基于KRB-YOLOv5s的煤矸识别方法

Recognition method for coal and gangue based on KRB-YOLOv5s

为解决煤矿高粉尘、低照度、高噪声与堆叠等复杂环境因素导致的煤矸识别精度低、漏检与误检问题,提出一种基于KRB-YOLOv5s算法的煤矸识别方法。采用K均值聚类(K-means++)算法对数据集进行重新聚类,以得到更精准的锚框参数;在YOLOv5s主干网络中引入大核卷积结构重参数(RepLKNet)网络,通过大核卷积架构提取目标更高层级的特征信息;在YOLOv5s颈部引入加权双向特征金字塔(Bi FPN)网络,通过双向跨尺度连接对目标多尺度特征进行快速捕捉与融合。在煤矸数据集上开展实验,结果表明:与其他YOLO系列检测算法相比,KRB-YOLOv5s算法在高粉尘、低照度、高噪声与堆叠工况下的综合检测性能最佳,识别精度均值(m AP)达94.5%,比YOLOv5s算法提高了3.3个百分点。研究结论为煤矿复杂工况下煤矸智能分选提供参考。

To solve the problems of low recognition accuracy,missed detection,and false detection of coal and gangue caused by complex environmental factors such as high dust,low illumination,and high noise in coal mines,a recognition method for coal and gangue based on KRB-YOLOv5s is proposed.The K-means++algorithm is used to re-cluster the dataset to obtain more accurate anchor box parameters.The RepLKNet network is introduced into the YOLOv5s backbone network to extract higher-level feature information of the target through a large kernel convolutional architecture.A BiFPN network is introduced into the neck of YOLOv5s to quickly capture and fuse multi-scale features of the target through bidirectional cross scale connections.Experiments are conducted on a dataset of coal and gangue,and the results show that compared with other YOLO series detection algorithms,the KRB-YOLOv5s algorithm has the best comprehensive detection performance under high dust,low illumination,and high noise conditions,with an average recognition precision(mAP)of 94.5%,which is 3.3 percentage points higher than that of YOLOv5s algorithm.The research conclusions provide a reference for intelligent sorting of coal and gangue under complex working conditions in coal mine.