摘要
准确识别护帮板支护状态,判断护帮板是否与采煤机发生干涉,是实现煤矿安全生产的重要一环。提出了一种基于改进YOLOv5s的护帮板异常检测方法。建立了护帮板数据集hb_data2021,对YOLOv5s模型进行改进。根据基于改进YOLOv5s的护帮板状态检测结果的标签分类,判断护帮板状态是否异常。为了减小YOLOv5s模型的参数量,采用MobileNetV3和轻量级注意力机制NAM (normalization-based attention module,标准化注意力模块)替换主干特征提取网络。为了提高护帮板检测精度,改进损失函数为α-CIoU,并进行知识蒸馏。实验结果表明:蒸馏后的网络平均精度提高了1.0%,参数量减小了33.4%,推理加速34.2%;基于改进YOLOv5s的护帮板异常检测方法效果良好,将其部署在NVIDIA Jetson Xavier平台上,可以满足实时检测视频的要求。将检测模型移植到巡检机器人的嵌入式平台上,可以实现护帮板异常检测,满足煤矿工业实际需求。
It is an important link to realize the safe production of coal mine to accurately identify the support state of the side guard and judge whether the side guard interferes with the shearer. This paper presented an anomaly detection method of side guard based on improved YOLOv5s, which set up a data set of side guard called hb_data2021 and improved the YOLOv5s model. It could be judged whether the state of the side guard was abnormal, according to the label classification based on the detection results of the improved YOLOv5s. In order to reduce the parameters of YOLOv5s model, MobileNetV3 and the lightweight attention mechanism NAM(normalization-based attention module) were used to replace the backbone feature extraction network. In order to improve the detection accuracy of side guard, the loss function was improved to α-CIoU and knowledge distillation was conducted. The experimental results showed that after distillation, the average precision of the network was improved by 1.0%, the parameters was reduced by 33.4%, and the reasoning speed was accelerated by 34.2%;the abnormal detection method of side guard based on the improved YOLOv5s had a good effect. It could be deployed on the NVIDIA Jetson Xavier platform to meet the requirements of real-time video detection.Transplanting the detection model to the embedded platform of the patrol robot can realize the abnormal detection of the side guard and meet the actual needs of the coal industry.
作者
张旭辉
闫建星
麻兵
鞠佳杉
沈奇峰
吴雨佳
ZHANG Xu-hui;YAN Jian-xing;MA Bing;JU Jia-shan;SHEN Qi-feng;WU Yu-jia(School of Mechanical Engineering,Xi'an University of Science and Technology,Xi'an 710054,China;Shaanxi Key Laboratory of Mine Electromechanical Equipment Intelligent Monitoring,Xi'an 710054,China)
出处
《工程设计学报》
CSCD
北大核心
2022年第6期665-675,共11页
Chinese Journal of Engineering Design
基金
国家自然科学基金资助项目(51974228,51834006)
陕西省创新人才计划项目(2018TD-032)
陕西省重点研发计划项目(2018ZDCXL-GY-06-04)。
关键词
护帮板
目标检测
轻量化
边框损失函数
嵌入式设备
side guard
target detection
lightweight
frame loss function
embedded device
