面向多类路面病害的智能集成检测方法

An Intelligent Integrated Detection Method for Multiple Types of Pavement Distress

为提升路面病害检测效率,弥补当前病害排查手段低效费时的弊端,本文以深度学习技术为支撑,结合图像分类与目标检测技术,提出一种面向多类型路面病害的集成性智能检测方法。该方法将路面种类分类、路面病害判别和路面病害检测3个功能模块进行有效集成。首先,调整VGG-16算法,并加入SE注意力机制;其次,优化YOLOv7检测网络,添加小目标检测层的同时融入CBAM前馈卷积注意模块。结果表明,调整后的VGG-16网络在路面种类分类、路面病害判别任务上准确率均在98%以上,优化后的YOLOv7使沥青、混凝土、砌块路面检测平均精度分别提高了3.00%,1.80%,3.90%。经实地测试,3个模块平均准确率分别为99.72%,98.28%,91.52%,整体方法综合准确率为89.69%。研究结果为路面病害快速筛查,实现整体评估提供有效参考。

In order to improve the pavement distress detection efficiency and make up for the inefficient and time-consuming limitations of current distress detection methods,an integrated intelligent detection method for multi-type pavement distress was proposed based on deep learning technology and combined image classification and target detection technology.The method effectively integrates three functional modules of pavement classification,pavement distress identification and pavement distress detection.Firstly,VGG-16 algorithm is adjusted and SE attention mechanism is added.Secondly,the YOLOv7 detection network is optimized,and the CBAM feedforward convolutional attention module is integrated while the small target detection layer is added.The results show that the adjusted VGG-16 network is more than 98%accurate in the task of pavement classification and pavement distress identification,and the optimized YOLOv7 can improve the average accuracy of asphalt,concrete and block pavement detection by 3.00%,1.80%and 3.90%,respectively.Through field tests,the average accuracy of the three modules is 99.72%,98.28%and 91.52%respectively,and the comprehensive accuracy of the whole method is 89.69%.The results of this study provide an effective reference for the rapid screening and overall evaluation of pavement distress.