铁道建筑

针对地铁轨道板表面裂缝检测误报率高、准确度低的问题，本文提出了一种基于改进TOOD模型的地铁轨道板表面裂缝检测算法。使用自走行地铁探伤车沿线路进行数据采集，获取高清轨道板表面数据，并使用图像增强算法对数据进行预处理，以模拟不同天气状况和拍摄环境，扩大数据适用范围。修改TOOD模型的骨干网络为ResNeXt，可增强模型特征提取能力，并插入全局注意力机制（GAM）模块来提升模型对细小裂缝的识别能力，同时将损失函数修改为CIoU，提高模型的回归精度。在裂缝数据集上对模型的能力进行了验证。结果表明：相较于传统TOOD模型，改进后均值平均精度（mAP）提升了8.2%，达到75.6%；在复杂环境下，本文模型对地铁轨道板细小裂缝表现出良好的鲁棒性和检测能力，可提高地铁的运维检修效率。

Addressing the issues of high false alarm rates and low accuracy in the detection of surface cracks on subway track slabs， this paper proposes a subway track slab surface crack detection algorithm based on an improved TOOD model. A self-propelled subway flaw detection vehicle is used to collect data along the line， obtaining high-definition track slab surface data. Image enhancement algorithms are applied to preprocess the data， simulating different weather conditions and shooting environments to expand the data's applicability. The backbone network of the TOOD model is modified to ResNeXt， enhancing the model's feature extraction capability. A Global Attention Mechanism (GAM) module is inserted to improve the model's ability to recognize fine cracks. Additionally， the loss function is modified to CIoU， improving the model's regression accuracy. The model's performance is validated on a crack dataset. The results show that， compared to the traditional TOOD model， the improved model achieves an 8.2% increase in mean Average Precision (mAP)， reaching 75.6%. In complex environments， the proposed model exhibits good robustness and detection capability for fine subway track slab cracks， which can improve the efficiency of subway operation， maintenance， and repair.