基于深度学习的磁芯表面缺陷检测研究
2023,31(3):36-42
摘要：在产品表面缺陷智能检测过程中,存在缺陷样本收集困难、样本不平衡、目标尺寸小和难以定位等问题。针对磁芯表面缺陷检测中存在的问题进行了研究,提出了一种基于深度学习的图像增强和检测方法,首先利用结合高斯混合模型的深度卷积生成对抗网络生成磁芯缺陷图像,然后结合泊松融合方法产生增强的数据集,最后基于YOLO-v3网络,实现了磁芯表面缺陷的智能检测。实验表明,该方法能够生成质量更高、缺陷更明显的图像,检测准确度提升了5.6%。
关键词：磁芯;缺陷检测;深度卷积生成对抗网络;图像融合;深度学习
Research on Magnetic Core Surface Defect Detection Based on Deep Learning
Abstract：In the process of intelligent detection of surface defects, there would be several problems in practical applications, such as it is difficulty to collect magnetic core samples with defects, small defect target and the disequilibrium on defective samples and the hardness when locating defects. This article studies the existing problems that appear when detecting the surface defects on magnetic core,an image enhancement and detection method which based on deep learning is proposed. Primarily, the images of magnetic core defects are generated by deep convolutional generative adversarial network, and then the Poisson Blending method is used to produce the enhanced data set. Finally, the intelligent defect detection is achieved grounded on YOLO-v3 network. The experimental results indicate that the proposed method can yield images with higher quality and more clear defects, and the accuracy of defect detection is enhanced by 5.6%.
Key words：Magnetic core; defect detection; deep convolution generative adversarial networks; image fusion; deep learning
收稿日期：2022-07-20
基金项目：浙江省科技计划项目(NO.2019C011123),浙江省基础公益研究计划项目(NO.LGG19F030011)
