Infrared And Visible Image Fusion Algorithm Based On Convolutional Neural Network

Infrared and visible image fusion is a process in which different types of images collected by different sensors are fused together through certain algorithms in the same scene,so as to realize the complementary advantages of infrared and visible images and solve the problem of insufficient information of a single image,thus providing higher quality and more accurate information.It lays the foundation for target recognition,target tracking and other fields,and is widely used in machine vision,object detection,military and other fields.For the problems of infrared and visible image fusion,this paper introduces convolution neural network on the basis of multi-scale decomposition method,which can obtain more detail features.This paper takes advantage of its advantages to complete the following tasks:(1)To solve the problem that the details and textures of fused images are not clear enough,this paper proposes a non-subsampled Shearlet Transform(NSST)and Convolution Nerual Network,CNN)infrared and visible image fusion algorithm.The method first uses NSST to decompose the source image.Secondly,similarity matching and fusion rules were adopted for low-frequency images,and high-frequency images were input into IFCNN network for feature extraction.Then,L2 regularization,convolution operation and maximum strategy were used to obtain the final high-frequency images.The fusion result is obtained by inverse transformation.The final experimental results show that the proposed method retains the edge details of the image well and has good visual effect.(2)To solve the problems of fuzzy edges and low contrast of fused images,this paper proposes a new method based on Window Empirical Mode Decomposition.WEMD and Generative Adversarial Networks(GAN)infrared and visible image fusion algorithm.Firstly,WEMD is used to decompose the image into Intrinsic Mode Function(IMF)and residual components.The residual components are fused by simple weighted average because they contain less detail information.Principal Component Analysis(PCA)was used for the fusion of IMF Components,and then the preliminary fusion result graph was reconstructed.The preliminary fusion result graph was input into the improved generative adversarial network.In order to better extract useful information,The generator and discriminator are added into the dense connection to complete the information of infrared and visible light,and finally the fusion result is obtained.Experimental results show that the proposed method highlights the target of infrared image,improves the contrast,and has obvious advantages in objective evaluation indexes.In this paper,the infrared and visible image fusion system is designed,and the other eight image fusion algorithms are evaluated subjectively and objectively.By comparing the algorithm,the effect of the proposed algorithm can be seen intuitively.In this system,the fusion image and evaluation indexes can be viewed more conveniently,providing technical support for the infrared and visible image fusion.