基于数字图像的山区公路涵洞病害识别方法研究

Research on Identification Methods of Diseases in Mountainous Highway Culverts Based on Digital Images

摘要:

涵洞病害检测对于确保结构安全和公路运行至关重要，传统的检测方法通常依赖人工进入涵洞进行目视检查，但这种方法受限于山区涵洞中空间狭窄及较高的安全风险，导致检测效率低下且精度不高，病害的跟踪和修复也存在困难。因此，针对以上问题提出了一种基于YOLOv8模型的涵洞病害自动识别方法。该方法通过对预处理后的图像数据集进行YOLOv8模型训练，生成训练模型以识别涵洞病害图像。实验结果表明，采用该方法能够有效识别涵洞的洞身、盖板、涵底铺砌的裂缝、表面脱落等病害，模型检测精度达到0.831，调和平均数为0.796，表明该方法具有较好的病害识别性能。

Abstract:

Culvert disease detection is essential to ensure the structural safety and highway operation, and the traditional detection methods usually rely on the manual entry into culverts for visual inspection, but this method is limited by the narrow space and high safety risk in mountainous culverts, resulting in low detection efficiency and low accuracy. And it is also difficult to track and repair diseases. Therefore, aiming at the above problems, an automatic identification method of culvert disease based on YOLOv8 model is proposed. In this method, the YOLOv8 model is trained on the preprocessed image data set to generate a training model to identify the culvert disease images. The experimental results show that the used method can effectively identify the cracks and surface shedding diseases of the culvert body, cover plate and culvert bottom paving. The detection accuracy of the model reaches 0.831, and the average harmonic value is 0.796, indicating that the method has a relatively good performance in disease identification.