The Algorithm Of Single Image Dehazing Based On Deep Learning

With the development of society and industrialization,the emissions of various industrial gases,natural disaster gas and human life gas in the global environment are increasing rapidly,and the global climate is affected.There are often hazy weather in many countries and regions.In this hazy weather,the particles and smoke floating in the atmosphere will absorb and scatter the sunlight,which leads to the serious degradation of the image quality collected by all outdoor image acquisition equipment,and has ineffaceable impact on the judgment of computer vision system.The algorithm research of image dehazing is to eliminate the influence of the hazy weather on the image quality of the image acquisition equipment through the algorithm,so it is urgent to research and develop an algorithm and related application system that can real-time move the haze from image and improve the image quality.However,there are some problems in the existing image dehazing algorithms,such as poor dehazing of high-resolution image,halo effect and the dehazing effect in large-area sky area.In order to get better dehazing effect,this paper first analyzes and analogically deduces the atmospheric scattering model,establishes the hazy image model with noise,and combines the refinenet and residual network to dehazing the high-resolution hazy image;secondly,the contour information is extracted through the creation of the guidance decomposition module,and then the condition generation network is combined with it.In this way,it can be used to solve the possible halo phenomenon in image generation.Finally,we select large datasets such as RESIDE and NYU for experiments,through subjective and objective evaluation methods,we prove that the proposed method has better effect than the existing methods.The main contents and innovations of this paper are as following:(1)As a result of the difficulty of existing technology in feature extraction of high-resolution haze image,and the effect is not good,so this paper constructs 18 layer residual network as the feature extractor to extract four different scales of hazy image.In order to deal with the problem that the Refine Net module is too large,some of its cell structures are modified to fuse the proposed features.Secondly,the whole network is recursively called to learn the non-linear mapping between the hazy image and the clear image.The experimental results show that the algorithm achieves the best results compared with some of the leading edge algorithms,and improves the peak signal-to-noise ratio andthe structure similarity by 0.605 db and 0.015 on average.(2)In order to solve the halo phenomenon that may appear in the generation of clear and haze-free images,as well as the restoration requirements of some visual tasks for the contour information of the scene.Firstly,the guided decomposition module is constructed by the decomposition module and the guided filter layer to extract the contour information of the hazy image;secondly,the generator in the network is antagonized by the common structure improvement condition of the residual network and the u-net,and the intermediate features of the generator are fused by the feature mapping channel to restore the clear image.Experimental results show that the algorithm has achieved good results objectively,and achieved the highest index value in some data sets.Compared with the conditional countermeasure generation network,it has improved the peak signal-to-noise ratio and structure similarity by 2.71 db and 0.023,and has achieved a more clear visual effect on contour restoration subjectively.(3)Finally,in order to consider the configuration of the image dehazing machine,based on the above two depth learning methods and some traditional dehazing algorithms,the prototype of the image dehazing system is designed,which can provide effective help for the corresponding image acquisition personnel to obtain clear and haze-free images.In this paper,two kinds of single image dehazing algorithms based on depth learning are proposed to improve the quality of image dehazing,and the prototype of image dehazing system is designed,which is conducive to the recovery of low-quality images in some related industries.