Rebar Cross-section Detection Based on Improved YOLOv5s Algorithm

Rebar is an indispensable material in the construction industry and the counting of rebar is a very important part of the production, transportation, and use process. Due to the problems of dense cross-sections and mutual adhesion of cross-sections boundaries for bundles of rebar, most of the rebar is still counted manually. To address the above problems, this paper proposes an improved YOLOv5s-based rebar cross-section detection method, which aims to solve the problems of missed and false detection in dense small object detection. The K-means algorithm is used to re-cluster the anchor box size; the ECANet (Efficient Channel Attention Net) module is integrated with the backbone network to adjust the attention weight of the feature map and improve the feature extraction ability of the network; To address the problem that small object with a scale smaller than 8×8 in the network cannot be detected, a new detection layer scale is redesigned by fusing the high-level semantic information with the low-level features to reduce the network's miss detection rate for small object. Experiments were conducted on a homemade rebar cross-section dataset with mAP0.5 of 94.3%, which is 6.2% improvement compared with the baseline model YOLOv5s. The improved network was able to better identify the cross-sectional features of the rebar, thus significantly reducing the leakage rate in rebar counting detection.