利用深度可变形轮廓模型进行结构体裂纹视觉检测

doi:10.16180/j.cnki.issn1007-7820.2023.09.006

摘要/Abstract

摘要：

目前针对裂纹检测的深度学习实例分割方法主要通过目标检测生成一个边界框进行逐个像素掩码分割,不仅影响结构体裂纹轮廓的检测效果,且伴随着复杂的后处理代价。针对这一问题,文中提出利用深度可变形轮廓算法模型Deep Snake对结构体裂纹进行识别检测。通过对结构体裂纹数据集进行数据增强以提高模型鲁棒性,同时运用迁移学习将大型图片数据集COCO上的预训练网络参数迁移学习到结构体裂纹分割模型中作为初始化。在自制的裂纹图像数据集上进行实验,结果表明训练后的模型能够正确识别的裂纹对象并同时完成多个裂纹目标的分割,在检测平均时间为0.12 s 的前提下AP50达到75.4%,与其他深度学习模型及边缘检测算法进行的比较结果也体现出Deep Snake算法的优越性。

关键词: 深度学习, 实例分割, 目标检测器, Deep Snake, 结构体裂纹, 数据增强, 迁移学习, 裂纹图像数据集

Abstract:

At present, the deep learning instance segmentation method for crack detection mainly generates a boundary box through target detection to segment pixel by pixel mask, which will affect the detection effect of structural crack contour, and is accompanied by complex post-processing cost. To solve this problem, this study proposes to use deep snake algorithm model of deep deformable contour to identify and detect structural cracks. The robustness of the model is improved by data enhancement of the structural crack data set. At the same time, the pre training network parameters on the large image data set coco are transferred to the structural crack segmentation model as initialization by transfer learning. The experimental results on the self-made crack image data set show that the trained model can correctly identify the crack object and complete the segmentation of multiple crack targets at the same time. On the premise of the average detection time of 0.12 s, the AP50 reaches 75.4%. The comparison between the proposed methool and other deep learning models and edge detection algorithms also reflect the advantages of Deep Snake algorithm.

Key words: deep learning, instance segmentation, object detector, Deep Snake, structural cracks, data augmentation, transfer learning, crack image data set

中图分类号:

TP391.41

赖彬,王森. 利用深度可变形轮廓模型进行结构体裂纹视觉检测[J]. 电子科技, 2023, 36(9): 35-40.

Bin ,WANG Sen. Visual Detection of Structural Cracks Using Depth Deformable Contour ModelLAI[J]. Electronic Science and Technology, 2023, 36(9): 35-40.

