Research And Implementation Of Single Image Dehazing Based On Deep Learning

Under the haze weather,the fine particles in the air and their scattering in the atmosphere make the images obtained by various imaging devices become blurred,which reduces the application value of the images.Therefore,it is particularly important to reduce the influence of haze on imaging and improve the image quality.The key to realize image defogging is to accurately estimate the transmission rate through image analysis.In order to understand the distribution of transmission rate through a large number of foggy image analysis,the most effective method at present is to use deep learning network for modeling analysis.This paper designed the three kinds of convolution structure of neural network for single image to fog.(1)The resnet-based network structure is used to design a trainable end-to-end system for estimating media transmission.The system bypasses the input information directly to the output,to sovle the problem of information loss and protect the integrity of information.The whole network only needs to learn the different parts of input and output,so as to simplify the learning objectives and difficulties.The system adopts the full convolutional network structure of Res Net,and takes two 3x3 convolutional networks concatenated together as a residual module to construct the network structure.(2)Two trainable end-to-end systems are designed based on the Dense Net network architecture.The first network combines Dense Net with the full convolution network.The network uses the idea of dense block and multi-scale,rich image feature information is extracted.The second network combines Dense Net with a de-noising self-encoder to extract and combine features through encoding and decoding.The above two network systems take the haze image as the input and output its medium transmission diagram,and then use the medium transmission diagram to restore the fog-free image through the atmospheric scattering model.Experiments show that this algorithm improves the quality and efficiency of fogging image.