Research On High Resolution Remote Sensing Image Dehazing Method Based On Deep Learning

Nowadays,remote sensing technology is developing rapidly.Remote sensing images are widely used in land resources exploration,industry and agriculture,disaster relief and other fields.However,the remote sensing imaging equipment is affected by fog,which makes the remote sensing image blurred and the utilization of remote sensing image reduced.Therefore,it has certain research significance for removing fog from remote sensing images.In order to improve the quality of image haze removal,a pre-condition is set before removing fog from remote sensing images.The transfer learning method based on VGG19 is used to recognize the fog images.After the image is identified,the image is defogged by using conditional generative adversarial nets method.This method can get rid of the atmospheric scattering model and can be used to remove fog directly from the image.The main work is as follows.(1)In this paper,the recognition task of fog image is realized by means of VGG19 and transfer learning.First,the remote sensing image is divided into foggy and clear images.By fine-tuning the VGG19 network,the convolution layer and pooling layer pairs in the network are retained,and the final classification part is changed into two categories.By training the network,the foggy image can be recognized finally.(2)The remote sensing image is de-fogged by the method of conditional generative adversarial nets.Before the model training,fog images and clear images are paired.The generator network adopts the network structure of upsampling and downsampling.In order to ensure the similarity between the generated image and the original clear image,a distance function is added to the loss function.The network really achieves end-to-end image dehazing.(3)Designing a high resolution remote sensing image defogging system.Using VGG19 transfer learning method and conditional generative adversarial nets defogging method,a high resolution remote sensing image defogging system is designed,which integrates image recognition module and image defogging module.Experimental results show that the algorithm is effective and feasible in practical application.