基于CE-YOLOv8的炮孔识别方法  

Blast Hole Identification Method Based on CE-YOLOv8

作　　者：胡启国[1] 刘洋 余芃林 任渝荣 余汛 HU Qiguo;LIU Yang;YU Penglin;REN Yurong;YU Xun(School of Mechatronics&Vehicle Engineering,Chongqing Jiaotong University,Chongqing 400074,China;Liupanshui Huaan Blasting Engineering Co.,Ltd.,Liupanshui 553000,Guizhou,China)

机构地区：[1]重庆交通大学机电与车辆工程学院,重庆400074 [2]六盘水华安爆破工程有限公司,贵州六盘水553000

出　　处：《重庆交通大学学报（自然科学版）》2024年第11期130-138,共9页Journal of Chongqing Jiaotong University(Natural Science)

基　　金：国家自然科学基金项目(52175042)。

摘　　要：为了实现工程爆破中炸药的自动装填,现针对炮孔检测这一前置任务进行研究。由于实际工程中存在炮孔形状各异、碎石多、检测背景复杂等问题,导致检测过程中易出现遗漏和误判,因此,基于YOLOv8进行了改进,并提出了一种更高效的炮孔检测方法CE-YOLOv8。首先,将骨干网络中的C2f模块替换成C2f_DCN模块,增加了网络的空间变形适应性,从而能更准确地提取到炮孔特征,提高炮孔的检测精度。其次,在YOLOv8特征提取网络中加入改进C-CBAM注意力机制,对CBAM注意力机制的输入特征进行分组,并增加一个并行分支实现多尺度特征处理,最后通过跨空间学习模块将信息进行融合,提升模型的感受野和表征能力。最后,引入了E-IOU作为评估锚定框相互关系的计算方法,克服了传统IOU在梯度传递方面的缺陷,加快了网络的收敛。实验结果表明:改进后的模型对炮孔具有较好的识别效果,其中平均检测精度提升至98.6%,精确度和召回率分别达到95.7%和96.1%。改进的方法在爆破工程中识别炮孔从而实现智能化炸药装填具有较广泛的应用前景。In order to realize the automatic loading of explosives in engineering blasting,the pre-task of blast hole detection was studied.Due to the problems of different hole shapes,many gravels and complex detection backgrounds in the actual engineering,which were prone to omissions and misjudgments in the detection process.Therefore,a more efficient blast hole detection method CE-YOLOv8 was proposed on the improvement of YOLOv8.Firstly,the C2f module in the backbone network was replaced by a C2f_DCN module,increasing the spatial deformation adaptability of the network,so that the features of the blast hole could be extracted more accurately and the detection accuracy of the blast hole can be improved.Secondly,the improved attention mechanism C-CBAM was added into the YOLOv8 feature extraction network,the input features of the CBAM attention mechanism were grouped,and a parallel branch was added to realize multi-scale feature processing.The information was fused through the cross-space learning module to improve the receptive field and representation ability of the model.Finally,E-IOU was introduced as a calculation method to evaluate the correlation between anchor boxes,which overcame the shortcomings of traditional IOU in gradient transfer and accelerated the convergence of the network.The experimental results show that the improved model has a good identification effect on blast holes,in which the mean average Precision(mAP)is increased to 98.6%,and the Precision and Recall reaches 95.7%and 96.1%,respectively.The improved method has a wide application prospect for identifying blast holes in blasting engineering,so as to realize intelligent explosive loading.

关 键 词：机电工程 目标检测 注意力机制 炸药装填 炮孔识别 

分 类 号：U445.53[建筑科学—桥梁与隧道工程] TP242.6[交通运输工程—道路与铁道工程]