随着矿井智能化技术的不断推进，基于计算机视觉方法，自动实现瓦斯抽采的智能计数成为可能。但是，由于煤矿井下作业场所狭小，钻机被遮挡的情况经常发生，影响计数准确性。因此，本文提出一种结合改进YOLOv8(You Only Look Once Version 8)和SiamBan(Siamese Box Adaptive Network)算法的井下钻机运行监测方法，通过分析钻机运动轨迹，计算出钻杆数量，实现钻杆的准确计数。

（1）针对钻场环境中的钻机目标被遮挡，导致漏检、误检的问题，提出一种改进的YOLOv8钻机检测算法。首先，设计了深度坐标注意力模块(Deep Coordinate Attention，DCA)作为骨干网络，其包含坐标注意力分支和深度可分离卷积分支，能够更有效地从复杂背景中分离出钻机目标，降低对遮挡物的关注程度；其次，使用渐进式特征融合网络(Asymptotic Feature Pyramid Network，AsymptoticFPN)进行特征融合，从而有效地检测在不同遮挡尺度下的钻机目标；最后，引入抗遮挡损失函数(Repulsion Loss)，通过增强钻机与遮挡物之间的互斥作用，从而减少钻机与矿工重叠检测引起的误检情况。为了验证该方法的有效性，在自建的钻机遮挡数据集上进行实验验证。与其他方法相比，本文提出改进的YOLOv8钻机检测算法对钻机目标检测精度提高了4.6%，遮挡物的检测精度提升了1.9%。

（2）针对钻场环境中的钻机目标被遮挡，导致目标跟踪偏移、丢失的问题，提出一种改进的SiamBan钻机跟踪算法。在SiamBan网络基础上，设计了一种新的遮挡判断机制(Peak Oscillation Quality，POQ)，通过分析孪生网络输出响应图对遮挡情况进行判定和分级。当出现部分遮挡导致跟踪偏移时，根据历史运动信息，引入余弦相似度函数来确定当前帧响应图的最大峰值，从而确定钻机位置；当出现完全遮挡导致跟踪框丢失时，使用联合长短期记忆网络的预测模块预测钻机位置。为了验证该方法的有效性，在自建的钻机遮挡数据集上进行实验验证。实验结果表明，所提改进算法可以有效改善跟踪框的稳定性。与其他方法相比，本文提出改进的SiamBan钻机跟踪算法的跟踪成功率平均提升了3%，跟踪精确度平均提升了1.8%。

（3）以提出的方法为基础，分析钻机的运动轨迹，并采用滤波算法对运行结果进行去噪，以获得平滑的运动轨迹，通过处理运行轨迹中的局部峰值，进一步得到钻杆数量。此外，本文设计并开发了一个钻杆计数系统，该系统基于B/S架构，采用Vue和Django进行前后端开发，并集成本文所提算法。通过该原型系统，验证所提算法的应用可行性，为智能煤矿井下钻杆计数提供理论依据和技术支撑。

With the continuous advancement of mine intelligent technology, it has become possible to automatically realize intelligent counting of gas drainage based on computer vision methods. However, due to the small working space underground in coal mines, drilling rigs are often blocked, which affects the accuracy of counting. Therefore, this article proposes an underground drilling rig operation monitoring method that combines the improved YOLOv8 (You Only Look Once Version 8) and SiamBan (Siamese Box Adaptive Network) algorithms. By analyzing the movement trajectory of the drilling rig, the number of drill pipes is calculated and the accuracy of the drill pipes is achieved.

(1) Aiming at the problem that the drilling rig target in the drilling field environment is blocked, resulting in missed detection and false detection, an improved YOLOv8 drilling rig detection algorithm is proposed. First, the Deep Coordinate Attention module (Deep Coordinate Attention, DCA) is designed as the backbone network, which includes the coordinate attention branch and the depth separable convolution branch, which can more effectively separate the drilling target from the complex background and reduce occlusion. The degree of attention of objects; secondly, use the Asymptotic Feature Pyramid Network (AsymptoticFPN) for feature fusion to effectively detect drilling rig targets under different occlusion scales; finally, introduce the anti-occlusion loss function (Repulsion Loss), By enhancing the mutual exclusion between the drilling rig and the obstruction, false detections caused by overlapping detection of the drilling rig and miners are reduced. In order to verify the effectiveness of this method, experimental verification was conducted on a self-built drilling rig occlusion data set. Compared with other methods, the improved YOLOv8 drilling rig detection algorithm proposed in this article improves the drilling rig target detection accuracy by 4.6% and the obstruction detection accuracy by 1.9%.

(2) Aiming at the problem that the drilling rig target is occluded in the drilling field environment, resulting in target tracking deviation and loss, an improved SiamBan drilling rig tracking algorithm is proposed. Based on the SiamBan network, a new occlusion judgment mechanism (Peak Oscillation Quality, POQ) is designed to determine and classify occlusion situations by analyzing the twin network output response map. When partial occlusion occurs, causing tracking offset, a cosine similarity function is introduced based on historical motion information to determine the maximum peak value of the current frame response map, thereby determining the position of the drilling rig; when complete occlusion occurs and the tracking frame is lost, joint long and short-term memory is used The network’s prediction module predicts rig locations. In order to verify the effectiveness of this method, experimental verification was conducted on a self-built drilling rig occlusion data set. Experimental results show that the proposed improved algorithm can effectively improve the stability of the tracking frame. Compared with other methods, the tracking success rate of the improved SiamBan drilling rig tracking algorithm proposed in this article has increased by an average of 3%, and the tracking accuracy has increased by an average of 1.8%.

(3) Based on the proposed method, analyze the movement trajectory of the drilling rig, and use a filtering algorithm to denoise the operation results to obtain a smooth movement trajectory. By processing the local peaks in the operation trajectory, the number of drill pipes is further obtained. In addition, this article designed and developed a drill pipe counting system, which is based on B/S architecture, uses Vue and Django for front-end and back-end development, and integrates the algorithm proposed in this article. Through this prototype system, the application feasibility of the proposed algorithm is verified, and the theoretical basis and technical support are provided for intelligent coal mine underground drill pipe counting.