Table 2 Hyper parameter setting of network
From: Flight maneuver intelligent recognition based on deep variational autoencoder network
Name | Structure | Parameter |
|---|---|---|
Encoder | Connection layer | Input: 22, output: 128 |
Connection layer | Input: 128, output: 64 | |
Connection layer | Input: 64, output: 10 | |
Parameters of the split | Split the mean and variance | Input: 10, output: 2*5 |
Heavy parameterized | Compute hidden layer variables | Input: 2*5, output: 5 |
Decoder | Connection layer | Input: 5, output: 64 |
Connection layer | Input: 64, output: 128 | |
Connection layer | Input: 128, output: 22 | |
Down-sampling | Convolution layer | Convolution kernel size: 3 Convolution kernel number: 64 Fill a null: 1 |
Pooling layer | 2 × 2 | |
Convolution layer | Convolution kernel size: 3 Convolution kernel number: 128, Fill a null: 1 | |
Pooling layer | Pooling of nuclear: 2 × 2 | |
Up-sampling | Transpose convolution layer | Convolution kernel size: 2 Convolution kernel number: 64 Step away from: 2 |
Convolution layer | Convolution kernel size: 3 Convolution kernel number: 64 Fill a null: 1 | |
Transpose convolution layer | Convolution kernel size: 2 Convolution kernel number: 16 Step away from: 2 | |
Convolution layer | Convolution kernel size: 3 Convolution kernel number: 1 Fill a null: 1 |