A real-time deviation detection method for coal conveyor belt based on REDLNet

To tackle the challenges posed by inadequate detection performance in current algorithms for detecting conveyor belt deviations during coal transportation, this study presents a real-time algorithm designed for deviation detection in coal conveyor systems. Initially, an algorithm leveraging deep edge feature constraints is employed, successfully addressing issues of missed and false detections caused by indistinct edge features of the conveyor belt in low-light conditions. Furthermore, the study incorporates edge feature gradient constraints as a loss function, which n enhances the speed of convergence during the model training phase. Following this, the edge points identified on the coal conveyor belt are fitted using the least squares approach to derive accurate edge lines. Lastly, a deviation threshold is established in accordance with safety production standards for coal mining, and the fitting results are compared against this threshold to facilitate the detection of deviations within the conveyor belt. In this paper, the Real-time Belt Deviation Detection Method is based on Lightweight Neural Network and Feature Gradient Constraints (REDLNet) algorithm was compared with ultra-fast structure-aware Deep Lane Detection (UFSLD) and lightweight Convolutional neural network (L-CNN) to verify its superior performance. The results show that the algorithm proposed in this paper can efficiently and real-time detect the deviation of coal conveyor belts compared to existing algorithms, providing effective guarantees for the safety production of coal mines