Application Of Machine Vision Technology In Coal Gangue Recognition

Coal is one of the most important fossil energy sources in China.In the process of coal mining,large amounts of coal gangues are often mixed with coals.The coal gangue has a low carbon content and a small combustion benefit,so the coal gangue must be sorted.At present,artificial drainage is still the most common sorting method,and the manual sorting method is time-consuming,labor-intensive and efficient.In view of these drawbacks,this paper adopts the machine vision method to sort coal gangue and coal automatically.The difference in visual aspect between coal gangue and coal also provides the possibility that machine vision can be applied to the identification of coal gangue.In order to complete the separation of coal gangue and coal,the experiment simulates the segmentation and identification of coal gangue and coal under the black belt background.In this paper,the acquired image is enhanced by gray equalization,Laplacian image enhancement and logarithmic transformation and then the enhanced image is filtered by Gaussian filtering.For the image after pre-processing,image segmentation methods such as threshold image segmentation,k-nearest neighbor image segmentation and watershed algorithm based on improved distance transform,are employed to separate coal gangue and coal from a black background.The experimental results indicate that the watershed image segmentation algorithm based on distance transform is robust and not susceptible to illumination and shadow.In addition,the segmentation result is more accurate.Then the classification models are established to classify coal gangue and coal.In this paper,traditional machine classifiers are employed in classification tasks and data after feature extraction is as input of the classifiers.Support vector machine and random forest are selected as the classification model,and HOG features and gray level co-occurrence matrix are extracted as input feature vectors.The experiment compares the effects of four combinations on coal gangue and coal sorting.Finally,the experimental results indicate that the coal gangue sorting algorithm based on HOG feature extraction and support vector machine achieves the best performance with the accuracy of 91.9%.The result demonstrate the effectiveness of the methods in the separation of coal gangue and coal.Finally,convolutional neural network is employed to classify coal gangue and coal.Firstly,a four-layer shallow convolutional neural network is constructed to observe the classification effect of coal gangue and coal.Experiments results show that the shallow convolution neural network can achieve an accuracy of 92.5%.The identification of coal gangue is basically completed,and machine vision has a potential to identify the coal gangue from coal.The paper has 30 pictures,5 tables,and 79 references.