FUSION OF VISIBLE IMAGE AND SAR IMAGE BASED ON CONVOLUTIONAL NETWORK

With the need for image development, a single image can no longer comprehensively represent a video scene, and the development of fused images is imminent. Image fusion refers to the process of extracting the most favorable information from each channel through the integrated processing of image processing and computer technology, and finally combining them into a high-quality image. At present, most of the image fusion techniques are mainly focused on visible and infrared images, and the research on synthetic aperture radar and visible light images is limited. In this paper, we focus on image fusion based on visible light and synthetic aperture radar images. First, this paper introduces the definition of image fusion and analyzes the image fusion potential of visible light image and synthetic aperture radar image. Second, the advantages and disadvantages of the two existing image fusion methods are summarized, and a convolutional neural network-based image fusion method is proposed to solve the existing problems. This method mainly processes the high-frequency part and low-frequency part of the image separately. The low-frequency part is mainly processed by linear combination, and the high-frequency part is fused by extracting the corresponding salient features based on VGG19 network. In order to compare with other image fusion algorithms, we conduct experimental tests using the OS dataset to obtain specific definitions of various evaluation metrics of fused images, and find that the proposed algorithm improves the evaluation metrics of fused images. Finally, we conclude this paper and present the remaining challenges in image fusion research and predict possible future research directions.