Research Article
Discrimination of Fresh Tobacco Leaves with Different Maturity Levels by Near-Infrared (NIR) Spectroscopy and Deep Learning
Table 4
Parameter settings of the convolutional neural networks for upper, middle, and lower leaves data sets.
| | Layers | Model parameters | Output shape |
| | Input layer | NIRS data of 454 × 1 dimension | | | Conv1D (C1) | 128 convolutional kernels of the size 13 × 1, the Relu function and BN mechanism, stride = 1 | 450 × 128 | | MaxPooling1D (S2) | Maxpooling, pooling size = 2 × 1, stride = 1 | 225 × 128 | | Conv1D (C3) | 64 convolutional kernels of the size 13 × 1, the Relu function and BN mechanism, stride = 1 | 221 × 64 | | MaxPooling1D (S4) | Maxpooling, pooling size = 1 × 1, stride = 1 | 221 × 64 | | Flatten (F5) | Flatten the feature vector of the S4 layer into 1 vector | 14144 × 1 | | Dense (F6) | 100 output neurons fully connected to all neurons in layer F5, the Relu function | 100 × 1 | | Dense (F7) | 5 output neurons consistent with the number of maturity levels | 5 × 1 | | Output layer | The softmax function | |
|
|
