The optimization of swarm intelligence algorithms is a main way to improve swarm intelligence algorithms. As the swarm intelligence algorithms are more and more widely used in all kinds of model optimization， production scheduling， path planning and other problems， the demand for performance of intelligent algorithms is also getting higher and higher. As an important means to optimize swarm intelligence algorithms， subgroup strategies can balance the global exploration ability and local exploitation ability flexibly， and has become one of the research hotspots of swarm intelligence algorithms. In order to promote the development and application of subgroup strategies， the dynamic subgroup strategy， the subgroup strategy based on master-slave paradigm， and the subgroup strategy based on network structure were investigated in detail. The structural characteristics， improvement methods and application scenarios of various subgroup strategies were expounded. Finally， the current problems and the future research trends and development directions of the subgroup strategies were summarized.

Aiming at the problems such as weak development ability， low population diversity， and premature convergence to local optimum in Dragonfly Algorithm （DA）， an HDASDE （Hybrid Dragonfly Algorithm based on Subpopulation and Differential Evolution） was proposed. Firstly， the basic dragonfly algorithm was improved： the chaotic factor and purposeful Levy flight were integrated to improve the optimization ability of the dragonfly algorithm， and a chaotic transition mechanism was proposed to enhance the exploration ability of the basic dragonfly algorithm. Secondly， opposition-based learning was introduced on the basis of DE （Differential Evolution） algorithm to strengthen the development ability of DE algorithm. Thirdly， a dynamic double subpopulation strategy was designed to divide the entire population into two dynamically changing subpopulations according to the ability that the subpopulation can improve the algorithm’s ability to jump out of the local optimum. Fourthly， the dynamic subgroup structure was used to fuse the improved dragonfly algorithm and the improved DE algorithm. The fused algorithm had good global exploration ability and strong local development ability. Finally， HDASDE was applied to 13 typical complex function optimization problems and three-bar truss design optimization problem， and was compared with the original DA， DE and other meta-heuristic optimization algorithms. Experimental results show that， HDASDE outperforms DA， DE and ABC （Artificial Bee Colony） algorithms in all 13 test functions， outperforms Particle Swarm Optimization （PSO） algorithm in 12 test functions， and outperforms Grey Wolf Optimizer （GWO） algorithm in 10 test functions. And it performs well in the design optimization problem of three-bar truss.

Aiming at the problems of slow detection and low recognition accuracy of road traffic signs in Chinese intelligent driving assistance system， an improved road traffic sign detection algorithm based on YOLOv3 （You Only Look Once version 3） was proposed. Firstly， MobileNetv2 was introduced into YOLOv3 as the basic feature extraction network to construct an object detection network module MN-YOLOv3 （MobileNetv2-YOLOv3）. And two Down-up links were added to the backbone network of MN-YOLOv3 for feature fusion， thereby reducing the model parameters， and improving the running speed of the detection module as well as information fusion performance of the multi-scale feature maps. Then， according to the shape characteristics of traffic sign objects， K-Means++ algorithm was used to generate the initial cluster center of the anchor， and the DIOU （Distance Intersection Over Union） loss function was introduced to combine DIOU and Non-Maximum Suppression （NMS） for the bounding box regression. Finally， the Region Of Interest （ROI） and the context information were unified by ROI Align and merged to enhance the object feature expression. Experimental results show that the proposed algorithm has better performance， and the mean Average Precision （mAP） of the algorithm on the dataset CSUST （ChangSha University of Science and Technology） Chinese Traffic Sign Detection Benchmark （CCTSDB） can reach 96.20%. Compared with Faster R-CNN （Region Convolutional Neural Network）， YOLOv3 and Cascaded R-CNN detection algorithms， the proposed algorithm has better real-time performance， higher detection accuracy， and is more robustness to various environmental changes.

