基于改进RetinaNet的高铁无砟轨道板表面裂缝检测

doi:10.3778/j.issn.1002-8331.2110-0419

摘要/Abstract

摘要： 针对高铁无砟轨道板表面裂缝尺度差异大、裂缝类别不平衡等问题，提出了基于改进RetinaNet的裂缝检测方法。为了缓解下采样与特征金字塔横向连接压缩而导致的细微信息丢失的问题，利用多级特征金字塔融合ResNet-50主干网络中提取的不同层次的深浅特征，实现了图像特征信息的充分表达；为了解决检测过程中表面裂缝的分类和定位置信度之间不匹配的问题，提出自适应锚点学习使锚点与网络模型同时进行优化，提高了对小尺度裂缝的检测精度；为了缓解裂缝类别不平衡对检测性能的影响，引入焦点损失函数（Focal Loss）作为分类损失函数，并在其中添加类平衡权重项因子，提升了对小类别裂缝的检测精度。实验结果表明，改进RetinaNet检测网络对高铁无砟轨道板不同类别的裂缝均获得了较好的效果，平均检测精度（mAP）达到72.58%，较之原始RetinaNet检测网络提高了3.60个百分点，有效实现了对不同尺度裂缝的准确检测。

关键词: 目标检测, 高铁无砟轨道板, 裂缝检测, RetinaNet, 多级特征金字塔, 锚点, Focal Loss

Abstract: A crack detection method based on improved RetinaNet is proposed to address the problems of large difference between scales and unbalanced crack types in the surface of ballastless slab track of high-speed railway. To alleviate the problem of subtle information loss caused by downsampling and horizontal connection compression of feature pyramid, multi-level feature pyramid network is used to integrate different depth features extracted from ResNet-50 backbone network to achieve full expression of image feature information. To solve the problem of mismatching between the classification and positioning confidence of surface cracks in the detection process, adaptive anchor learning is proposed to optimize the anchor and the network model at the same time, which improves the detection accuracy of small-scale cracks. To alleviate the impact of crack category imbalance in detection performance, Focal Loss function is introduced as the classification loss function, and weight factor of class balance is added to improve the detection accuracy of small types of cracks. The experimental results show that the improved RetinaNet detection network achieves good results on different crack types in the ballastless slab track of high-speed railway, and the mean average precision（mAP） is 72.58%, which is 3.60 percentage points higher than that of the original RetinaNet detection network, effectively realizes the accurate detection of cracks of different scales.

Key words: object detection, ballastless slab track of high-speed railway, crack detection, RetinaNet, multi-level feature pyramid network（MLFPN）, anchor, Focal Loss

张诗慧, 罗晖, 裴莹玲, 余俊英, 徐杰. 基于改进RetinaNet的高铁无砟轨道板表面裂缝检测[J]. 计算机工程与应用, 2023, 59(6): 310-317.

ZHANG Shihui, LUO Hui, PEI Yingling, YU Junying, XU Jie. Surface Crack Detection in Ballastless Slab Track of High-Speed Railway Based on Improved RetinaNet[J]. Computer Engineering and Applications, 2023, 59(6): 310-317.

