Belt Abnormal High-Temperature Detection Algorithm Based on Improved YOLOv7-Tiny

This paper endeavors to enhance the current coal conveyor belt anomalous high-temperature detection algorithms by addressing their limitations in terms of precision and efficiency. We introduce a novel algorithm grounded in the optimized YOLOv7-tiny model, which leverages the Normalized Wasserstein Distance (NWD) loss function to enhance the representation of small targets. Furthermore, we incorporate the DyHead detection head to enrich feature representation and elevate detection capabilities. Our experimental findings validate that our enhanced algorithm surpasses the original YOLOv7-tiny algorithm, exhibiting notable gains in both detection accuracy and speed, especially when detecting small and visually similar targets.