Recognition Methods for Coal and Coal Gangue Based on Deep Learning

Abstract

Recognizing coal and coal gangue is an important part of the coal industry and is mainly conducted via human sorting at present. Consequently, considerable manpower is needed, which adds a burden to enterprises and results in low efficiency. As an important branch of artificial intelligence, deep learning has been widely applied in many fields, especially in machine vision and voice recognition, its performance is greatly improved compared with the performances of traditional learning methods, and it also has good a transfer learning ability. This paper proposed an improved YOLOv4 algorithm as a classic deep learning method for the intelligent and highly accurate recognition of coal and coal gangue. Compared to other algorithms, YOLOv4 has a better anchor value by applying cluster analysis to different data sets, a good anti-interference ability due to using the Laplacian operator and Gaussian filter to reduce the impacts from mine dust and shock and acquires richer detailed information by increasing the number of layers of the feature pyramid. The experimental results show that compared with the other four algorithms of YOLOv4, YOLOv3, SSD and Faster-RCNN, the improved YOLOv4 proposed in this paper exhibits better detection accuracy, a better detection speed and robust performance.