Research Article
A Novel Deep Convolutional Neural Network Based on ResNet-18 and Transfer Learning for Detection of Wood Knot Defects
Table 6
Accuracy of different models.
| | Different models | Accuracy | Different models | Accuracy |
| | Add a logsoftmax classifier and a NLLLoss function | 14.08% | Add a ReLU and a fully connected layer | 14.08% |
| | Add a softmax classifier | 14.40% | Without ReLU | 84.78% |
| | Without batch normal | 96.07% | Add a convolutional layer, a BN layer, and a ReLU function | 96.24% |
| | Without squeeze-and-excitation basic-block module | 98.53% | ReSENet-18 (our method) | 99.02% |
|
|
