基于动态特征蒸馏的水工隧洞缺陷识别方法

doi:10.11772/j.issn.1001-9081.2020101596

计算机应用 ›› 2021, Vol. 41 ›› Issue (8): 2358-2365.DOI: 10.11772/j.issn.1001-9081.2020101596

所属专题：前沿与综合应用

基于动态特征蒸馏的水工隧洞缺陷识别方法

黄继爽^1,2, 张华^1,2,3, 李永龙^1,2,3, 赵皓^1,2, 王皓冉^3,4, 冯春成^1,2,3

1. 西南科技大学信息工程学院, 四川绵阳 621010;
2. 特殊环境机器人技术四川省重点实验室(西南科技大学), 四川绵阳 621010;
3. 清华四川能源互联网研究院, 成都 610213;
4. 水沙科学与水利水电工程国家重点实验室(清华大学), 北京 100084

收稿日期:2020-10-14 修回日期:2020-12-07 出版日期:2021-08-10 发布日期:2021-01-27
通讯作者: 黄继爽
作者简介:黄继爽(1993-),男,四川绵阳人,硕士研究生,主要研究方向:深度学习、模型压缩;张华(1969-),男,四川绵阳人,教授,博士,主要研究方向:辐射环境机器人、模式识别、智能系统;李永龙(1983-),男,山西太原人,高级工程师,博士研究生,主要研究方向:模式识别、智能系统;赵皓(1991-),男,四川南充人,博士研究生,主要研究方向:模式识别、智能系统;王皓冉(1988-),男,河南南阳人,助理研究员,博士,主要研究方向:库坝安全、智能巡检;冯春成(1992-),男,四川达州人,博士研究生,主要研究方向:模式识别、智能系统。
基金资助:
国家重点研发计划项目（2019YFB1310503）；国家“十三五”核能开发科研项目（20161295）；四川省科技创新创业苗子工程（2020JDRC0130）；中国大唐集团有限公司科学技术项目（CDT-TZK/SYD［2018］-010）。

Hydraulic tunnel defect recognition method based on dynamic feature distillation

HUANG Jishuang^1,2, ZHANG Hua^1,2,3, LI Yonglong^1,2,3, ZHAO Hao^1,2, WANG Haoran^3,4, FENG Chuncheng^1,2,3

1. School of Information Engineering, Southwest University of Science and Technology, Mianyang Sichuan 621010, China;
2. Robot Technology Used for Special Environment Key Laboratory of Sichuan Province(Southwest University of Science and Technology), Mianyang Sichuan 621010, China;
3. Sichuan Energy Internet Research Institute under Tsinghua University, Chengdu Sichuan 610213, China;
4. State Key Laboratory of Hydroscience and Engineering(Tsinghua University), Beijing 100084, China

Received:2020-10-14 Revised:2020-12-07 Online:2021-08-10 Published:2021-01-27
Supported by:
This work is partially supported by the National Key Research and Development Program of China (2019YFB1310503), the Nuclear Energy Development and Scientific Research Project of the National "13th Five-Year Plan" (20161295), the Sichuan Science and Technology Innovation and Entrepreneurship Seedling Project (2020JDRC0130), the Science and Technology Project of China Datang Corporation Limited (CDT-TZK/SYD[2018]-010).

摘要/Abstract

摘要： 针对水工隧洞缺陷识别任务中现有深度卷积神经网络（DCNN）对缺陷图像特征提取能力不足、识别种类少、推理耗时长的问题，提出一种基于动态特征蒸馏的缺陷自主识别方法。首先，通过深度曲线估计网络对图像进行优化，从而改善低照度环境下的图像质量；其次，构建加入注意力机制的动态卷积模块取代传统静态卷积，并且把得到的动态特征用于训练教师网络以获得更好的模型特征提取能力；最后，在知识蒸馏框架中融合鉴别器结构，以构造一种动态特征蒸馏损失，并通过鉴别器将动态特征知识从教师网络传递到学生网络，从而在大幅减少模型推理时间的同时实现六类缺陷的高精度识别。在四川某水电站水工隧洞缺陷数据集上对该方法和原有残差网络进行对比实验，结果表明该方法可达到96.15%的识别准确率，其模型参数量和推理时间分别降低到原来的1/2和1/6。通过实验结果可知，将缺陷图像的动态特征蒸馏信息融合到识别网络中能够提高水工隧洞缺陷的识别效率。

