In this paper, a method for measuring crack propagation length in the structural fatigue process is presented. A convolutional neural network is presented to eliminate noise interference of matching marks and recognize crack features. The initial region of the crack band is obtained through the convolution neural network result. Based on the initial region, an improved algorithm for tip recognition is proposed to calculate the exact position coordinates for crack tips. Finally, according to the position coordinates, the information of crack length is obtained. By increasing the number of cameras, cracks in different directions and locations can be detected simultaneously. The relationship between crack propagation length and the loading values, such as force and fatigue cycles, can be obtained by the proposed method. Meanwhile, compared with measuring means of the electromagnetic vortex, the effectiveness and accuracy of the proposed method are validated.

中文翻译：

---

结构疲劳过程中裂纹扩展检测方法

本文提出了一种测量结构疲劳过程中裂纹扩展长度的方法。提出了一种卷积神经网络来消除匹配标记的噪声干扰并识别裂纹特征。裂纹带的初始区域是通过卷积神经网络的结果得到的。基于初始区域，提出了一种改进的尖端识别算法来计算裂纹尖端的精确位置坐标。最后，根据位置坐标，得到裂纹长度信息。通过增加摄像头的数量，可以同时检测不同方向和位置的裂缝。裂纹扩展长度与载荷值（如力和疲劳循环）之间的关系可以通过所提出的方法获得。同时，