Research On Coal-rock Interface Identification Method For Shearer Based On Image Information Mining

The exploration and research of coal-rock interface identification methods has always been a hot and difficult point of research at home and abroad.At present,more than 20 kinds of coalrock interface identification technologies have been proposed.Although these methods have achieved some achievements,due to the complexity and special nature of the coal mine underground environment Many methods have problems such as low recognition accuracy and narrow application range.with the development of machine vision technology,the relates research of coal-rock interface recognition method based on image recognition has made progress,which provides a new idea for realizing the accurate recognition of coal-rock interface.Therefore,for the coal-rock interface images collected from the coal mining face,this paper starts with the relevant theories of image recognition,and on this basis,the coal-rock image recognition research is carried out.The specific work is reflected in the following aspects:(1)Make a dataset of coal and rock images.At present,the research on image-based coal and rock identification is in the preliminary stage,and there is a lack of public coal and rock image data sets.In this paper,an explosion-proof camera is used to obtain coal and rock images from the fully mechanized mining face of Dongsheng Coal Mine,Wangtao Township,Qinyuan County,Changzhi City,Shanxi Province.sample.Due to the poor underground conditions,the number of samples obtained is small.In this paper,the image data set is augmented by various methods,and the augmented images are marked to construct a coal rock image data set.(2)Design a feature extraction method and a classifier selection method suitable for coal and rock images to realize image classification.In this paper,the general process of image classification is described,the image preprocessing operation is performed on the coal and rock image data set,the texture features of the coal and rock images are extracted by the method of gray co-occurrence matrix,and KNN,RF and BP neural networks are selected as the classifiers,respectively.The extracted feature vector is input into the classifier for training to realize coalrock image classification.(3)A network model of coal and rock image classification based on deep learning is designed.Comparing experiments on a variety of different deep learning networks,such as VGG16,Res Net50,Mobile Net V2 and Efficient Net V2,combined with transfer learning to train the network,and verified it on the coal rock data set in this paper.The results show that the Efficient Net V2 model has better performance than other models.(4)A semantic segmentation method of coal-rock interface image based on Trans U-Net is designed.By introducing the combined structure of CNN-Transformer,the encoder part in the original U-NET structure is replaced.Using the pytorch deep learning framework,on the coal and rock image segmentation dataset,a comparative experiment with FCN,U-Net and U-Net++network models was carried out.The experimental results verify the effectiveness and superiority of Trans U-Net in coal and rock image classification.