Table 2.
ResNet-Swish-BiLSTM structural description.
Layer Name | Activation | Learnable |
---|---|---|
Image Input | 224 × 224 × 3 | – |
Convolution | 112 × 112 × 64 | Weights 7x7x3x64 Bias 1 × 1 × 64 |
Batch Normalize | 112 × 112 × 64 | offset 1 × 1 × 64 Scale 1 × 1 × 64 |
Swish Activation | 112 × 112 × 64 | |
Multi Plan | 112 × 112 × 64 | |
Convolution | 56 × 56 × 64 | Weights 3 × 3 × 64 × 64 Bias 1 × 1 × 64 |
Batch Normalize | 56 × 56 × 64 | Offset 1 × 1 × 64 Scale 1 × 1 × 64 |
Convolution Block | 28 × 28 × 128 | Weights 3 × 3 × 128 × 128 Bias 1 × 1 × 128 |
Identity Block | 28 × 28 × 128 | |
Convolution Block | 14 × 14 × 256 | Weights 3 × 3 × 258 × 256 Bias 1 × 1 × 256 |
Identity Block | 14 × 14 × 256 | |
Convolution Block | 7 × 7 × 512 | Weights 3 × 3 × 128 × 512 Bias 1 × 1 × 512 |
Identity Block | 7 × 7 × 512 | |
Convolution Block | 3 × 3 × 1024 | Weights 3 × 3 × 1024 × 1024 Bias 1 × 1 × 1024 |
Identity Block | 1 × 1 × 1024 | |
Residual Block-1 | 1 × 1 × 1024 | |
AP (Average Pooling) | 1 × 1 × 1024 | |
Residual Block-2 | 3 × 3 × 1024 | |
Featureinput | 9216 | |
BiLSTM1 | 500 | inputWeights:500 × 9216, |
FC1 | 200 | Weights:200 × 500,Bias: 200 × 1 |
BiLSTM2 | 200 | inputWeights:2500 × 9216, Bias:2500 × 9216 |
Dropout 25% | 200 | – |
FC2 2 Fully Connected | 2 | Weights:2 × 200, Bias:2 × 1 |
SoftMax | 2 | – |
Classification | 2 | – |