融合轻量级网络和双重注意力机制的煤块检测方法

Coal block detection method integrating lightweight network and dual attention mechanism

  • 摘要: 针对现有煤矿井下带式输送机上煤块检测方法存在检测精度低、检测速度慢等问题,提出了一种融合轻量级网络和双重注意力机制的改进YOLOv4模型,并将其应用于带式输送机煤块检测。改进YOLOv4模型采用K-means聚类算法重新聚类先验框,使先验框更适应检测目标尺寸;通过引入MobileNet轻量级网络模型改进主干网络结构,以减少模型的参数量和计算量,提高检测速度;嵌入具有双重注意力机制的卷积块注意模块,用于提高模型对目标特征的敏感度,抑制干扰信息,提高目标检测精度。实验结果表明,改进YOLOv4模型能准确检测出不同尺寸的煤块;相较于YOLOv4模型,改进YOLOv4模型权重文件减少了36.46%,精确率提高了2.16%,召回率提高了20.4%,平均精度均值提高了14.37%,漏检率降低了16%,检测速度提升了19帧/s,处理单张图像耗时减少了1.31 s,提高了煤块检测精度和检测速度。

     

    Abstract: In order to solve the problems of low detection precision and slow detection speed of existing coal block detection methods on belt conveyor in underground coal mine, an improved YOLOv4 model integrating lightweight network and dual attention mechanism is proposed, and it is applied to coal block detection of belt conveyor. The improved YOLOv4 model uses K-means clustering algorithm to re-cluster the prior frames, so that the prior frames are more suitable for the size of the detected target. The model improves the backbone network structure by introducing the MobileNet lightweight network model to reduce the amount of model parameters and calculations, and improve the detection speed. A convolution block attention module with dual attention mechanism is embedded to improve the sensitivity of the model to target characteristics, suppress interference information and improve the precision of target detection. The experimental results show that the improved YOLOv4 model can detect coal blocks of different sizes accurately. Compared with the YOLOv4 model, the improved YOLOv4 model weight file is reduced by 36.46%, the accuracy rate is increased by 2.16%, the recall rate is increased by 20.4%, the average accuracy is increased by 14.37%, the missed detection rate is decreased by 16%, the detection speed is increased by 19 frames/s, the processing time for a single image is reduced by 1.31 s, which improves the detection precision and speed of coal block detection.

     

/

返回文章
返回