一种基于pix2pix改进的工业缺陷数据增强方法

An improved industrial defect data augmentation method based on pix2pix

基于深度学习的目标检测算法在工业检测中应用广泛,为解决工业缺陷数据不足的问题,提出了一种基于pix2pix改进的缺陷数据增强方法。从加强生成器和判别器对图像中缺陷区域的注意力出发,针对pix2pix进行了如下改进:(1)仅将整幅图像的缺陷区域作为判别器的输入,以此提升生成器对缺陷区域的注意力,同时,判别器采用了更小的卷积核提取缺陷区域的特征;(2)仅将图像中所有缺陷区域的平均生成对抗损失作为该图像的生成对抗损失,使网络更加关注缺陷区域的特征学习。在工业LED缺陷数据集上的实验结果表明,本方法生成的缺陷具有更逼真的视觉效果和更低的FID指数,同时有效提升了基于RetinaNet算法的缺陷检测精度。

The object detection method based on deep learning is widely used in industrial inspection. In order to solve the problem of insufficient industrial defect data, an improved defect data augmentation algorithm based on pip2 pix is proposed. Starting from the enhancement of the generator and discriminator’s attention to the defect area in the image, the following improvements have been made to pix2 pix:(1)Only the defect area of the entire image is used as the input of the discriminator to enhance the generator’s attention to the defect area. At the same time, the discriminator uses a smaller convolution kernel to extract the characteristics of the defect area.(2)Only the average generation confrontation loss of all defect regions in the image is used as the generation confrontation loss of the image, so that the network pays more attention to the defects regional feature learning. The experimental results on the industrial LED defect dataset show that the defects generated by the proposed method have more realistic visual effects, lower FID, and effectively improve the accuracy of defect detection based on the RetinaNet algorithm.