单轨吊作为煤矿井下辅助运输的重要设备形式之一，具有运载能力大、爬坡能力强等优点，在煤矿智能化建设背景下无人驾驶是其必然的发展方向。为确保无人驾驶单轨吊在井下巷道内的安全行驶，对轨道接缝和关键目标的可靠检测尤为重要，从提高单轨吊无人驾驶安全通过性出发，对单轨吊无人驾驶目标识别算法与轨道接缝检测方法两大主要方面展开了研究。首先，对矿井图片数据集进行了增强处理，提高了其多样性；提出了一种加入改进通道注意力机制ECA_s和EIOU回归损失函数的YOLOv5算法，并对改进后的YOLOv5算法进行试验分析，结果表明，采用改进后的YOLOv5算法，识别准确率提高了9.1%，mAP提升了3.6%。其次，建立了基于机器视觉的单轨吊轨道接缝检测方法，采用图像预处理、直方图信息统计、形态学处理等技术，根据标定系数计算轨道接缝距离，结果表明，基于机器视觉的轨道接缝检测算法处理一张接缝图像的检测误差仅为0.3 mm。最后，开展了单轨吊无人驾驶目标检测试验，结果表明，改进后的YOLOv5目标检测算法平均精度达到90.3%，具有更高的检测准确率；单轨吊轨道接缝检测算法处理接缝图像的平均检测误差为0.73 mm、最大检测误差不超过1.1 mm，具有更高的检测精度。为保障煤矿井下单轨吊无人驾驶安全通过性和可靠性，提供了务实可行的检测方案和精确可靠的检测算法。

As one of the important forms of equipment for underground auxiliary transportation in coal mines, monorail cranes have the advantages of large carrying capacity and strong climbing ability, and unmanned driving is its inevitable development direction in the context of intelligent construction of coal mines. In order to ensure the safe driving of the unmanned monorail crane in the underground roadway, it is particularly important for the reliable detection of track joints and key targets. Firstly, the mine image dataset was enhanced to improve its diversity. A YOLOv5 algorithm with improved channel attention mechanism ECA_s and EIOU regression loss function was proposed, and the improved YOLOv5 algorithm was experimentally analyzed, and the results showed that the recognition accuracy was increased by 9.1% and mAP by 3.6% by using the improved YOLOv5 algorithm. Secondly, a machine vision-based single-track crane track seam detection method was established, and the track seam distance was calculated according to the calibration coefficient by using image preprocessing, histogram information statistics, morphological processing and other technologies, and the results showed that the detection error of a seam image processed by the track seam detection algorithm based on machine vision was only 0.3 mm. Finally, the unmanned target detection test of the monorail crane is carried out, and the results show that the average accuracy of the improved YOLOv5 target detection algorithm reaches 90.3%, which has a higher detection accuracy. The average detection error of the seam image processed by the monorail crane rail seam detection algorithm is 0.73 mm, and the maximum detection error is not more than 1.1 mm, which has higher detection accuracy. In order to ensure the safe passage and reliability of unmanned monorail cranes in coal mines, a pragmatic and feasible detection scheme and an accurate and reliable detection algorithm are provided.