河南工程学院学报（自然科学版）

为提高服装生产中织物瑕疵检测的速度和准确率，提出了一种结合改进深度生成对抗网络(DCGAN)与改进YOLOv5模型的服装生产瑕疵检测(IDPGP)方案。消融实验结果显示，完整IDPGP方案的检测准确率达95%,误检率和漏检率分别控制在3%和2%左右，处理速度约150 ms。在实际应用测试中，该方案对3 000个样本的平均分类准确率接近98%,优于CA-EfficientDet方案的88%。IDPGP方案有助于企业对服装生产质量进行控制，也为服装生产瑕疵检测技术的研发提供了参考。

To improve the speed and accuracy of fabric defect detection in clothing production, a defect detection scheme for IDPGP clothing production combining improved DCGAN and improved YOLOv5 is proposed. The results of the ablation experiment showed that the detection accuracy of the complete IDPGP scheme reached 95%, with false detection rate and missed detection rate controlled at 3% and 2%, respectively. And the processing speed was 150 ms. At the same time, in practical application testing, the average classification accuracy of this scheme for 3 000 samples is close to 98%, which is better than the 88% of the CA-EfficientDet scheme. The IDPGP scheme helps to control the quality of clothing production for enterprises and provides references for the research and development of defect detection technology in clothing production.