show that， compared with traditional deep learning prediction methods， the proposed model has more accurate prediction performance； compared with Convolutional-Long Short-Term Memory （Conv-LSTM） network， the model has shorter running time. Furthermore， the model migration is realized by the proposed model under the condition of ensuring high-precision prediction， which saves about 49% of training time compared to training prediction model for each lane.

To solve the problem of Fine Particulate Matter (PM2.5) concentration prediction, a PM2.5 concentration prediction model was proposed. First, through introducing the comprehensive meteorological index, the factors of wind, humidity, temperature were comprehensively considered; then the feature vector was conducted by combining the actual concentration of SO2, NO2, CO and PM10; finally the Least Squares Support Vector Machine (LS-SVM) prediction model was built based on feature vector and PM2.5 concentration data. The experimental results using the data from the city A and city B environmental monitoring centers in 2013 show that, the forecast accuracy is improved after the introduction of a comprehensive weather index, error is reduced by nearly 30%. The proposed model can more accurately predict the PM2.5 concentration and it has a high generalization ability. Furthermore, the author analyzed the relationship between PM2.5 concentration and the rate of hospitalization, hospital outpatient service amount, and found a high correlation between them.

The cluster head election of current Wireless Sensor Network (WSN) mainly relies on energy and location, but ignores the trust. In this paper, using the trust as the basis of reliability and with the energy as the priority, trust value was considered to improve the system reliability. Meanwhile, an election generates more candidate nodes, every node bears cluster head in turn, so it reduces the frequency of the election and improves the efficiency of the election of cluster head. Thus, a Cluster Head Trust Elections (CHTE) based on trust and energy was proposed. The experiments show that the electing algorithm is efficient on data packet correctness of sink node and Mean Time Between Failures (MTBF) within WSN environments.

On the basis of analyzing the mixed Client/Server (C/S) and Browser/Server (B/S) architecture pattern and AJAX technology, a novel mixed architecture pattern was proposed, which can unify the foreground interaction method of B/S and C/S and make the servers share effectively, thus enhance the scalability and maintainability of the system. According to the proposed pattern, the architecture of the spacecraft dynamical application platform was given, as a reference to the system with similar architecture.