Research On Scratch Detection And Inpainting Method Of Coal Photomicrograph

The quantitative analysis of the microscopic components of coal has important research significance for the definition of coal process performance.The observation of coal polished section with an optical microscope is an important means for quantitative analysis of the microscopic components.With the development of computer vision technology,researchers continue to try to use image processing methods to analyze the microscopic components of coal.However,during the preparation and use of the coal polished section,it is easy to cause coal scratches on its surface,which affects the pass rate of coal polished section preparation and the results of image analysis.Therefore,it is necessary to detect coal scratches in coal photomicrograph and repair the image by technical means.At present,domestic and foreign researches on the scratch detection and inpainting of coal photomicrographs are still insufficient.Most of the existing researches are based on traditional image processing methods.The detection accuracy is not high and the false detection and missed detection are serious.The image inpainting result has obvious semantic information missing.Based on the semantic segmentation algorithm and generative adversarial technology,this paper studies the scratch detection and inpainting of coal photomicrographs.Around this topic,the main work of this article is as follows:Aiming at the problem of scratch detection in coal photomicrographs,this paper proposes a semantic segmentation detection method based on dual attention mechanism.First,this paper proposes a dual attention model that combines spatial attention and channel attention.This model can adaptively update the attention weight of feature information and improve the accuracy of coal scratch detection.Then,this paper combines the attention model with the semantic segmentation network,and obtains the coal scratch detection model through deep learning training.The experimental results show that the method proposed in this paper can accurately output the semantic segmentation results of coal scratches.The objective evaluation indicators of pixel accuracy and mean intersection-of-union respectively reached 92.64% and 87.56%,which are significantly better than traditional methods and basics networks.Aiming at the inpainting of coal photomicrographs,this paper studies the image inpainting strategy based on the generative adversarial model.First of all,this paper proposes a coal photomicrographs inpainting network using shallow feature prior information and lightweight gated convolution,which can generate patch images with contextual semantic information relevance.Then,this paper designs a weighted objective loss function for the image inpainting network,and improves the effect of coal photomicrograph generation through model training.The experimental results show that the coal photomicrograph complemented by the method proposed in this article has rich texture details,and the peak signal-to-noise ratio and structural similarity index of the image respectively reached 39.96 d B and 99.15%,which is an increase of 8.69 d B and 1.99% compared with the traditional image inpainting methods,with better image inpainting performance.This paper has 49 pictures,7 tables,and 82 references.