关键词: 水工隧洞, 缺陷识别, 动态卷积, 知识蒸馏, 模型压缩

Abstract: Aiming at the problems that the existing Deep Convolutional Neural Network (DCNN) have insufficient defect image feature extraction ability, few recognition types and long reasoning time in hydraulic tunnel defect recognition tasks, an autonomous defect recognition method based on dynamic feature distillation was proposed. Firstly, the deep curve estimation network was used to optimize the image to improve the image quality in low illumination environment. Secondly, the dynamic convolution module with attention mechanism was constructed to replace the traditional static convolution, and the obtained dynamic features were used to train the teacher network to obtain better model feature extraction ability. Finally, a dynamic feature distillation loss was constructed by fusing the discriminator structure in the knowledge distillation framework, and the dynamic feature knowledge was transferred from the teacher network to the student network through the discriminator, so as to achieve the high-precision recognition of six types of defects while significantly reducing the model reasoning time. In the experiments, the proposed method was compared with the original residual network on a hydraulic tunnel defect dataset of a hydropower station in Sichuan Province. The results show that this method has the recognition accuracy reached 96.15%, and the model parameter amount and reasoning time reduced to 1/2 and 1/6 of the original ones respectively. It can be seen from the experimental results that fusing the dynamic feature distillation information of the defect image into the recognition network can improve the efficiency of hydraulic tunnel defect recognition.

Key words: hydraulic tunnel, defect recognition, dynamic convolution, knowledge distillation, model compression

中图分类号:

TP391.41

黄继爽, 张华, 李永龙, 赵皓, 王皓冉, 冯春成. 基于动态特征蒸馏的水工隧洞缺陷识别方法[J]. 计算机应用, 2021, 41(8): 2358-2365.

HUANG Jishuang, ZHANG Hua, LI Yonglong, ZHAO Hao, WANG Haoran, FENG Chuncheng. Hydraulic tunnel defect recognition method based on dynamic feature distillation[J]. Journal of Computer Applications, 2021, 41(8): 2358-2365.

参考文献

[1] 陈益民, 周垂一. 中国水利水电地下工程数据统计(截至2016年底)[J]. 隧道建设,2017,37(6):778-779.(CHEN Y M,ZHOU C Y. China water conservancy and hydropower underground engineering data statistics (to the end of 2016)[J]. Tunnel Construction,2017,37(6):778-779.)
[2] 刘志宽. 输水隧洞检测和安全评价[D]. 大连:大连理工大学, 2017:9-11.(LIU Z K. Detection and safety assessment of water conveyance tunnel[D]. Dalian:Dalian University of Technology, 2017:9-11.)
[3] 姚学练, 贺福强, 平安, 等. 基于HSI颜色空间与灰度波动相结合的复杂桥梁蜂窝麻面的图像分割[J]. 计算机应用,2019,39(3):882-887. (YAO X L,HE F Q,PING A,et al. Image segmentation for complex voids and pits of bridge based on combination of HSI color space and gray fluctuation[J]. Journal of Computer Applications,2019,39(3):882-887.)
[4] 王永会, 陈荣. 基于分数阶傅里叶变换和频谱增强的路面裂缝图像识别方法[J]. 计算机应用,2020,40(S1):189-194. (WANG Y H,CHEN R. Pavement crack image recognition method based on fractional Fourier transform and spectral enhancement[J]. Journal of Computer Applications,2020,40(S1):189-194.)
[5] 张建云, 刘九夫, 金君良. 关于智慧水利的认识与思考[J]. 水利水运工程学报,2019(6):1-7.(ZHANG J Y,LIU J F,JIN J L. Understanding and thinking of smart water conservancy[J]. HydroScience and Engineering,2019(6):1-7.)
[6] 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 S, WU X, ZHANG Y H, et al. Image crack detection with fully convolutional network based on deep learning[J]. Journal of Computer-Aided Design and Computer Graphics, 2018,30(5):859-867.)
[7] 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 B, ZHANG H,WANG S,et al. Study on detection method of dam surface cracks based on full convolution neural network[J]. Journal of Hydroelectric Engineering,2020,39(7):52-60.)
[8] 梁雪慧, 程云泽, 张瑞杰, 等. 基于卷积神经网络的桥梁裂缝识别和测量方法[J]. 计算机应用,2020,40(4):1056-1061. (LIANG X H,CHENG Y Z,ZHANG R J,et al. Bridge crack classification and measurement method based on deep convolutional neural network[J]. Journal of Computer Applications,2020,40(4):1056-1061.)
[9] DUNG C V,ANH L D. Autonomous concrete crack detection using deep fully convolutional neural network[J]. Automation in Construction,2019,99:52-58.
[10] FENG C C,ZHANG H,WANG H R,et al. Automatic pixel-level crack detection on dam surface using deep convolutional network[J]. Sensors,2020,20(7):No. 2069.
[11] HE K M,ZHANG X Y,REN S Q,et al. Deep residual learning for image recognition[C]//Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE,2016:770-778.
[12] HU J,SHEN L,SUN G. Squeeze-and-excitation networks[C]//Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE,2018:7132-7141.
[13] GULRAJANI I,AHMED F,ARJOVSKY M,et al. Improved training of Wasserstein GANs[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems. Red Hook,NY:Curran Associates Inc.,2017:5767-5779.
[14] GUO C L,LI C Y,GUO J C,et al. Zero-reference deep curve estimation for low-light image enhancement[C]//Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE,2020:1777-1786.
[15] YANG B,BENDER G,LE Q V,et al. CondConv:conditionally parameterized convolutions for efficient inference[C]//Proceedings of the 2019 Annual Conference on Neural Information Processing Systems. La Jolla, CA:NIPS Foundation, 2019:1305-1316.
[16] HINTON G,VINYALS O,DEAN J. Distilling the knowledge in a neural network[EB/OL]. (2015-03-09)[2020-06-10]. https://arxiv.org/pdf/1503.02531.pdf.
[17] WANG H R,WANG S,FENG C C,et al. Diversion tunnel defects inspection and identification using an automated robotic system[C]//Proceedings of the 2019 Chinese Automation Congress. Piscataway:IEEE,2019:5863-5868.
[18] 周志华. 机器学习[M]. 北京:清华大学出版社,2016:23-33. (ZHOU Z H. Machine Learning[M]. Beijing:Tsinghua University Press,2016:23-33.)
[19] MOLCHANOV P, TYREE S, KARRAS T, et al. Pruning convolutional neural networks for resource efficient inference[EB/OL]. (2017-06-08)[2020-06-10]. https://arxiv.org/pdf/1611.06440.pdf.
[20] ZHANG Y K,ZHANG J,WANG Q,et al. DyNet:dynamic convolution for accelerating convolutional neural networks[EB/OL]. (2020-04-22)[2020-06-10]. https://arxiv.org/pdf/2004.10694.pdf.
[21] YUAN L, TAY F E, LI G L, et al. Revisiting knowledge distillation via label smoothing regularization[C]//Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE,2020:3902-3910.
[22] AHN S,HU S X,DAMIANOU A,et al. Variational information distillation for knowledge transfer[C]//Proceedings of the 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway:IEEE,2019:9155-9163.
[23] CHATTOPADHYAY A,SARKAR A,HOWLADER P,et al. Grad-CAM++:generalized gradient-based visual explanations for deep convolutional networks[C]//Proceedings of the 2018 IEEE Winter Conference on Applications of Computer Vision. Piscataway:IEEE,2018:839-847.

