Research On Dehazing Algorithm Of Coal Mine Images In Complex Scenarios

With the development of modern industrial technology,the integration of coal industry and artificial intelligence technology has become an industry trend and urgent need.However,the water droplets and coal dust in the coal mine environment will disperse the reflected light of the object,resulting in a significant decrease in the image quality acquired by the optical imaging system,which will have a serious impact on the work of workers safety helmet detection,worker tracking,belt foreign object recognition and so on in the intelligent coal mines.Therefore,it is of great significance to study how to restore foggy image in coal mines and reduce the impact of fog and dust particles on image quality.However,some existing dehazing algorithms only apply to outdoor foggy images,but cannot be applied to coal mine scenes,which is due to the complexity of coal mine scenes compared with outdoor scenes.For example,in the open-pit coal mine highillumination scene,there will be uneven fog distribution,sky area and single color in the image;in the underground coal mine high-illumination scene,there are often many artificial light sources,such as miners’ headlamps,roadway lighting and safety lights,which will affect the image and make it difficult to see the details in the image;The lowillumination scenes of open-pit coal mines and underground coal mines are often caused by environmental light,resulting in uneven brightness distribution,low contrast,color distortion and blurred images in the images.This paper has studied in detail the causes of image degradation under high and low illuminance in open-pit coal mines and underground,and has delved into the theoretical knowledge and methods of image dehazing.On this basis,some meaningful practical works have been carried out on existing image dehazing methods,such as improvement,perfection and introduction of new ideas,including but not limited to.Firstly,aiming at the problem that the sky area recovery is not true in the high illumination image dehazing processing of open-pit coal mine and the dehazing at the light source in the high illumination image of underground mine is not complete,this paper proposes a dehazing algorithm based on boundary constraint.The basic idea of this method is: firstly,the boundary constraint of the image is used to obtain the coarse transmittance value of the image;then,the dark channel prior algorithm is used to correct the transmittance value obtained by the boundary constraint,and a more detailed and accurate rough transmittance map is obtained.Finally,the context regularization is used to refine the transmittance,and the image after defogging is obtained by combining the atmospheric scattering model,thus improving the defogging effect at the sky area of the open-pit coal mine and the underground light source.Secondly,aiming at the problem of wrong segmentation of bright and dark regions in the process of dehazing low illumination images in underground mines,this paper proposes a bright and dark region segmentation algorithm.First of all,after many experiments,it is found that the Otsu method can effectively segment the bright and dark regions in the low illumination environment,but it will also segment the light reflected by some object surfaces.Therefore,this paper modifies the Otsu method so that it can more accurately segment the bright and dark areas of the underground.Thirdly,aiming at the problem of overexposure in the process of image enhancement,this paper proposes a transmittance modification function.It uses the image enhancement characteristics to increase the transmittance value of the bright area,and combines the modified transmittance of the bright area with the transmittance of the dark area,and then weakens the splicing gap through Gaussian filtering to ensure image integrity.Lastly,in order to verify the effectiveness of the two algorithms proposed in this paper,this paper not only evaluates the image from a subjective perspective,but also uses the four image quality evaluation methods of peak signal-to-noise ratio,structural similarity,feature similarity and information entropy to evaluate the experimental results from an objective perspective.Figure[32] Table[10] Reference[88]...