Research on bolt contour extraction and counting of locomotive running gear based on deep learning

다국어 초록 (Multilingual Abstract)

The detection of abnormal running gear is essential to a locomotive’s daily maintenance, with the posture and quantity of various small bolts being important indicators to judge whether the locomotive is running safely. Traditional detection algorithms are easily affected by light changes, stain coverage, and image distortion, which is difficult to meet the detection requirements. Thus this paper proposes a deep learning based on bolt detection method that is appropriate for locomotive running gears. First, a bolt segmentation network was developed based on an improved U-netthat compensates the image information loss after multiple cross fusions involving the fusion of front and back convolution layer feature images. Furthermore, the proposed network utilizes the PReLU activation function and employs the concept structure to optimize the convolution method. This strategy aims to improve further the model’s segmentation accuracy and convergence speed. On this basis, we exploited several morphological transformations to improve the contour detection accuracy and ensure the bolt counting accuracy. The experimental results on the mainline running train data highlight that, compared with U-net, the proposed network’s recall rate and the mean intersection over union value are increased by 5.38 and 14.3, respectively. Furthermore, the bolt counting method’s loss function and mean absolute errors are significantly reduced compared with the contour extraction algorithm.

참고문헌 (Reference)

1 K. A. Philbrick, "What does deep learning see? Insights from a classifier trained to predict contrast enhancement phase from CT images" 211 (211): 1184-1193, 2018

2 Q. Ronneberger, "U-Net: Convolutional Networks for Biomedical Image Segmentation" Springer 2015

3 S. Suzuki, "Topological structural analysis of digitized binary images by border following" 30 (30): 32-46, 1985

4 Y. Y. Zhang, "Single-image crowd counting via multi-column convolutional neural network" 589-597, 2016

5 G. R. Wu, "Scalable high-performance image registration framework by unsupervised deep feature representations learning" 63 (63): 1505-1516, 2015

6 C. Szegedy, "Rethinking the inception architecture for computer vision" 2818-2826, 2016

7 C. Wu, "Research on Trouble Detection Method of Emu's Running Gear Based on Image Processing" Southwest Jiaotong University 2018

8 L. M. Xie, "Research of Key Components Detection Algorithm of Locomotive Running-gear Based on Implicit Shape Model" Southwest Jiaotong University 2016

9 Y. Xu, "Remote sensing image segmentation method based on deep learning model" 39 (39): 2905-2914, 2019

10 T. C. Huynh, "Quasiautonomous bolt-loosening detection method using visionbased deep learning and image processing" 105 : 102844-, 2019

11 J. Cho, "Medical image deep learning with hospital PACS dataset" 2015

12 C. Szegedy, "Inception-v4, Inception-resNet and the impact of residual connections on learning" 4278-4284, 2017

13 K. He, "Delving deep into rectifiers: surpassing human-level performance on imagenet classification" 1026-1034, 2015

14 K. He, "Deep residual learning for image recognition" 770-778, 2016

15 M. Lai, "Deep learning for medical image segmentation"

16 Sang-Yun Lee ; 이상권, "Deep convolutional neural network with new training method and transfer learning for structural fault classification of vehicle instrument panel structure" 대한기계학회 34 (34): 4489-4498, 2020

17 S. Y. Wang, "Deep Learning for Medical Image Analysis" Academic Press 245-269, 2017

18 Y. Yang, "Combining optimized UNet and residual learning for cell membrane segmentation" 40 (40): 3313-3318, 2019

19 H. C. Pham, "Bolt-loosening monitoring framework using an imagebased deep learning and graphical model" 20 (20): 3382-, 2020

20 X. F. Zhao, "Bolt loosening angle detection technology using deep learning" 26 (26): e2292-, 2019

21 S. Ioffe, "Batch normalization: accelerating deep network training by reducing internal covariate shift" PMLR 448-456, 2015

22 Q. Huang, "Automatic recognition of bolts on locomotive running gear based on laser scanner 3D measurement" 45 (45): 170532-1-170532-8, 2018

23 유현석 ; 배은경 ; 문영진 ; 권지훈 ; 최재순, "Automatic control of cardiac ablation catheter with deep reinforcement learning method" 대한기계학회 33 (33): 5415-5423, 2019

24 F. Q. Zhou, "Automated visual inspection of target parts for train safety based on deep learning" 12 (12): 550-555, 2018

25 D. Q. Tran, "Artificial intelligence-based bolt loosening diagnosis using deep learning algorithms for laser ultrasonic wave propagation data" 20 (20): 5329-, 2020

26 B. Niu, "An infrared image denoising method based on double-tree complex wavelet and morphology" 26 (26): 49-52, 2019

27 G. Litjens, "A survey on deep learning in medical image analysis" 42 (42): 60-88, 2017