融合深度学习与特征投影曲线的钢筋绑扎节点检测方法

Rebar Binding Crosspoints Detection Method Based on Deep Learning and Feature Projection Curve

针对传统人工钢筋绑扎过程中操作效率低、成本高等问题,将深度学习技术应用于钢筋绑扎节点检测,提出一种融合深度学习与特征投影曲线的钢筋绑扎节点检测方法.首先,通过原始数据采集和增强构建3300幅图像的钢筋绑扎节点数据集,基于TensorFlow深度学习框架构建MobileNetV3-SSD钢筋绑扎节点识别模型,实现钢筋已绑扎/未绑扎节点的自动识别及分类,初步提取预测框区域图像;然后,对提取的未绑扎节点预测框区域进行图像预处理,提出一种由控制因子α与特征投影曲线相结合的节点定位方法实现未绑扎节点精确定位;最后,通过性能测试实验对方法评估,并根据评价指标确定识别模型与定位方法的最优阈值.实验结果表明,采用MobileNetV3-SSD钢筋绑扎节点识别模型的宏精确率和宏召回率分别为95.40%和96.70%;采用特征投影曲线定位方法的定位准确率达90.45%,平均相对误差为6.15%;所提方法可实现快速、非接触的智能化检测,具有良好的鲁棒性和应用价值.

Focused on the problems of cumbersome operation,low efficiency,and high cost in the traditional manual rebar binding process,a rebar binding crosspoints detection method based on deep learning and fea-ture projection curve is proposed.Firstly,a dataset of 3300 rebar binding crosspoints is built by original data collection and data enhancement,and a MobileNetV3-SSD rebar binding crosspoints recognition model is built based on TensorFlow deep learning framework to achieve automatic recognition and classification of rebar bound or unbound crosspoints,which is used to extract the prediction box of unbound crosspoints.Secondly,an image post-processing procedure is developed for the extracted prediction box region of un-bound crosspoints,and a crosspoints localization method combining control factorαand feature projection curve is introduced to achieve the localization of unbound crosspoints.Finally,the algorithm is evaluated by performance test experiments to determine the optimal threshold of the recognition model and localization method.Experimental results show that the macro precision and macro recall of the proposed model are 95.40%and 96.70%,respectively,and the localization accuracy of the proposed method is 90.45%with an average relative error of 6.15%.The proposed method can achieve fast,non-contact and intelligent detection of rebar binding crosspoints,which has good robustness and application value.