基于改进YOLOv8的轻量化钢筋端面检测算法研究

Lightweight Rebar End Detection Algorithm Based on Improved YOLOv8

【目的】解决端面密集、直径尺度不统一、端面边界粘连、端面与背景融合以及端面之间存在遮挡等问题下的钢筋精确计数。【方法】提出了一种改进的轻量化YOLOv8模型框架,引入了空间和通道重建卷积模块(SCConv)和针对小目标检测的归一化Wasserstein距离(NWD)损失函数。【结果】消融试验的结果表明,SCConv模块可以在大幅减少网络参数的情况下保持网络性能,而NWD损失函数可以有效提高钢筋端面检测模型的精度,为高精度、轻量化的钢筋计数问题提供有效的解决方案。

【Purposes】 Rebar plays an indispensable role in construction engineering;however, challenges such as densely packed end faces, non-uniform diameter scales, adhesive boundaries, background fusion, and occlusions between end faces have made precise counting a significant challenge. In recent years, deep learning has made remarkable strides in the field of dense object counting. Nonetheless, deep leaming faces limitations because of the need for large-scale data and computational resources, hindering its practical application. 【Methods】 In response to these challenges, an enhanced YOLOv8 model framework is introduced for rebar end detection. The framework incorporates Spatial and Channel reconstruction Convolutional(SCConv) modules and the Normalized Wasserstein Distance(NWD) loss function tailored for small object detection. 【Findings】 Experimental results from ablation tests demonstrate that the SCConv module significantly reduces network parameters while maintains network performance. Furthermore, the NWD loss function notably enhances the accuracy of rebar end detection in large models. This research provides an effective solution for achieving high-precision and lightweight rebar counting.