Research On Improved Image Dehazing Algorithm Based On Multi-scale Convolutional Neural Networks

In recent years,the frequency of occurrence of bad weather such as fog and sputum has increased,and the suspended particles in the fog directly affect people access to clear images from the outdoors.The image quality of the degraded image obtained in the foggy day is blurred,which seriously affects the recognition and extraction of image feature information.Clearing and defogging of atomized images has important research value in the fields of traffic monitoring,military and civil aerial photography,target tracking,and remote sensing satellites.Aiming at many existing image dehazing methods,it is generally time-consuming,inefficient and easy to be distorted.Based on the mechanism of deep learning technology,this paper conducts in-depth research and discussion on the method of conventional convolutional neural network model.An improved image dehazing algorithm based on multi-scale convolutional neural networks.Its research direction is as follows:Firstly,this paper briefly introduces the basic foggy imaging model and some traditional defogging methods based on physical models and non-physical models.The study found that traditional methods based on physical models applied to image dehazing have generally achieved better results.Based on the deep learning fundamental framework,this paper trains the dataset based on the convolutional neural network model to realize the automatic extraction of image features.Based on this,a multi-scale convolutional neural networks based on the fusion of deep information and shallow information is explored.The improved neural network algorithm is used to quickly and efficiently achieve a single image clarify and defogging process.Secondly,based on the convolutional neural network training dataset,this paper designs a convolutional neural network model with two different scale architectures,which aims to realize the automatic extraction of the fine transmittance of deep and shallow feature information.The experiment first designs a set of coarse-scale networks for estimating the transmittance of the deep information of the image,and then sets a set of fine-scale networks with the functions of refining the local shallow structure of the image to obtain the fine transmittance.Finally,based on the foggy image degradation model combined The estimated atmospheric light value restores a clear,fog-free image.Finally,the experiment uses the fog image synthesized by the Middlebury Stereo Database,and the outdoor real image to perform the dehazing test,then compares and analyzes the results of several defogging algorithms.In this paper,the peak signal-to-noise ratio(PSNR),structural similarity(SSIM)and running time are used to verify the feasibility and effectiveness of the algorithm.The experimental results show that the performance index of the algorithm is the highest,the feature of image is clear after restoration,and the retained structural information is closer to the original clear fog-free image.The running efficiency is higher than several de-fogging algorithms,and the overall performance is good defogging effect.