基于3D-Attention与多尺度的矿井人员行为识别算法  

Mine personnel behavior recognition algorithm based on 3D-Attention and multi-scale

作　　者：李占利[1] 权锦成 靳红梅[1] Li Zhanli;Quan Jincheng;Jin Hongmei(College of Computer Science and Technology,Xi'an University of Science and Technology,Xi'an 710054,Chin)

机构地区：[1]西安科技大学计算机科学与技术学院,西安710054

出　　处：《国外电子测量技术》2023年第7期95-104,共10页Foreign Electronic Measurement Technology

摘　　要：为了解决现有行为识别模型在矿井环境下识别率低,对矿井下环境的适应性较弱,不具备适用性的问题,提出基于3D-Attention与多尺度(CSAD)的矿井人员行为识别算法。针对国内外现有矿工行为数据集较为匮乏的问题,自建矿工行为数据集;其次,针对煤矿井下视频动态变化的问题,提出3D多尺度卷积模块,通过学习不同尺度的特征,提升模型的泛化性,增强模型对不同煤矿环境的适应性;考虑到模型在煤矿井下环境中识别率较低的问题,提出改进的A3D-Net注意力模块,使模型更加专注于识别区域的特征提取,进而提升模型的准确率。实验结果表明,在公共数据集UCF101、KTH上进行实验,提出的CSAD模型准确率分别达到89.9%、92.7%,在自建矿工行为数据集上进行试验,模型准确率达到74.98%,在使用视频增强预处理后,准确率达到了76.42%。In order to solve the problem that the existing behavior recognition model has low recognition rate,weak adaptability to the mine environment,and no applicability in the mine environment.A mine personnel behavior recognition algorithm based on 3D-Attention and multi-scale(CSAD)is proposed.Aiming at the problem of lack of existing miners'behavior data sets at home and abroad,a self-built miners'behavior data set was built.Secondly,a 3D multi-scale convolution module was proposed to improve the generalization of the model and enhance the adaptability of the model to different coal mine environments by learning the features of different scales.Considering the problem of low recognition rate of the model in the coal mine environment,an improved A3D-Net attention module is proposed to make the model focus more on the feature extraction of the recognition area,there by improving the accuracy of the model.Experimental results show that the accuracy of the proposed CSAD model is 89.9%and 92.7%on the public data sets UCF101 and KTH,and the accuracy of the model is 74.98%on the self-built miner behavior data set,and the accuracy is 76.42%after using video enhancement preprocessing.

关 键 词：深度学习 煤矿井下行为识别 注意力机制 神经网络 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]