参考文献

[1] 王平，徐浩，陈嵘，等.路基上CRTS Ⅱ型板式轨道裂纹影响分析[J].西南交通大学学报，2012，47（6）：929-934.
WANG P，XU H，CHEN R，et al.Influence analysis of CRTS Ⅱ plate track crack on subgrade[J].Journal of Southwest Jiaotong University，2012，47（6）：929-934.
[2] 丁义南.基于车致振动响应的CRTS Ⅱ型板式无砟轨道损伤识别研究[D].石家庄：石家庄铁道大学，2018.
DING Y N.Damage identification of CRTS Ⅱ type slab ballastless track based on dynamic response induced by vehicle analysis[D].Shijiazhuang：Shijiazhuang Tiedao University，2018.
[3] 罗会兰，陈鸿坤.基于深度学习的目标检测研究综述[J].电子学报，2020，48（6）：1230-1239.
LUO H L，CHEN H K.Survey of object detection based on deep learning[J].Electronic Journals，2020，48（6）：1230-1239.
[4] REN S，HE K，GIRSHICK R，et al.Faster R-CNN：towards real-time object detection with region proposal networks[J].IEEE Transactions on Pattern Analysis & Machine Intelligence，2017，39（6）：1137-1149.
[5] DAI J，LI Y，HE K M，et al.R-FCN：object detection via region-based fully convolutional networks[C]//Advances in Neural Information Processing Systems，December 5-10，2016，Barcelona，Spain.Canada：NIPS，2016：379-387.
[6] REDMON J，DIVVALA S，GIRSHICK R，et al.You only look once：unified，real-time object detection[C]//Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition.Washington，DC：IEEE Computer Society，2016：779-788.
[7] LIU W，ANGUELOV D，ERHAN D，et al.SSD：single shot multi-box detector[C]//Proceeding of the 2016 European Conference on Computer Vision.Cham：Springer，2016：21-37.
[8] LIN T Y，GOYAL P，GIRSHICK R，et al.Focal loss for dense object detection[J].IEEE Transactions on Pattern Analysis and Machine Intelligence，2020，42（2）：318-327.
[9] 陈科圻，朱志亮，邓小明，等.多尺度目标检测的深度学习研究综述[J].软件学报，2021，32（4）：1201-1227.
CHEN K Q，ZHU Z L，DENG X M，et al.Deep learning for multi-scale object detection：a survey[J].Journal of Software，2021，32（4）：1201-1227.
[10] 李红光，于若男，丁文锐.基于深度学习的小目标检测研究进展[J].航空学报，2021，42（7）：107-125.
LI H G，YU R N，DING W R.Research development of small object traching based on deep learning[J].Journal of Aviation，2021，42（7）：107-125.
[11] KONG T，SUN F，HUANG W，et al.Deep feature pyramid reconfiguration for object detection[C]//European Conference on Computer Vision，2018：172-188.
[12] SUN F C，KONG T，HUANG W B，et al.Feature pyramid reconfiguration with consistent loss for object detection[J].IEEE Transactions on Image Processing，2019，28（10）：5041-5051.
[13] ZOU Q，ZHANG Z，LI Q，et al.DeepCrack：learning hierarchical convolutional features for crack detection[J].IEEE Transactions on Image Processing，2019，28（3）：1498-1512.
[14] WANG J，CHEN K，YANG S，et al.Region proposal by guided anchoring[C]//IEEE/CVF Conference on Computer Vision and Pattern Recognition（CVPR），Long Beach，CA，USA，2019.
[15] CHENG X，YU J B.RetinaNet with difference channel attention and adaptively spatial feature fusion for steel surface defect detection[J].IEEE Transactions on Instrumentation and Measurement，2021，70：1-11.
[16] TAN J R，WANG C B，LI B Y，et al.Equalization loss for long-tailed object recognition[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition（CVPR），2020：11659-11668.
[17] 崔子越，皮家甜，陈勇，等.结合改进VGGNet和Focal Loss的人脸表情识别[J].计算机工程与应用，2021，57（19）：171-178.
CUI Z Y，PI J T，CHEN Y，et al.Facial expression recognition combined with improved VGGNet and Focal Loss[J].Computer Engineering and Applications，2021，57（19）：171-178.
[18] ZHAO Q，SHENG T，WANG Y，et al.M2det：a single-shot object detector based on multi-level feature pyramid network[C]//Proceedings of the AAAI Conference on Artificial Intelligence，2019：9259-9266.
[19] OKSUZ K，CAM B，KALKAN S，et al.Imbalance problems in object detection：a review[J].IEEE Transactions on Pattern Analysis and Machine Intelligence，2020，43（10）：3388-3415.
[20] CUI Y，JIA M，LIN T，et al.Class-balanced loss based on effective number of samples[C]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition（CVPR），2019：9260-9269.
[21] ZOU Q，CAO Y，LI Q，et al.CrackTree：automatic crack detection from pavement images[J].Pattern Recognition Letters，2012，33（3）：227-238.