Research And Implementation On Infrared And Visible Image Fusion Technology Based On Super Resolution

Visible images have the characteristics of high resolution,clear texture details and rich colors,but they are also easily affected by illumination conditions,and clear visible images cannot be obtained in low visibility environments.Infrared images are not affected by illumination conditions,and clear images can also be obtained in low illumination environment.However,infrared images represent the radiation distribution of objects,which is different from human perception and is not conducive to human observation.Moreover,infrared image is limited by its imaging principle,so it is impossible to obtain the same high-resolution image as visible image,such as infrared image with 2k resolution.To sum up,in order to obtain high-resolution fused images,combined with the different characteristics of visible images and infrared images,a certain fusion strategy is adopted to generate fused images,which can be widely used in target recognition,tracking and other fields.In this paper,the purpose of research is to improve the quality of fused images.Deep learning super-resolution reconstruction algorithm is used to improve the resolution of infrared image and on this basis,the image quality and performance index of fused image are improved.The main contents and innovations of this paper are as follows:(1)To solve the problem of low resolution of infrared image,this paper puts forward an improvement of infrared image super-resolution reconstruction algorithm,that is,an infrared image super-resolution reconstruction algorithm framework based on residual network structure,and on this basis,combined with attention mechanism,enhances the local details of infrared image.Experimental results show that the improved infrared super-resolution reconstruction algorithm effectively enhances the highfrequency details of infrared images,and improves the objective index PSNR by 2%compared with EDSR algorithm.(2)Based on infrared image super-resolution restoration algorithm,a fusion network model of infrared and visible images is proposed.According to infrared image superresolution restoration algorithm,the method of extracting features of visible images and infrared images is improved,an effective fusion strategy is adopted for image fusion.The performance index of fusion image is compared with that of classical image fusion algorithm to verify the effectiveness of fusion algorithm.Experimental results show that the fusion algorithm effectively improves the objective index of the fused image,and the visual quality is clearer and the contrast is more obvious.(3)In order to improve the practicability of the algorithm,the hardware system of fusion algorithm based on super-resolution reconstruction is built in this paper.Design parallel optical axis to collect visible images and infrared images,register the collected images and verify the effectiveness of the super-segmentation fusion algorithm proposed in this paper;At the same time,the implementation of the algorithm on ZYNQ heterogeneous hardware platform is studied.Combined with the performance characteristics of ZYNQ chip,the resource utilization of the algorithm is optimized,and the software and hardware collaborative algorithm is designed to accelerate the whole neural network.Experimental results show that the hyper-sub-fusion algorithm deployed on the hardware platform can achieve good acceleration effect.