Image enhancement method in high-dust environment based on deep learning and atmospheric scattering model

Abstract

In view of the image degradation caused by a large number of suspended particles such as coal dust and water mist during underground mining, this paper proposes an image enhancement method based on depth learning and atmospheric scattering model, combined with the characteristics of inconsistent illumination components and inconsistent dust concentration of the collected image. Firstly, the input image is decomposed into reflection component and incident component by neural network; Secondly, referring to the characteristic values of intensity, standard deviation and area, the incident component is decomposed into image blocks by quadtree decomposition method, and the consistency of illumination components in image blocks is achieved; Then, based on the decomposition of the incident component and referring to the dust concentration value, the reflection component is further decomposed into image blocks, realizing the consistency between the illumination component and the dust concentration in the image block; Finally, based on the prior knowledge such as saturation and information entropy, the transmission estimation is completed, and the image enhancement under the high dust environment in the coal mine is realized by combining the principle of atmospheric scattering model. Experimental analysis shows that the image enhancement method proposed in this paper has achieved good results in adding visible edge ratio, contrast restoration, image clarity and so on, and provides a new idea for image enhancement in the high dust environment of coal mines with uneven illumination and uneven concentration of suspended particles.