Research on Coal and Gangue Identification Algorithm by Improved YOLOv7

In view of the low accuracy, missed and false detection of the current coal gangue recognition algorithm, this paper proposes a coal and gangue recognition algorithm by improved YOLOv7. The FReLU activation function and the lightweight normalization based attention mechanism (NAM) are introduced into the YOLOv7 algorithm; at the same time, K-means++ algorithm is utilized for the re-clustering of anchor boxes; EIOU is used as the bounding box regression loss function instead of CIOU. The experimental results demonstrate that the improved YOLOv7 algorithm achieves mean average precision (mAP) of 98.96%, the F1 value of 95.98%, and the single image recognition time of 29.77ms for each individual image. The improved YOLOv7 algorithm demonstrates precise identification of coal and gangue in diverse lighting conditions, reduce the probability of missed and false detection, and meet the real-time requirements.