TABLE 1.
Structures | 1D terahertz absorption spectrums | 1D near-infrared hyperspectral spectrums | 2D terahertz absorption image | 2D near-infrared hyperspectral image |
F | 128 | 128 | 64 | 128 |
F2 | 64 | 64 | 32 | 64 |
R1 | 0.01 | 0.01 | 0.05 | 0.05 |
R2 | 0.01 | 0.01 | 0.05 | 0.05 |
KS1 | 2 × 1 | 2 × 1 | 2 × 2 | 2 × 2 |
KS2 | 2 × 1 | 2 × 1 | 2 × 2 | 2 × 2 |
AP | 2 × 1 | 2 × 1 | 2 × 2 | 2 × 2 |
D1 | 256 | 256 | 512 | 64 |
D2 | 3 | 3 | 3 | 3 |
T | 200 | 200 | 120 | 120 |
F1 and F2 mean the number of filters in the first and second convolutional layer; R1 and R2 mean the L2 regularization number of the first and second convolutional later; KS1 and KS2 mean the size of filters in the first and second convolutional layer; AP means the size of filter in the average pooling layer; D1 and D2 mean the number of neurons in the first and second fully connected layer; T means the number of training epoch.