图/表 7

图1

图2

图3

图4

图5

表1

图6

参考文献 18

[1]	伍旭东, 王勇, 王瑛. 基于深度主动学习的路面裂纹检测[J]. 信息与电脑, 2021, 33(7):76-80.
	Wu Xudong, Wang Yong, Wang Ying. Pavement crack detection based on deep active learning[J]. Information and Computer, 2021, 33(7):76-80.
[2]	邓鑫, 刘桂华. 一种改进阈值分割的荧光磁粉裂纹提取算法[J]. 测控技术, 2020, 39(9):87-93.
	Deng Xin, Liu Guihua. An improved threshold segmentation algorithm for crack extraction of fluorescent magnetic particles[J]. Measurement and Control Technology, 2020, 39(9):87-93.
[3]	梁放, 宋学, 肖颖, 等. 基于边缘检测的含缺陷岩石裂纹扩展分析[J]. 安徽建筑, 2019, 26(12):127-128.
	Liang Fang, Song Xue, Xiao Ying, et al. Analysis of crack propagation in defective rock based on edge detection[J]. Anhui Architecture, 2019, 26(12):127-128.
[4]	崔洁. 基于随机森林的激光超声表面缺陷识别方法研究[D]. 太原: 中北大学, 2019:33-38.
	Cui Jie. Research on laser ultrasonic surface defect recognition method based on random forest[D]. Taiyuan: North Central University, 2019:33-38.
[5]	夏翔, 宋逸君, 俞传富, 等. 基于BEMD和SVM的路面裂缝识别方法[J]. 电子器件, 2020, 43(1):224-228.
	Xia Xiang, Song Yijun, Yu Chuanfu, et al. Pavement crack identification method based on BEMD and SVM[J]. Electronic Devices, 2020, 43(1):224-228.
[6]	Ye X W, Jin T, Chen P Y. Structural crack detection using deep learning-based fully convolutional networks[J]. Advances in Structural Engineering, 2019, 22(16):3412-3419. doi: 10.1177/1369433219836292
[7]	Lee D, Kim J, Lee D. Robust concrete crack detection using deep learning-based semantic segmentation[J]. International Journal of Aeronautical and Space Sciences, 2019, 20(1):287-299. doi: 10.1007/s42405-018-0120-5
[8]	张莹, 刘子龙, 万伟. 基于Faster R-CNN的无人机车辆目标检测[J]. 电子科技, 2021, 34(11):11-20.
	Zhang Ying, Liu Zilong, Wan Wei. UAV vehicle target detection based on Faster R-CNN[J]. Electronic Science and Technology, 2021, 34(11):11-20.
[9]	刘继丹, 孙吉, 仲天舒, 等. 改进SSD卷积神经网络在钢板边裂纹检测中的应用[J]. 冶金自动化, 2020, 44(4): 43-47.
	Liu Jidan, Sun Ji, Zhong Tianshu, et al. Application of improved SSD convolutional neural network in steel plate edge crack detection[J]. Metallurgical Automation, 2020, 44(4):43-47.
[10]	蔡逢煌, 张岳鑫, 黄捷. 基于YOLOv3与注意力机制的桥梁表面裂痕检测算法[J]. 模式识别与人工智能, 2020, 33(10):926-933. doi: 10.16451/j.cnki.issn1003-6059.202010007
	Cai Fenghuang, Zhang Yuexin, Huang Jie. Bridge surface crack detection algorithm based on YOLOv3 and attention mechanism[J]. Pattern Recognition and Artificial Intelligence, 2020, 33(10):926-933. doi: 10.16451/j.cnki.issn1003-6059.202010007
[11]	李太文, 范昕炜. 基于Faster R-CNN的道路裂缝识别[J]. 电子技术应用, 2020, 46(7):53-56.
	Li Taiwen, Fan Xinwei. Road crack recognition based on Faster R-CNN[J]. Application of Electronic Technology, 2020, 46(7):53-56.
[12]	王森, 伍星, 张印辉, 等. 基于深度学习的全卷积网络图像裂纹检测[J]. 计算机辅助设计与图形学学报, 2018, 30(5):859-867.
	Wang Sen, Wu Xing, Zhang Yinhui, et al. Full convolutional network image crack detection based on deep learning[J]. Journal of Computer Aided Design and Graphics, 2018, 30(5):859-867.
[13]	陈波, 张华, 汪双, 等. 基于全卷积神经网络的坝面裂纹检测方法研究[J]. 水力发电学报, 2020, 39(7):52-60.
	Chen Bo, Zhang Hua, Wang Shuang, et al. Research on dam surface crack detection method based on fully convolutional neural network[J]. Journal of Hydroelectric Power Generation, 2020, 39(7):52-60.
[14]	李画, 李明晶, 李凯, 等. Mask RCNN 模型在路面缺陷检测中的应用[J]. 科学技术创新, 2020(29):131-132.
	Li Hua, Li Mingjing, Li Kai, et al. Application of Mask RCNN model in pavement defect detection[J]. Science and Technology Innovation, 2020(29):131-132.
[15]	章世祥, 张汉成, 李西芝, 等. 基于机器视觉的路面裂缝病害多目标识别研究[J]. 公路交通科技, 2021, 38(3):30-39. doi: 10.3969/j.issn.1002-0268.2021.03.005
	Zhang Shixiang, Zhang Hancheng, Li Xizhi, et al. Research on multitarget recognition of pavement cracks and diseases based on machine vision[J]. Highway Traffic Science and Technology, 2021, 38(3):30-39. doi: 10.3969/j.issn.1002-0268.2021.03.005
[16]	Zhang Q Q, Zhu Z J, Bai Y Q, et al. Distracted driving detection based on the improved CenterNet with attention mechanism[J]. Multimedia Tools and Applications, 2022, 81(6):7993-8005. doi: 10.1007/s11042-022-12128-3
[17]	李豪, 袁广林, 李从利, 等. 基于2D循环卷积和难度敏感轮廓交并比损失的Deep Snake[J]. 模式识别与人工智能, 2021, 34(11):1004-1016. doi: 10.16451/j.cnki.issn1003-6059.202111004
	Li Hao, Yuan Guanglin, Li Congli, et al. Deep Snake based on 2D circular convolution and difficulty sensitive contour intersection loss[J]. Pattern Recognitionand Artificial Intelligence, 2021, 34(11):1004-1016. doi: 10.16451/j.cnki.issn1003-6059.202111004
[18]	Xu H Y, Xu X L, Zuo Y B. Applying morphology to improve Canny operator's image segmentation method[J]. The Journal of Engineering, 2019(23):8816-8819.

方法	AP/%	AP50/%	AP75/%	时间/s
Mask R-CNN	0.453	0.647	0.538	0.465
PANet	0.486	0.709	0.584	0.356
Deep Snake	0.526	0.754	0.612	0.121