基于改进Faster R-CNN算法的扣件缺陷检测

Fastener Defect Detection Based on Improved Faster R-CNN Algorithm

快速、高效检测高速铁路扣件缺失与缺陷情况,对保障铁路轨道线路维护和运营安全意义重大。采用目标检测技术提出一种基于改进Faster R-CNN的高速铁路扣件缺陷检测与提取算法,相较于人工巡检与静态检测能显著提升检测速度和精度。首先,使用ResNet-101网络代替VGG-16网络,结合FPN提取图像特征并进行特征图融合。然后,采用ROI Align替代ROI Pooling避免两次量化,消除了Faster R-CNN模型自身的边界框偏差。最后,采用SoftNMS算法代替传统NMS算法抑制非极大值,提升检测效率。实验表明,该方法在提取、检测铁路扣件缺陷的精度方面,相较于传统Faster R-CNN方法提升了10.58%,具有一定的实用价值。

Fast and efficient detection of missing and defective fasteners on high-speed railways is of great significance to ensure the maintenance and operational safety of railway track lines.A high-speed railway fastener defect detection and extraction algorithm based on improved Faster R-CNN is proposed using object detection technology,which can significantly improve detection speed and accuracy compared to manual inspection and static detection.Firstly,using ResNet-101 network instead of VGG-16 network,combined with FPN to extract image features and perform feature map fusion.Then,ROI Align was used instead of ROI Pooling to avoid double quantization and eliminate the bounding box bias of the Faster R-CNN model itself.Finally,the Soft-NMS algorithm is used to replace the traditional NMS algorithm to suppress non maximum values and improve detection efficiency.The experiment shows that this method improves the accuracy of extracting and detecting railway fastener defects by 10.58%compared to the traditional Faster R-CNN method,and has certain practical value.