为了克服传统密度估计方法受限于算法配置工作量高、高等级密度样本数量有限等因素无法大规模应用的缺点,提出一种基于监控视频的全景密度估计方法。首先,通过自动构建场景的权重图消除成像过程中射影畸变造成的影响,该过程针对不同的场景自动鲁棒地学习出对应的权值图,从而有效降低算法配置工作量;其次,利用仿真模拟方法通过低密度等级样本构建大量高密度等级样本;最后,提取训练样本的面积、周长等特征用于训练支持向量回归机(SVR)来预测每个场景的密度等级。在测试过程中,还通过二维图像与全景地理信息系统(GIS)地图的映射,实时展示全景密度分布情况。在北京北站广场地区的深度应用结果表明,所提全景密度估计方法可以准确、快速、有效地估计复杂场景中人群密度动态变化。

In order to overcome the weakness of direction and the poorness of purpose in multidimensional search and the premature convergence, this paper presented an improved particle swarm optimization algorithm. For both the best and the worst information of the cognitive part and the best and the worst information of the social part, the improved algorithm respectively assigned different learning factors, and the algorithm has a greater ability to learn. Each particle associatively memorized the best information and the worst information in its history, and then found the optimal position in accordance with the principle of chasing the best and avoiding the worst. Associative memory overcomes the weakness of direction and the poorness of purpose in multidimensional search. The principle of chasing the best and avoiding the worst keeps the diversity of population, helps to improve the convergence speed, and overcomes the premature convergence. Simulation test of the benchmark function has verified the validity of the algorithm.

To improve the resource utilization, the task resource requirement features of processor, network, disk and so on were efficiently estimated through analyzing the way of the task request, measuring the performance of application program in task or simulating to run the task. Afterwards, according to the features of resource requirement, the tasks could be classified into processor bound task, communicate bound task, disk bound task and others. And then the classified tasks were integrated with specific virtual machines to make all kinds of resources to be used efficiently. The research shows that the method can classify the task efficiently. And compared to unclassified method, it can reduce the times of virtual machines migration or integration.

The estimation of the content popularity in the Content Distributed/Delivery Network (CDN) system mainly relies on the experience of administrators, which implies strong subjectivity and cannot guarantee the Quality of Service (QoS). In the paper, the authors firstly preprocessed the data, and obtained the initial knowledge base to predict the film popularity. This paper used data mining techniques to learn the existing knowledge and predict the popularity of films. Thus, the films in the CDN system could be deployed more effectively and efficiently. The movie popularity predicted by Bayesian network classier was compared with the movie popularity predicted by decision tree. On the premise of the same correct classification rate and other classification parameters, the time taken to build model in the Bayesian network classifier can be shorter. Therefore, the Bayesian network classifier was preferred. The method can solve the inaccurate deployment caused by the administrators subjectivities and improve the efficiency of the CDN system.

Hyperspectral image has high spectral dimension, vast data and altitudinal interband redundancy, which brings problems to image classification. To effectively reduce dimensionality and improve classification precision, a new extraction method of nonlinear manifold learning feature based on Supervised Local Linear Embedding (SLLE) for classification of hyperspectral image was proposed in this paper. A data point's k Nearest Neighbours (NN) were found by using new distance function which was proposed according to prior class-label information. Because the intra-class distance is smaller than inter-class distance, classification is easy for SLLE algorithm. The experimental results on hyperspectral datasets and UCI data set demonstrate the effectiveness of the presented method.

The storage, look-up and view frustum culling of nodes in 3D scene are the key problems which effect the rendering efficiency in large scale scene. The paper introduces an improved quatree algorithm to store and look up nodes and proposes an iterative algorithm in place of recursion algorithm. And we implement the radial detection and view frustum culling based on this algorithm. The experimental results show that FPS is increased a lot in this way. The algorithm features in hiberarchy in itself and sequence in coding, which avoids large computation in view frustum culling algorithm.

In conventional region of interest (ROI) progressive image transmission algorithms, human visual specialties are not taken into account. An improved algorithm based on the properties of wavelet transform and human visual specialties was proposed. Firstly, an image with low resolution was transmitted so that user could decide if the image was needed. On the condition that there is no impact on the subjective quality of ROI, the transmission of ROI coefficients which are not important for visual quality could be delayed. And the bandwidth could be utilized to transmit the important back ground (BG) coefficients. Furthermore, depending on the bit rates, an expansion factor was evaluated. To guarantee the image with better visual effects, the transmission of BG image was controlled by the expansion factor. The experimental results show that the improved algorithm is effective and the expansion factor can accommodate image transmission.