基于异常特征对抗学习的工业图像异常检测方法
|
|
王天飞,周文俊,项圣,贺宇航,彭博
|
Industrial image anomaly detection method based on adversarial learning of abnormal features
|
|
Tianfei WANG,Wenjun ZHOU,Sheng XIANG,Yuhang HE,Bo PENG
|
|
| 表 2 不同方法图像级/像素级AUROC结果对比 |
| Tab.2 Image- and pixel-level AUROC comparison results for different models |
|
| 类别 | AUROC | | US | AE-SSIM | RIAD | PaDim | CutPaste | CLGAN | MB-PFM | ATSNM | 本研究算法 | | 1）注：斜线前、后数据分别表示图像级以及像素级AUROC结果 | | bottle | 99.0/97.81） | 88.0/93.0 | 99.9/98.4 | 99.9/98.3 | 98.2/97.6 | 97.6/92.6 | 100.0/98.4 | 100.0/98.3 | 96.8/98.5 | | capsule | 86.1/96.8 | 61.0/94.0 | 88.4/92.8 | 91.3/98.5 | 98.2/97.4 | 98.2/98.4 | 94.5/94.3 | 93.7/98.5 | 96.2/97.3 | | grid | 81.0/89.9 | 69.0/94.0 | 99.6/98.8 | 96.7/97.3 | 100.0/97.5 | 99.3/98.7 | 98.0/98.8 | 95.2/98.7 | 100.0/99.6 | | leather | 88.2/97.8 | 46.0/78.0 | 100.0/99.4 | 100.0/99.2 | 100.0/99.5 | 100.0/99.7 | 100.0/96.4 | 100.0/99.5 | 100.0/99.9 | | pill | 87.9/96.5 | 60.0/91.0 | 83.8/95.7 | 93.3/95.7 | 94.9/95.7 | 98.1/97.3 | 96.5/95.2 | 93.7/96.5 | 98.2/96.5 | | tile | 99.1/92.5 | 52.0/59..0 | 98.7/89.1 | 98.1/98.1 | 94.6/90.5 | 96.5/94.1 | 99.6/96.2 | 95.9/97.9 | 100.0/98.6 | | transistor | 81.8/97.8 | 52.0/90.0 | 90.9/87.7 | 97.4/97.5 | 96.1/93.0 | 96.4/93.3 | 97.8/97.8 | 91.6/87.5 | 95.0/92.1 | | zipper | 91.9/95.6 | 80.0/88.0 | 98.1/97.8 | 90.3/98.5 | 99.9/99.3 | 99.3/97.8 | 97.4/98.2 | 96.3/98.5 | 100.0/99.1 | | cable | 86.2/91.9 | 61.0/82.0 | 81.9/84.2 | 92.7/96.7 | 81.2/90.0 | 98.3/95.6 | 98.8/96.7 | 91.3/96.8 | 93.2/96.7 | | carpet | 91.6/93.5 | 67.0/87.0 | 84.2/96.3 | 99.8/99.1 | 93.9/98.3 | 98.2/97.8 | 100.0/99.2 | 97.8/98.3 | 96.5/99.2 | | hazelnut | 93.1/98.2 | 54.0/97.0 | 83.3/96.1 | 92.0/98.2 | 98.3/97.3 | 99.0/98.1 | 100.0/99.1 | 99.8/98.4 | 100.0/98.8 | | metalnut | 82.0/97.2 | 54.0/89.0 | 88.5/92.5 | 98.7/97.2 | 99.9/93.1 | 97.9/96.8 | 100.0/97.2 | 98.6/96.7 | 98.9/97.8 | | screw | 54.9/97.4 | 51.0/92.0 | 84.5/98.8 | 85.8/98.5 | 88.7/96.7 | 95.2/94.9 | 91.8/97.7 | 92.1/98.9 | 96.6/98.9 | | toothbrush | 95.3/97.9 | 74.0/96.0 | 100.0/98.9 | 96.1/98.8 | 99.4/98.1 | 98.2/96.6 | 88.6/98.6 | 91.4/98.9 | 100.0/97.9 | | wood | 97.7/92.1 | 83.0/73.0 | 93.0/85.8 | 99.2/94.9 | 99.1/95.5 | 98.9/96.9 | 99.5/95.6 | 98.8/96.9 | 97.6/94.5 | | 平均 | 89.7/95.3 | 63.4/88.0 | 91.3/94.2 | 95.3/97.5 | 96.1/96.0 | 97.5/95.1 | 97.5/97.3 | 95.7/97.4 | 98.2/97.8 |
|
|
|