Concerning the problem of low object detection precision of DEtection TRansformer （DETR） for small targets， an object detection algorithm with optimized feature extraction structure， called CF?DETR （DETR combined CSP?Darknet53 and Feature pyramid network）， was proposed on the basis of DETR. Firstly， CSP?Darknet53 combined with the optimized Cross Stage Partial （CSP） network was used to extract the features of the original image， and feature maps of 4 scales were output. Secondly， the Feature Pyramid Network （FPN） was used to splice and fuse the 4 scale feature maps after down?sampling and up?sampling， and output a 52×52 size feature map. Finally， the obtained feature map and the location coding information were combined and input into the Transformer to obtain the feature sequence. Through the Forward Feedback Networks （FFNs） as the prediction head， the category and location information of the prediction object was output. On COCO2017 dataset， compared with DETR， CF?DETR has the number of model hyperparameters reduced by 2×10⁶， the average detection precision of small objects improved by 2.1 percentage points， and the average detection precision of medium? and large?sized objects improved by 2.3 percentage points. Experimental results show that the optimized feature extraction structure can effectively improve the DETR detection precision while reducing the number of model hyperparameters.

The method of point cloud reconstruction based on slicing technology needs to extract boundary points from slicing planar points. In order to solve the problem of extracting boundary points and overcome the drawback of low efficiency and bad result of current algorithms, a boundary points extraction method of planar point cloud based on multi-threshold was proposed. In the algorithm, k adjacent points were selected from the judged points, then the angle between the nearest points were calculated and the maximum angle was limited because there existed the biggest angle, thus the boundary points could be rapidly extracted. By analyzing the value of multi-threshold, testing the method to extract boundary points of different point cloud and comparing the proposed method with other three methods, the method accurately and better extracted boundary points regardless of the shapes. The experimental results show that the proposed method can well extract the boundary points on the condition of guaranteeing the original characteristic information and improves the speed and efficiency of point cloud reconstruction.

Concerning the optimization of the overall complexity problem on the high-performance computing platform, a new algorithm named Group Mosquito Host-Seeking Algorithm (GMHSA) was proposed. GMHSA was an intelligent optimization algorithm inspired by mosquitoes sucking blood behavior. It involved max-min fairness and group interaction behavior. The producer group was chosen according to the concept of leader decision and the leadership functions were constructed to make each group maintain their own superiority as well as getting rid of local optimal solution. The algorithm was tested by Traveling Salesman Problem (TSP) and compared with other swarm intelligent algorithms. In the parallel experiment of 16 nodes, the speedup of GMHSA was 15.8, which was nearly linear speedup. Moreover, it could be directly used to solve transport problems and other practical optimal problems. The results indicate that GMHSA has highly parallelism and scalability, and it is an effective measurement for solving complex optimal problems involving behavior.

This paper proposes a classified strain limiting method which deals with unreal stretch deformation in clothing simulation with physical-based mass-spring model. The method primarily include two processing module. The first module is classification. it uses velocity computed by integrating system as input parameters first, and then judges some point whether needs strain limiting by the energy method, finally through the judged result divides point set into two types : needing strain limiting and not. The second module is strain limiting. It defines the threshold value of spring deformation and three variables for representing the restrictive proportions in principal strain direction, and then by the line strain theory computes the strain tensor of spring, finally obtains specific restrictive proportions and updates the position of corresponding point. The method could guarantee natural simulation results and eliminate unreal stretch deformation, and does not require all elements of the strain limiting for processing, reducing the computational cost to ensure real-time. Results indicate that the method have good effect and efficiency.

In this paper, a new joint Maximum A Posterior (MAP) formulation was proposed to integrate image registration into blind image Super-Resolution (SR) reconstruction to reduce image registration errors. The formulation was built upon the MAP framework, which judiciously combined image registration, blur identification and SR. A cyclic coordinate descent optimization procedure was developed to solve the MAP formulation, in which the registration parameters, blurring function and High Resolution (HR) image were estimated in an alternative manner given to the two others, respectively. The experimental results indicate that the proposed algorithm has considerable effectiveness in terms of both quantitative measurement and visual evaluation.