Research On Fusion Image Generation Based On Deep Neural Network

Image fusion is an information fusion technology that brings practical application value to many fields.For image fusion tasks,reducing the loss of information in the fusion process can achieve better fusion results.However,the existing methods have some information missing.For example,the human filter bank cannot extract all the frequency characteristics of the image when detecting the feature activation degree,and the image cannot be completely transformed after the scale transformation,and commonly used.The fusion rules are too absolute and so on when making decisions.This paper focuses on image fusion technology based on Convolutional Neural Network(CNN),and designs a new fusion framework that is compatible with many mainstream CNNs models.Multi-sensor image fusion is achieved by using feature maps obtained by convolutional neural networks.The method proposed in this paper can effectively optimize three aspects of image fusion:1)Complete activation detection,regional detection is one of the focuses in the field of image fusion.Based on the hierarchical feature extraction neural network has powerful feature extraction ability,it can effectively avoid the shortage of traditional filters;2)Accurate image reconstruction,transforming the image by convolution,fusion,etc.,and then using the inverse reconstruction iteration to generate the fused image.Convolution transformations maximize the complete representation of the image.3)Differentiable fusion rules,this paper proposes a continuous and differentiable fusion rule,the purpose of which is to preserve the details of the source image as much as possible,and embed end-to-end fusion network training.Finally,the experiment proves that the method proposed in this paper has obvious improvement in the above three aspects.This method can obtain the superior level of the current mainstream methods both in subjective and objective evaluation indicators.