摘要
煤炭开采过程中原煤夹杂的矸石对煤质有较大的影响,基于图像识别技术的智能化矸石识别对推动煤炭无人化开采具有重要意义。经典的卷积神经网络识别煤和矸石的方法,在网络模型训练过程关注了全局特征,而对局部显著区域特征的关注有待于进一步加强。为此,针对井下大巷皮带主煤流运输过程中的矸石识别问题,提出一种改进空间通道注意力与残差融合算法,该算法将改进的空间注意力机制和改进的通道注意力机制引入残差网络,实现煤矸石识别。在空间注意力维度上,通过全局最大池化和全局标准差池化实现煤与矸石特征提取。在通道注意力维度上,通过全局平均池化和全局最大池化两路实现煤与矸石特征提取。模型以煤和矸石的实时监测图像为输入,以被识别对象的类别为输出,实现煤和矸石的识别。实验表明:所提出的融合模型对矸石的识别准确率为96.2%,相比于残差网络、空间注意力残差网络、通道注意力残差网络、注意力卷积网络,矸石识别精度分别提高了4.1%,3.4%,2.6%和1.5%。
The gangue in raw coal has an impact on coal quality in the mining.The intelligent gangue identification based on the image recognition technology is of great significance for the unmanned coal mining.The classical convolutional neural network method for recognizing coal and gangue focuses on the global features in the process of network model training,ignoring the local salient region features.Aiming at the gangue monitoring problem in the coal flow transport process,an improved recognition model of the convolution neural network is constructed by fusing with the attentions.For the model,the improved attentions of the channel and the space are introduced into the convolution neural network.In the spatial attention dimension,the gangue feature is extracted by using the globally maximum pool and the globally standard deviation pool.In the channel attention dimension,the gangue feature is extracted by using the globally average pool and the globally maximum pool.To realize the recognition of the coal and the gangue,the constructed model takes the real-time monitoring image of the coal or the gangue as input,and the category probability of the identified object as output.Finally,the real-time image recognition experiments of the coal or gangue on the coal flow belt in Caojiatan Coal Mine of Shaanxi Province show that the accuracy of the improved attention convolution network model is as high as 95.2%.Compared with the classical convolution network,the spatial attention network and the channel attention network,the recognition accuracy of the gangue is increased by 4.1%,3.4%,2.6%,1.5%,respectively.
作者
韩存地
朱兴攀
符立梅
董立红
刘安强
李远成
许犇
汪梅
HAN Cundi;ZHU Xingpan;FU Limei;DONG Lihong;LIU Anqiang;LI Yuancheng;XU Ben;WANG Mei(Caojiatan Coal Industry Co.,Ltd.,Yulin 719000,China;SHCCIG Yubei Coal Industry Co.,Ltd.,Yulin 719000,China;College of Computer Science and Engineering,Xi’an University of Science and Technology,Xi’an 710054,China;College of Electrical and Control Engineering,Xi’an University of Science and Technology,Xi’an 710054,China)
出处
《西安科技大学学报》
CAS
北大核心
2021年第6期1113-1121,共9页
Journal of Xi’an University of Science and Technology