X-ray image defect recognition method for pipe weld based on improved convolutional neural network
-
摘要: 针对卷积神经网络(CNN)应用于焊缝探伤图像识别时，目标区域占比小，局部信息冗余，激活函数小于零时出现硬饱和区导致模型对输入变化较敏感、网络参数难以训练的问题，采用超像素分割算法(SLIC)和改进的ELU激活函数构建CNN模型进行焊缝探伤图像缺陷识别. 首先，在CNN模型中使用ELU激活函数，在缓解梯度消失时对输入噪声产生更好的鲁棒性，同时，利用SLIC算法对图像像素进行像素块处理的特点，增大焊缝探伤图像中感兴趣区域的占比，降低局部冗余信息，提高模型在训练过程中的特征提取能力. 通过对焊缝探伤图像感兴趣区域提取并与所述CNN模型进行对比试验. 结果表明，该方法在焊缝探伤图像特征提取、训练耗时及识别准确率方面较传统卷积神经网络有更好的表现.Abstract: When convolution neural network (CNN) is applied to weld flaw detection image recognition, the target area is small, the local information is redundant, and the hard saturation region of activation function is less than zero, which makes the model sensitive to input change and difficult to train the network parameters. The super pixel segmentation algorithm (SLIC) and the improved ELU activation function are used to construct CNN model for weld flaw detection image defect recognition. First, the ELU activation function is used in the CNN model to generate better robustness to the input noise when the response gradient disappears, At the same time, the SLIC algorithm is used to deal with the pixels of the image, which increases the proportion of the region of interest in the weld flaw detection image, reduces the local redundant information, and improves the feature extraction ability of the model in the training process. Through the extraction of the region of interest of weld flaw detection image and the establishment of the CNN model described in this paper, the results show that the proposed method has better performance than the traditional convolution neural network in feature extraction, training time and recognition accuracy of weld flaw detection image.
-
Keywords:
- weld defect recognition /
- convolution neural network /
- SLIC algorithm /
- ELU function
-
-
![]()
图 6 迭代正确率及交叉熵损失
Figure 6. Iterative correct rate and cross entropy loss. (a) iterative accuracy rate; (b) iterative cross entropy loss
表 1 模型构建方式
Table 1 Model construction method
CNN模型名称 训练图像 激活函数 CNN-1 未进行SLIC处理 RELU CNN-2 未进行SLIC处理 ELU CNN-3 进行SLIC处理 ELU 
下载: 导出CSV
表 2 各模型迭代耗时
Table 2 Iteration time of each model
CNN模型名称 训练图像 激活函数 耗时减幅 CNN-1 未进行SLIC处理 RELU 0 CNN-2 未进行SLIC处理 ELU 1.07% CNN-3 进行SLIC处理 ELU 12.87%
下载: 导出CSV
表 3 焊缝探伤图像识别结果表
Table 3 Weld flaw detection image recognition result table
测试样本 CNN-1分类及识别结果 CNN-2分类及识别结果 CNN-3分类及识别结果 1号无缺陷 (0.997，0.002，0.000，0.000) (0.991，0.000，0.001，0.008) (0.999，0.000，0.000，0.000) 2号无缺陷 (0.999，0.000，0.000，0.000) (0.998，0.001，0.000，0.001) (0.999，0.000，0.000，0.000) 1号气孔 (0.000，0.000，0.000，0.999) (0.000，0.000，0.000，0.999) (0.000，0.000，0.000，1.000) 2号气孔 (0.000，0.000，0.000，0.999) (0.000，0.000，0.000，0.999) (0.000，0.000，0.000，0.999) 1号未熔合 (0.000，0.500，0.499，0.000) (0.000，0.043，0.956，0.001) (0.000，0.005，0.902，0.091) 2号未熔合 (0.000，0.138，0.797，0.064) (0.000，0.007，0.981，0.011) (0.000，0.001，0.549，0.448) 1号未焊透 (0.000，0.859，0.140，0.000) (0.000，0.936，0.063，0.000) (0.000，0.968，0.031，0.000) 2号未焊透 (0.000，0.957，0.042，0.000) (0.000，0.907，0.092，0.000) (0.400，0.598，0.001，0.000)
下载: 导出CSV
-
[1] 张志芬, 杨哲, 任文静, 等. 电弧光谱深度挖掘下的铝合金焊接过程状态检测[J]. 焊接学报, 2019, 40(1): 19 − 25. Zhang Zhifen, Yang Zhe, Ren Wenjing, et al. Condition detection in Al alloy welding process based on deep mining of arc spectrum[J]. Transactions of the China Welding Institution, 2019, 40(1): 19 − 25.