基于动态特征蒸馏的水工隧洞缺陷识别方法

Hydraulic tunnel defect recognition method based on dynamic feature distillation

PDF

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 12

编辑推荐

Metrics

[1]	张成, 万源, 强浩鹏. 基于知识蒸馏的深度无监督离散跨模态哈希[J]. 计算机应用, 2021, 41(9): 2523-2531.
[2]	张明明, 卢庆宁, 李文中, 宋浒. 基于联合动态剪枝的深度神经网络压缩算法[J]. 计算机应用, 2021, 41(6): 1589-1596.
[3]	张文烨, 尚方信, 郭浩. 基于Octave卷积的混合精度神经网络量化方法[J]. 计算机应用, 2021, 41(5): 1299-1304.
[4]	魏淳武, 赵涓涓, 唐笑先, 强彦. 基于多时期蒸馏网络的随访数据知识提取方法[J]. 计算机应用, 2021, 41(10): 2871-2878.
[5]	陈程军, 毛莺池, 王绎超. 基于激活-熵的分层迭代剪枝策略的CNN模型压缩[J]. 计算机应用, 2020, 40(5): 1260-1265.
[6]	谢斌红, 钟日新, 潘理虎, 张英俊. 结合剪枝与流合并的卷积神经网络加速压缩方法[J]. 计算机应用, 2020, 40(3): 621-625.
[7]	雷小康, 尹志刚, 赵瑞莲. 基于FPGA的卷积神经网络定点加速[J]. 计算机应用, 2020, 40(10): 2811-2816.
[8]	朱倩倩, 刘渊, 李甫. 深度神经网络的仿生矩阵约简与量化方法[J]. 计算机应用, 2020, 40(10): 2817-2821.
[9]	邓棋, 雷印杰, 田锋. 用于肺炎图像分类的优化卷积神经网络方法[J]. 计算机应用, 2020, 40(1): 71-76.
[10]	冀树伟, 杨喜旺, 黄晋英, 尹宁. 基于特征复用的卷积神经网络模型压缩方法[J]. 计算机应用, 2019, 39(6): 1607-1613.
[11]	高钦泉, 赵岩, 李根, 童同. 基于知识蒸馏的超分辨率卷积神经网络压缩方法[J]. 计算机应用, 2019, 39(10): 2802-2808.
[12]	陈世利黄玉秋张慧杨晓霞郭薇. 分形理论在近表面缺陷超声A扫检测中的应用[J]. 计算机应用, 2014, 34(11): 3365-3368.