Gong-kuang zidonghua (Oct 2022)
Intelligent identification and positioning of steel belt anchor hole in coal mine roadway support
Abstract
When the steel belt auxiliary bolt is used in the coal mine underground heading roadway, if the positioning of the steel belt anchor hole is not accurate, the drill bit is easy to cause equipment damage when hitting the steel belt or anchor net. There are large potential safety hazards. In order to solve the above problems, an intelligent identification and positioning method of steel belt anchor hole in coal mine roadway support based on improved YOLOv5s model is proposed. ① The definition of the anchor hole image is increased by the super-resolution(SR). The high-frequency information of the anchor hole edge in the image is prevented from being lost due to image blurring. ② Because the anchor hole is small and the camera has a certain distance from the anchor hole, it is easy to lose the characteristic information of the small anchor hole in the convolutional neural network. This affects the detection effect of the anchor hole. The coordinate attention mechanism (CA) module is added to the Backbone network of YOLOv5s model. The network layers of the characteristic extraction network in the YOLOv5s network are increased. The coordinate information of the target object is integrated into the convolutional network. The characteristic information of the anchor hole small target can be effectively extracted, and the success rate of anchor hole detection is improved. ③ The YOLOv5s network embedded in the CA module is trained to the anchor hole dataset reconstructed by SR, and the improved YOLOv5s model, namely SR-CA-YOLOv5s model, is obtained. ④ The SR-CA-YOLOv5s model combined with the binocular camera is used to identify and locate the anchor hole in real-time. The experimental results show that compared with the YOLOv5s model, the mean average precision of the SR-CA-YOLOv5s model is 96.8%, which is 3.1% higher than the YOLOv5s model. The SR-CA-YOLOv5s model has better detection capability and avoids missing detection to a certain extent. Although the frames per second (FPS) of the SR-CA-YOLOv5s model is reduced by 18.5 frames/s, its FPS remains at 166.7 frames/s, which does not affect the real-time detection function of the model. The actual test results show that the SR-CA-YOLOv5s model can accurately detect the anchor hole and obtain the three-dimensional coordinate of the anchor hole relative to the camera under different lighting conditions. The coordinate error is within 6 mm, and the FPS meets the real-time requirements.
Keywords
