基于RCR _YOLOv4的矿井巷道红外障碍检测研究

针对地下矿井巷道光线昏暗的道路上会出现落石或行人等行车障碍物，严重影响无人驾驶矿卡安全行驶的问题，提出了一种基于红外视觉识别的巷道障碍物快速检测优化模型RCR_YOLOv4。该模型利用K-Means++优化算法筛选巷道障碍物的先验框尺寸，并引入深度可分离卷积降低网络参数量和计算量，从而提高障碍目标的定位精度和检测效率。通过设计双通道注意力机制对网络特征融合模块进行优化，实现对无人矿卡行车障碍的高精度检测。结果表明，该目标检测模型对矿井道路障碍的检测准确率达到93.52%，检测速度达到60.6 FPS，能够为矿井巷道复杂环境下无人矿卡安全行驶提供保障。

Research on Infrared Obstacle Detection of Mine Roadway Based on RCR_YOLOv4

Aiming at the problem that driving obstacles such as falling rocks or pedestrians may appear on the dim road of underground mine roadway，which seriously affects the safe driving of unmanned mine card，a fast detection optimization model RCR_YOLOv4 of roadway obstacles based on infrared vision recognition was proposed.Firstly，the infrared camera was used for data acquisition，random cutting，random flipping，mirror flipping and other methods were used to expand the experimental data set.Labeling software was used for data Labeling，and the infrared obstacle data set of mine roadway was built and loaded into the obstacle detection model.Secondly，K-Means ++ optimization algorithm was used to screen the prior frame size of obstacles in the roadway，and depth separable convolution was introduced to reduce the number of network parameters and computation，so as to improve the positioning accuracy and detection efficiency of obstacle targets.The dual-channel attention mechanism is designed to optimize the network feature fusion module to realize the high-precision detection of the obstacle of unmanned mine jamming.The results show that the detection accuracy of the model can reach 93.52% and the detection speed can reach 60.6 FPS.Compared with the current popular target detection networks such as Faster_RCNN，SSD and YOLOv4，RCR_YOLOv4 also shows better comprehensive performance and can provide guarantee for the safe driving of unmanned mine cards in the complex environment of mine roadway.