关键词: 机电工程；目标检测；注意力机制；炸药装填；炮孔识别

Abstract: 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.

Key words: mechatronics engineering; object detection; attention mechanisms; explosive loading; blast hole identification