Research On Target Detection And Recognition In Image Fusion System

Aiming at the research of tank target detection in military complex environment,the target is easy to be occluded,the feature is submerged in the background,and the traditional single sensor image acquisition has poor target detection and recognition ability in complex background.In this paper,the target detection algorithm in the heterogeneous image fusion system is proposed,and the algorithm is systematically analyzed and simulated.Xilinx Zynq-7020 development platform is used to transplant and optimize the proposed algorithm,and an embedded target detection system for heterogeneous image fusion is built.The main research contents include the implementation of heterogeneous image data acquisition,infrared and visible image fusion algorithm and target detection algorithm on the embedded platform.As follows:The imaging principle of infrared and visible light sensors is studied.According to the different advantages of the two images,the fusion method of wavelet transform is adopted to fuse the infrared and visible light images.The corresponding high-frequency components and low-frequency components in the image are fused through the fusion rules.The thermal radiation information in the infrared image and the detailed texture information in the visible light image are combined to enhance the target contour and provide effective target features for target detection.A sliding window-based DPM+SVM target detection method is proposed,and a component model is built on the basis of HOG to enrich target features.It also proposes a target detection algorithm based on improved SSD+DSST,improves the traditional SSD,combines the feature pyramid FPN algorithm in the basic convolution network architecture,enhances the semantic information of the deep network,optimizes the target receptive field,makes full use of the convolution to calculate the target feature information to enhance the target detection ability,and adds the channel-space attention mechanism,Through the DSST method of scale discrimination,the problem of target loss is solved and the stable detection of continuous multi-frame targets is realized.The Zynq-7000 series development board is used to build an Ethernet wireless transmission system for binocular cameras.Through customized software on the PC side,the video wireless transmission between the development board and the upper computer is realized.Customize the algorithm IP core module at the FPGA side to realize hardware acceleration design,carry out hardware engineering block design,establish the hardware driver design at the ARM side and the module timing control and build the SVM classifier model,and realize the embedded target detection system of infrared and visible light fusion image.The PC terminal and ZYNQ platform jointly built a small target simulation system capable of target fusion,target trajectory setting,target recognition and tracking,and realized efficient simulation and algorithm performance test of preset scenes through the superposition and fusion of targets in different complex backgrounds.