随着智慧化矿山的建设和视频监控的广泛应用，采用计算机视觉技术对视频监控内容实时监测与行为分析已经受到越来越多的关注。由于以往胶带运输机的故障分析与速度监测无法实时进行智能预警与动态调控，胶带运输机的监测方式在煤矿智能化建设中面临巨大的挑战，亟需将计算机视觉技术和深度学习方法应用于煤矿相关研究中，实现煤矿视频监控下胶带运输机速度的实时准确监测。针对胶带运输机的速度智能化监测，以实现及时有效的风险预警和胶带运输机智能化调速为目标，本文以胶带运输机上的煤块为研究对象，提出了一种结合改进的YOLOv7-tiny和ByteTrack算法的胶带运输机运行速度监测方法。具体研究内容如下：

（1）针对煤矿环境下视频图像模糊和背景干扰导致YOLOv7-tiny算法对煤块的检测准确性不高存在误检问题，本文提出了一种改进的YOLOv7-tiny煤块检测算法。首先，将递归门控卷积模块引入主干网络，构建了ELAN-GN模块，用于捕获煤块的边缘细微特征及高阶特征，增强网络对煤块特征的表达能力。其次，将注意力机制SimAM模块嵌入路径聚合网络中，加强网络对煤块的感知能力，弱化背景信息的干扰。最后，采用引入角度损失的SIoU损失函数加速了网络模型的快速收敛，提高了模型的鲁棒性。实验结果表明，本文提出的改进YOLOv7-tiny算法在满足工业实时性要求的前提下，煤块检测准确率提高了5.7%，平均精度值提高了3.7%。

（2）针对煤块运动发生形变、轨迹漂移和跟踪丢失导致胶带运输机速度测量误差大等问题，本文提出了基于ByteTrack算法的胶带运输机速度测量方法。首先，采用改进的YOLOv7-tiny煤块检测算法作为检测器，实现图像中煤块的准确定位。其次，利用二次数据关联策略加强低分置信度煤块的关联匹配，同时采用引入置信度的卡尔曼滤波，将检测结果的置信度融入卡尔曼滤波的状态估计，使其卡尔曼滤波预测轨迹倾向于检测器的结果，提高煤块跟踪轨迹的稳定性。最后，通过胶带运输机两个支撑杆间的图像坐标及支撑杆间的距离简化相机标定方法，建立胶带运输机速度测量模型，实现胶带运输机的速度估计。实验结果表明，改进的ByteTrack算法有效改善了煤块跟踪的稳定性和连续性。同时，在胶带运输机运行视频数据上，验证了结合深度学习的胶带运输机速度测量方法的有效性，降低了胶带运输机的速度测量误差。

本文提出的基于深度学习的胶带运输机速度监测方法，不仅能够实现胶带运输机上煤块的准确检测，也能实现胶带运输机的运行速度测量，有效改善了胶带运输机在运行过程中的智能化监测以及智能化速度调控方式，对于煤矿的安全生产具有重要的意义。

With the construction of intelligent mines and the widespread use of video surveillance, the use of computer vision technology for real-time monitoring and behavior analysis of video surveillance content has received more and more attention. Since failure analysis and speed monitoring of belt conveyor in the past could not be carried out in real time for intelligent warning and dynamic regulation, the monitoring method of belt conveyor faces great challenges in the intelligent construction of coal mines, and there is an urgent need to apply computer vision technology and deep learning methods in coal mines related research to realize real-time accurate monitoring of belt conveyor speed under coal mine video surveillance. Aiming at the intelligent monitoring of the speed of the belt conveyor, with the goal of achieving timely and effective risk warning and intelligent speed regulation of the belt conveyor, this thesis proposes a method for monitoring the operating speed of the belt conveyor combining the improved YOLOv7-tiny and ByteTrack algorithms, taking the coal blocks on the belt conveyor as the research object. The specific research contents are as follows:

(1) To address the problem of false detection due to blurred video images and background interference in the coal mine environment resulting in low accuracy of the YOLOv7-tiny algorithm for coal block detection, an improved YOLOv7-tiny coal block detection algorithm is proposed in this thesis. First, the recursive gated convolution module is introduced into the backbone network, and the ELAN-GN module is constructed to capture the edge minutiae features and higher order features of the coal block to enhance the network's ability to express the features of the coal block. Second, the attention mechanism SimAM module is embedded in the path aggregation network to enhance the network's ability to perceive coal blocks and weaken the interference of background information. Final, the SIoU loss function with the introduction of angular loss accelerates the fast convergence of the network model and improves the robustness of the model. The experimental results on the self-built coal block dataset show that the improved YOLOv7-tiny algorithm proposed in this thesis improves the coal block detection accuracy by 5.7% and the average accuracy value by 3.7% while meeting the industrial real-time requirements.

(2) To address the problems of deformation of coal block motion、trajectory drift and tracking loss leading to large errors in belt conveyor speed measurement, this thesis proposes a belt conveyor speed measurement method based on ByteTrack algorithm. First, the improved YOLOv7-tiny coal block detection algorithm is used as a detector to achieve accurate positioning of coal blocks in images. Second, a secondary data association strategy is used to strengthen the association matching of low-score confidence coal blocks, while a Kalman filter with introduced confidence is used to integrate the confidence of detection results into the state estimation of Kalman filter, so that its Kalman filter prediction trajectory tends to the detector results and improves the stability of coal block tracking trajectory. Finally, the camera calibration method is simplified by the image coordinates between belt conveyor two support rods and the distance between support rods, and the belt conveyor speed measurement model is established to realize the speed estimation of the belt conveyor. The experimental results show that the improved ByteTrack algorithm effectively improves the stability and continuity of coal block tracking. Meanwhile, the effectiveness of the belt conveyor speed measurement method combined with deep learning is verified on the video data of the belt conveyor operation, and the speed measurement error of the belt conveyor is reduced.

The belt conveyor speed monitoring method based on deep learning proposed in this thesis can not only enable the accurate detection of coal blocks on the belt conveyor, but also the operation speed measurement of the belt conveyor, which effectively improves the intelligent monitoring of the belt conveyor during operation and the intelligent speed regulation method, which is of great significance for the safe production of coal mines.