Target detection in underground mines based on low-light image enhancement

The working environment in underground mines is complex, with extreme conditions such as dim lighting, high dust levels, and high humidity, leading to difficulties in feature extraction, low detection and positioning accuracy when existing target detection algorithms are applied underground. Therefore, a target detection algorithm based on low-light image enhancement is proposed. Firstly, the LIENet image enhancement algorithm is introduced to enhance the quality and brightness of low-light images, designing a dual gamma enhancement curve to adjust image pixels, achieving adaptive brightness adjustment through a few iterations under the guidance of a non-reference loss function. Secondly, a hierarchical feature extraction method HFE is proposed, with a dual-branch structure designed to capture long-term correlations and local correlations of input images separately, increasing attention to the corner regions crucial for the detection task. Finally, the HFE method is combined with a feature pyramid structure, able to effectively obtain comprehensive feature representation information through a top-down global feature adjustment method. The proposed method is validated on a self-built dataset, and compared with other excellent detection algorithms, the average detection accuracy of the proposed algorithm reaches the highest. The mAP@0.5 reaches 96.96%, and mAP@0.5:0.95 reaches 71.1%, demonstrating the excellent detection performance of the proposed algorithm under low-light conditions in mines.