[2] Min Xiangjun. The development and design of the repair welding procedure of the thick wall duplex stainless steel piping[J]. China Welding, 2017, 26(1): 60 − 64.
[3] 孙怡, 孙洪雨, 白鹏, 等. X射线焊缝图像中缺陷的实时检测方法[J]. 焊接学报, 2004, 25(2): 115 − 118. doi: 10.3321/j.issn:0253-360X.2004.02.029 Sun Yi, Song Hongyu, Bai Peng, et al. Real-time automatic detection of weld defects in X-ray images[J]. Transactions of the China Welding Institution, 2004, 25(2): 115 − 118. doi: 10.3321/j.issn:0253-360X.2004.02.029
[4] Yang Zhenzhen, Kuang Nan, Fan Lu, et al. Review of image classification algorithms based on convolutional neural networks[J]. Journal of Signal Processing, 2018, 34(12): 84 − 99.
[5] 周飞燕, 金林鹏, 董军. 卷积神经网络综述[J]. 计算机学报, 2017, 40(6): 1229 − 1251. doi: 10.11897/SP.J.1016.2017.01229 Zhou Feiyan, Jin Linpeng, Dong Jun. Review of image classification algorithms based on convolutional neural networks[J]. Chinese Journal of Computers, 2017, 40(6): 1229 − 1251. doi: 10.11897/SP.J.1016.2017.01229
[6] Boureau Y, Roux N L, Bach F, et al. Ask the locals: multi-way local pooling for image recognition[C]// IEEE International Conference on Computer Vision. IEEE, 2011: 2651-2658.
[7] Wu X, He R, Sun Z, et al. A light CNN for deep face representation with noisy labels[J]. IEEE Transactions on Information Forensics and Security, 2018, 13(11): 1 − 11. doi: 10.1109/TIFS.2018.2879134
[8] Gu J, Wang Z, Kuen J, et al. Recent advances in convolutional neural networks[J]. Computer Science, 2018, 77: 354 − 377.
[9] Achanta R, Shaji A, Smith K, et al. SLIC superpixels comparedto state of the art superpixel methods[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2012, 34(11): 2274 − 2282.
[10] Singh R, Om H. Newborn face recognition using deep convolutional neural network[J]. Multimedia Tools & Applications, 2017, 76(18): 1 − 11.
[11] Mery D, Riffo V, Zscherpel U, et al. The database of X-ray images for nondestructive testing[J]. Journal of Nondestructive Evaluation, 2015, 34(4): 1 − 12.
-
期刊类型引用(9)
1. 高博轩，赵弘，苗兴园. 基于改进GWO-GRNN的管道焊缝三维重构测量. 机床与液压. 2024(01): 1-10 . 
百度学术
2. 陈晓明，王丽，马良，周峰，袁山山. 钢筋工程焊缝质量检测技术研究进展. 北京理工大学学报. 2024(12): 1215-1224 . 百度学术
3. 唐昌华，吴琼，时兵，陈坚，李洪亮. 基于压缩感知的激光复合成像超分辨三维重构方法. 激光杂志. 2023(09): 109-113 . 百度学术
4. 李诺薇，邹维科，种法力. 基于3D激光扫描传感器的焊缝区域跟踪方法设计. 激光与红外. 2023(11): 1650-1656 . 百度学术
5. 余志，白小亮. 一种石油管内焊缝余高及刮槽深度的测量方法. 石油工业技术监督. 2022(05): 17-19 . 百度学术
6. 杨国威，张金丽. 基于光栅投影的焊后焊缝表面三维测量. 焊接学报. 2022(04): 100-105+112+119-120 . 本站查看
7. 马涛，丁克勤，舒安庆，刘亚男. 基于布里渊分布式光纤的储液罐变形场重构方法研究. 压力容器. 2021(04): 11-19 . 百度学术
8. 胡曦，余震，刘海生. 管道全位置焊缝三维测量研究. 石油机械. 2021(09): 129-136 . 百度学术
9. 李亮亮，郑世伟，单清群，左玉达. 转向架焊接结构件深度缺陷超声相控阵检测及三维可视化. 焊接. 2021(09): 44-50+64 . 百度学术
其他类型引用(12)
计量
- 文章访问数: 702
- HTML全文浏览量: 25
- PDF下载量: 75
- 被引次数: 21
