Table 3 Comparison of the performance of the CNN-based model with the ESG-ConvNeXt model on the X-SDD dataset.
From: An ESG-ConvNeXt network for steel surface defect classification based on hybrid attention mechanism
Model | Acc (%) | Pre (%) | Rec (%) | F1Score (%) | FPS (f/s) |
|---|---|---|---|---|---|
MobileNetV3 | 64.2 | 59.5 | 58.1 | 58.0 | 35.0 |
ConvNeXtV2 | 75.8 | 74.5 | 71.9 | 72.4 | 42.0 |
InceptionV3 | 76.8 | 76.7 | 72.8 | 73.4 | 15.9 |
VIT | 77.1 | 77.2 | 73.4 | 74.1 | 19.2 |
Swin-T | 82.5 | 80.0 | 79.8 | 79.4 | 3.8 |
DenseNet121 | 86.9 | 87.3 | 84.6 | 85.1 | 10.2 |
ESG-ConvNeXt | 90.5 | 90.7 | 89.5 | 89.7 | 6.0 |
