摘要
在荧光磁粉缺陷检测中,为快速有效地对金属轴上的点状、线型以及摩擦型缺陷进行分类检测,引入了深度学习技术,并与图像处理技术结合设计了一种改进型金属轴表面缺陷检测系统,克服了传统识别方式人工选定处理区域的局限性。利用基于YOLOv3算法的神经网络模型,对CCD相机获取的轴表面图像数据集进行训练和测试,对不同缺陷进行精确目标识别;采用图像处理技术对识别的目标进行缺陷定量分析。实验结果表明:该方法对不同缺陷类型能进行有效识别,在检测精度与检测效率上具有较高的提升。
In this paper, the technology of deep learning is introduced to quickly and efficiently classify the dot, line and friction type defect on the metal shaft for the fluorescent magnetic powder flaw detection. A kind of advanced metal shaft surface defect detection system is designed by combining the deep learning technology with image processing technology, which overcomes the limitations of traditional way of artificial choosing processing area. The neural network model based on YOLOv3 algorithm is used to train and test the data set of axis surface images obtained by CCD camera, and accurate target recognition is carried out for different defects. The image processing technology is used to analyze the defects quantitatively. Experimental results show that this method can effectively identify different defect types, and has a high improvement in detection accuracy and detection efficiency.
作者
刘硕
卜雄洙
谷世举
LIU Shuo;BU Xiongzhu;GU Shiju(School of Mechanical Engineering,Nanjing University of Technology,Nanjing 210094,China)
出处
《仪表技术》
2021年第3期49-53,共5页
Instrumentation Technology
关键词
荧光磁粉缺陷检测
YOLOv3算法
目标识别
图像处理
fluorescence magnetic particle defect detection
YOLOv3 algorithm
target recognition
image processing
