基于改进YOLOv5s的煤矿输送带异物检测算法
An Improved YOLOv5s-based Foreign Object Detection Algorithm for Coal Mine Conveyor Belts
张炳建 1郝洪涛 1刘秀娟 1李泽旭1
作者信息
- 1. 宁夏大学机械工程学院,宁夏 银川 750021
- 折叠
摘要
为了解决现有煤矿输送带异物检测模型存在的参数量大、占用计算机资源多、检测异物种类少等问题,对YOLOv5s目标检测算法进行了优化.首先,将轻量化卷积神经网络ShuffleNetv2 作为YOLOv5s骨干网络并对异物图像进行了特征提取,进而减少了模型参数量,提高了网络并行度;其次,将双向特征金字塔网络作为特征融合网络,融合了不同特征图尺度中的细节信息;最后,添加了坐标注意力机制,增强了特征提取能力,强化了异物目标关注度,从而提高了网络模型检测精度.实验结果显示,与原始模型相比,基于改进YOLOv5s的目标检测网络模型其参数量压缩为 3.60×106 个,检测帧率提升了 8.4%,表明该算法能够在计算资源较少的情况下实现快速、准确的煤矿输送带异物检测.
Abstract
In order to address the issues of large parameter size,high demand for computer resources,and limited types of detected foreign objects in existing detection models for coal mine conveyor belts,YOLOv5s object detection algorithm was optimized.Firstly,a lightweight convolutional neural network,ShuffleNetv2,was adopted as the backbone network of YOLOv5s to extract features from foreign object images,reducing the parameters of the model and improving network parallelism.Secondly,a bidirectional feature pyramid network was used as the feature fusion network to integrate detailed information from different feature map scales.Finally,a coordinate attention mechanism was added to enhance the feature extraction capability,strengthen the focus on foreign object targets,and thereby improve the detection accuracy of network model.Experimental results show that compared to the original model,the improved YOLOv5s-based object detection network model achieved a parameter compression of 3.6×106,and an 8.4%increase in detection frame rate,demonstrating that this method can achieve rapid and accurate foreign object detection on coal mine conveyor belts with fewer computational resources.
关键词
煤矿输送带/异物检测/YOLOv5s/特征融合网络/注意力机制Key words
coal mine conveyor belt/foreign object detection/YOLOv5s/feature fusion network/attention mechanism引用本文复制引用
基金项目
宁夏回族自治区重点研发计划项目(2023ZDYF0142)
宁夏自然科学基金项目(2021AAC03046)
出版年
2024
国家科技期刊平台
NSTL
万方数据