Research And Application Of Single Image Dehazing Algorithm Based On Convolutional Neural Network

With the development of modern industry,the emissions of various exhaust gases,automobile exhaust and domestic gas in the environment have greatly increased.It is well known that foggy weather is prone to occur in autumn and winter in southwestern China.In foggy weather conditions,particles floating in the atmosphere,such as dust and smoke,absorb and scatter light in large quantities,resulting in a decline in the quality of various captured images.A reduction of image quality can affect the performance of many computer vision systemsThe main task of image defogging technology is to eliminate the influence of foggy weather and environmental pollution on image quality,so that the image after defogging can be directly applied to further image processing work.We urgently need to develop an algorithm and related applications that can defog in real time and improve image quality.The existing image dehazing algorithms have problems such as difficulty in feature extraction,low filtering efficiency,and poor defogging effect on outdoor images in large sky areas.In order to ensure the good dehazing effect of the image,this study first analyzes the mechanism of fog image degradation,and then combines the convolutional neural network with the atmospheric light scattering model to establish the image defogging model.Secondly,the model is optimized.The fog-free image was successfully restored by this model.Finally,the RESIDE large data set was selected for experiment.The subjective and objective evaluation methods of the image proved that our proposed method has better performance than these existing methods.The main research contents of this paper are as follows.(1)For the problem of difficult feature extraction of the traditional manual dehazing algorithm,we used the three-layer convolutional neural network structure specially designed by Dehazenet network to extract the haze characteristics and achieve better results.(2)Aiming at the problem of low filtering efficiency of the existing defogging algorithm,this paper used several fast guide filtering algorithms which was one of the popular algorithms of edge smoothing to post-process the transmission map.We decreased the time complexity of the filter calculation from(46)(N)to O(N/S~2)optimizes the defogging time.(3)In view of the problem that the existing algorithm has poor defogging effect on outdoor images with large area sky area,here we adopted an improved version of the dark channel prior algorithm,which uses the channel estimation method when estimating atmospheric light.The method solved the problem of defogging on outdoor images with large area of sky.The experimental results showed that the defogging effect obtained by the algorithm is better.(4)In order to implement the improved algorithm into specific applications,this paper designed a GUI interface for displaying the input foggy image and defogged image,and the objective evaluation index of the improved algorithm,including peak signal to noise ratio PSNR and average.Structural similarity SSIM.After a series of experiments,the proposed algorithm has certain advantages,and solves many problems in the current defogging calculation of single image,so that the image after defogging can be directly applied to further image processing.