Infrared And Visible Image Fusion Based On Deep Learning

With the development of information fusion technology,the cooperative work of sensors is more and more widely used.The fusion of infrared and visible images is also a part of multisensor technology.Image fusion technology is to fuse simultaneous interpreting two or more than two images from different sensors to increase the integrity of image information,so that the viewer can get more visual information.Image fusion has been widely used in medical image,military scene,remote sensing image and machine vision.In this paper,deep learning is applied to the fusion of infrared and visible images,and three fusion methods of infrared and visible images are proposed.Firstly,the process of infrared and visible image fusion algorithm based on deep learning is studied.The image denoising,enhancement,registration and other aspects of the work,the deep learning model training,and then use the algorithm of this paper to fuse the infrared and visible images,and then use the objective evaluation parameters to evaluate the fused image objectively.Secondly,the infrared and visible light fusion method based on dense convolution neural network and visual saliency is studied.The infrared and visible light source images are decomposed into low frequency and high frequency parts respectively.The salient image of infrared image is extracted by visual saliency detection,and the infrared image features are enhanced.The low-frequency fusion image is obtained by fusing the saliency map of infrared image with the low-frequency part of infrared and visible image.The high-frequency part of the image is easy to lose information in the process of fusion.Therefore,the dense convolution neural network is used to process the features of the high-frequency part of the image,and then the maximum fusion strategy is used to fuse the high-frequency part.Finally,the fusion image of the low-frequency and high-frequency parts is reconstructed to obtain the final fused image.Thirdly,image fusion based on scale edge protection and dense convolution neural network is studied.The scale sensing edge protection filter is used to iterate infrared and visible images.Four images are obtained after four iterations,and the four images contain different image features.Then the fourth image is processed by Gaussian filter to get the image with more features,which is called the base layer image,and then the basic layer image is fused.Then,the basic layer image and the four layer filter image are fused by the trained dense convolution neural network,and the information difference of the fused image is calculated.These information differences are the edge information of the image.Finally,the detailed information calculated and the fusion image of the basic layer are fused to get the final fused image.Finally,the infrared and visible image fusion method based on DPN network is studied.The image is fused based on dual path network,and the image is decomposed by Nonsubsampled shear wave.The low-frequency part of the decomposed image is fused by the improved Laplacian energy of the guide filter and the fusion strategy of the larger one.The high-frequency part is fused by the fusion rules of dual path network.The experimental results show that compared with the traditional fusion methods of infrared and visible images,the fusion images obtained by the three fusion algorithms proposed in this paper get good results,which shows that the deep learning algorithm has superior performance in infrared and visible image fusion.