Research Article
A Novel Deep Convolutional Neural Network Based on ResNet-18 and Transfer Learning for Detection of Wood Knot Defects
Table 5
The evaluation index values of network.
| | Classes | Model | | | F1 | FAR |
| | Decayed knot | LeNet-5 | — | 0% | — | 3.11% | | AlexNet | 72% | 94.74% | 81.82% | 0.17% | | VGGNet-16 | 90.91% | 52.63% | 66.67% | 1.5% | | GoogLeNet | 65.22% | 78.95% | 71.43% | 0.68% | | MobileNet V2 | 100% | 68.42% | 81.25% | 1.00% | | ReSENet-18 | 100% | 94.74% | 97.30% | 0.17% |
| | Dry knot | LeNet-5 | 66.39% | 82.29% | 73.49% | 3.46% | | AlexNet | 96.81% | 94.79% | 95.79% | 0.97% | | VGGNet-16 | 90.48% | 98.96% | 94.53% | 0.20% | | GoogLeNet | 93.75% | 78.13% | 85.23% | 3.95% | | MobileNet V2 | 95.83% | 95.83% | 95.83% | 0.78% | | ReSENet-18 | 100% | 100% | 100% | 0% |
| | Edge knot | LeNet-5 | 91.49% | 94.51% | 92.98% | 0.97% | | AlexNet | 94.74% | 98.90% | 96.78% | 0.19% | | VGGNet-16 | 94.68% | 97.80% | 96.21% | 0.39% | | GoogLeNet | 97.67% | 92.31% | 94.92% | 1.33% | | MobileNet V2 | 96.70% | 96.70% | 96.7% | 0.58% | | ReSENet-18 | 100% | 100% | 100% | 0% |
| | Encased knot | LeNet-5 | 95.45% | 52.5% | 67.74% | 3.23% | | AlexNet | 100% | 90% | 94.74% | 0.70% | | VGGNet-16 | 100% | 90% | 94.74% | 0.70% | | GoogLeNet | 100% | 75% | 85.71% | 1.72% | | MobileNet V2 | 97.30% | 90% | 93.51% | 0.70% | | ReSENet-18 | 100% | 95% | 97.44% | 0.35% |
| | Horn knot | LeNet-5 | 83.33% | 51.02% | 63.29% | 4.13% | | AlexNet | 97.96% | 97.96% | 97.96% | 0.18% | | VGGNet-16 | 88.89% | 97.96% | 93.20% | 0.18% | | GoogLeNet | 87.27% | 97.96% | 92.31% | 0.18% | | MobileNet V2 | 88% | 89.80% | 88.89% | 0.89% | | ReSENet-18 | 100% | 100% | 100% | 0% |
| | Leaf knot | LeNet-5 | 70.89% | 84.85% | 77.24% | 1.88% | | AlexNet | 98.39% | 92.42% | 95.31% | 0.91% | | VGGNet-16 | 98.15% | 80.30% | 88.33% | 2.33% | | GoogLeNet | 98.25% | 84.85% | 91.06% | 1.81% | | MobileNet V2 | 90% | 95.45% | 92.64% | 0.55% | | ReSENet-18 | 97.01% | 98.48% | 97.74% | 0.18% |
| | Sound knot | LeNet-5 | 85.39% | 91.2% | 88.20% | 6.40% | | AlexNet | 96.8% | 96.8% | 96.8% | 2.22% | | VGGNet-16 | 95.33% | 98% | 96.65% | 1.41% | | GoogLeNet | 88.93% | 99.6% | 93.96% | 0.30% | | MobileNet V2 | 96.06% | 97.6% | 96.82% | 1.68% | | ReSENet-18 | 99.20% | 99.20% | 99.2% | 0.55% |
|
|
