Abstract
Coal gangue recognition presents significant challenges in the mining industry due to its inefficient and costly traditional treatment methods. The advent of deep learning techniques has introduced novel solutions for automating and online coal gangue processing. Despite the potential of deep learning models, challenges such as overfitting and the need for extensive labeled datasets hinder their effectiveness. You Only Look Once version 5 (YOLOv5), with its rapid inference speed and high accuracy, offers a suitable solution for real-time coal gangue detection. This research investigates the application of YOLOv5 for coal gangue image recognition, involving data preprocessing, model training, and optimization. Experimental results demonstrate that incorporating the multiple channel attention mechanism and lightweight content-aware re-assembly of features up-sampling operator significantly improves model confidence and recognition performance.