为解决煤矿井下胶带异物检测受煤尘干扰、光线不均、胶带高速运动造成传统检测算法精度低等问题,文章基于YOLOv7对矿井胶带异物检测算法进行优化。首先,通过自适应对比度增强算法,强化胶带监控图像对比度,提高目标图像轮廓清晰度;其次,在主干提取网络中提出多尺度混合残差注意力机制,增强YOLOv7对异物特征提取能力与对背景干扰能力;最后,采用加权双向特征金字塔网络与4检测头输出模型预测结果,提升网络对不同尺寸异物检测效率。通过实验可得,改进后的YOLOv7模型对井下胶带异物识别精度和速度优于YOLOv5、YOLOv7,对井下胶带异物识别精度和识别速度分别为93.6%、26f/s。识别平均准确率相较于YOLOv5模型、YOLOv7模型分别提高了3.9%,3.1%;平均召回率分别提高了4.1%,3.4%;检测时间分别有0.009s,0.005s的提升。

In order to solve the problems in the foreign matter detection on belt conveyor of underground coal mine caused by the low accuracy of traditional detection algorithms under the interference of coal dust, uneven light, and high-speed movement of the conveyor belt, the detection algorithm of mine belt foreign objects was optimized based on YOLOv7. Firstly, the adaptive contrast enhancement algorithm was used to strengthen the contrast of the belt monitoring image and improve the clarity of the target image contour. Secondly, a multi-scale mixed residual attention mechanism was proposed in the backbone extraction network to enhance YOLOv7’s ability to extract foreign body features and interfere with background.Finally, the weighted bidirectional feature pyramid network and detection heads were used to output the model prediction results to improve the efficiency of foreign object detection of different sizes. According to the experiment results, the improved YOLOv7 model was superior to YOLOv5 and YOLOv7 in the recognition accuracy and speed of foreign objects in the underground belt, and the recognition accuracy and speed of foreign objects in the underground belt were 93. 6% and f/ s, respectively.Compared with YOLOv5 and YOLOv7, the average recognition accuracy rate of the proposed model was increased by 3. 9% and 3. 1%, respectively; the average recall rate was increased by 4. 1% and 3. 4%, respectively; the detection time was improved by 0. 009 s and 0. 005 s respectively.