Key Technologies for Target Recognition of Conveyor Belt Inspection Robot in Low Illumination and Blurred Environments
Abstract：To address the issue of low recognition accuracy for inspection targets (especially small target defects) in underground coal mines, this paper proposes a deep learning-based intelligent detection method. The YOLOv5 model is used as the base network, and the Res2Net multi-scale feature representation module is introduced to replace the Bottleneck in the C3 module of the backbone network, thereby increasing the receptive field channels in the network. The CA attention mechanism is introduced in both the backbone network and the feature fusion network to improve channel utilization and early surface damage detection accuracy. The YOLOv5 model is trained using a dataset, and comparative experiments are conducted with the original YOLOv5 framework. Field tests show that the detection accuracy for conveyor belt misalignment distance ≥10 cm is ≥95%, for longitudinal tear length ≥10 cm it is 80%–95%, for foreign object detection with object size >80 mm it is 80%–95%, and for roller abnormality detection it is 96%. Compared to before optimization, the weighted detection accuracy increased from 83% to 95%.
Machine-generated Keywords：
Publication Date：2025-11-30
Online Publishing Date：2025-12-10(First online date of this platform, not the publication date of the document)
Pages：6( 65-70 )