基于轻量化与多信息融合的输送带块煤实时监测方法

Real Time Monitoring Method of Lump Coal on Conveyor Belt Based on Lightweight and Multi-information Fusion

煤炭在国家工业体系和社会发展中有着举足轻重的作用,针对煤炭运输过程中的块煤监测问题,提出一种基于轻量化和多信息融合(lightweightandmulti-informationfusion,LM)的实时监测方法——LMYOLOv5。首先,利用自适应直方图均衡化进行数据预处理;其次,引入Ghost轻量化卷积,减少计算量和特征提取的冗余性;最后,结合协同注意力(coordinate attention, CA)机制改善特征提取的倾向性,引入双向特征金字塔网络(bidirectional feature pyramid network, Bi FPN)机制实现跨阶段的信息融合。实验结果表明,改进后的LMYOLOv5算法有明显优势。参数量和浮点计算量分别减少约62.28%和67.66%。模型训练时长减少约21.78%,模型体积也从92.7 M压缩至35.1 M。此外,精确度和召回率分别提升约0.103%和1.395%,实时监测速度提升约38.05%。

Coal played a pivotal role in the country's industrial system and social development,to address the problem of monitoring lump coal during coal transportation,a real-time monitoring method LM YOLOv5 based on lightweight and multi-information fusion is proposed.Firstly,adaptive histogram equalization is utilized for data preprocessing.Secondly,Ghost lightweight convolution is introduced to reduce the computing volume and redundancy of feature extraction in the convolution process.Finally,combining with Coordinate attention(CA)mechanism to improve the tendency of feature extraction,and Bi FPN multi-information fusion mechanism is launched to realize cross-stage information fusion.The experimental results show that the improved LM YOLOv5 algorithm has obvious advantages.The number of parameters and floating-point operations by about 62.28%and 67.66%,respectively.The model training time is reduced by about 21.78%.The model volume is also compressed from 92.7M to 35.1M.In addition,the precision and recall rates improved by about 0.103%and 1.395%,respectively.The real-time monitoring performance is improved by about 38.05%.