Research On Pollution Source Image Classification And Quality Improvement Method Based On Deep Learning

The improvement of the quality of rain,haze and noise images(hereinafter referred to as pollution source images)is a very important technical basic work in the application of modern computer and vision technology.In recent years,the rise of deep learning technology has made pollution source images quality improvement methods more accurate.However,existing methods still exist problems such as poor generalization ability,poor decontamination effect of natural rain and haze images,and distortion and blurring of processed images.In response to the above problems,this article uses deep learning methods to propose corresponding improvement measures.The specific work is as follows:(1)This article created an image data set covering multiple pollution sources,including images with rain,haze,noise,and multiple pollution.After labeling,a total of 24,000 images were created.(2)Using the above image data set,this paper designs a pollution source image classification algorithm.The image is preprocessed first,and then the characteristic information of the input image data is captured by the residual network,and finally the convolutional neural network is used to realize the multi-label classification of different pollution source types in the image data set.(3)This paper proposes a multi-stage pollution source images quality improvement method to solve the problem of decontamination of different pollution source images in a targeted manner.First,the pollution source images classification algorithm is used to obtain the prior information of the image,and then in the dehazing stage,this paper designs a multi-scale convolutional neural network to solve the image dehazing algorithm,and obtains the transmittance and other parameters to evolve to obtain the dehazing image.The image is enhanced after processing.In the rain/de-noise stage,the residual network is used to extract the features of the rain/noise image,and the rain/denoise image is obtained.Finally,in the image detail restoration stage,this paper uses a generative adversarial network to supplement the lost image details after removing rain,haze and noise to improve image quality.The experimental results show that compared with the existing pollution source images quality improvement algorithms,the algorithm in this paper has stronger generalization ability,higher universality,and better processing performance.