小样本药用玻璃瓶缺陷生成方法及应用

Defect Generation Method and Application of Few Sample Medical Glass Bottles

中文关键词: 药包玻璃瓶  图像生成  StyleGAN2-ADA  缺陷检测  YOLOv7

英文关键词: medical glass bottles  image generation  StyleGAN2-ADA  defect detection  YOLOv7

中文摘要:

      目的 为了克服药包玻璃瓶缺陷样本不足带来的缺陷检测模型精度不准的问题，提出改进StyleGAN2-ADA的缺陷样本生成方法，提升模型鲁棒性。方法 首先，基于StyleGAN2-ADA算法，在无缺陷图像集上训练网络模型并作为骨干。其次，在骨干网络上添加缺陷感知残差块，生成缺陷掩码，在少量的缺陷图像数据集上训练网络模型操纵掩码区域的特征，模拟缺陷图像生成过程，合成缺陷图像。最后，采用YOLOv7检测网络验证该样本生成方法的效果。结果 实验结果表明，该方法在大量正常图像和少量缺陷图像基础上生成逼真且多样性的缺陷图像，应用该缺陷样本合成方法丰富数据集后，西林瓶缺陷检测平均准确率(mAP)达到97.3%，较原始数据集合和StyleGAN2-ADA算法分别提高了33.1%和4.1%。结论 该图像生成方法可以在少量缺陷样本下生成高质量的缺陷图像，优化不均衡数据集，增强模型训练的稳定性，提高药用玻璃包装产品的质量和合格率。

英文摘要:

      The work aims to propose an improved StyleGAN2-ADA defect sample generation method to enhance model robustness, so as to overcome the problem of inaccurate defect detection models due to insufficient defective sample data. First, a network model was trained as the backbone based on the StyleGAN2-ADA algorithm on a defect-free image dataset. Then, defect perception residual blocks were added to the backbone network to generate defect masks. The network model was trained on a dataset of few defective images to manipulate the features of the masked region and simulate the generation process defective images to synthesize defective images. Finally, YOLOv7 detection network was used to validate the effectiveness of this defect sample generation method. Experimental results showed that the proposed method could generate realistic and diverse defective images based on a large number of normal images and a small number of defective images. After enriching the dataset with this defect sample generation method, the average detection accuracy of penicillin vial defects (mAP) reached 97.3%, which was 33.1% higher than the original dataset and 4.1% higher than the StyleGAN2-ADA algorithm. In conclusion, this defect image generation method can generate high-quality defective images with few defective samples, optimize imbalanced datasets, enhance the stability of model training and improve the quality and pass rate of medicinal glass packaging products.