Skip to main content
. 2018 Jul 27;14:64. doi: 10.1186/s13007-018-0332-5

Table 2.

Layers property of the CNN architecture. The network consists of twenty-five layers. There are eight layers with learnable weights: five convolutional layers, and three fully connected layers

No. Layer name Description
1 Image input 227 × 227 × 3 true color images with zerocenter standardization
2 1st-level convolution 96 channels, 11 × 11 × 3 convolutions
3 ReLU Rectified linear units
4 Cross channel standardization Cross channel standardization with 5 channels per element
5 Max pooling 3 × 3 max pooling
6 2nd-level convolution 256 channels, 5 × 5 × 48 convolutions
7 ReLU Rectified linear units
8 Cross channel standardization Cross channel standardization with 5 channels per element
9 Max pooling 3 × 3 max pooling
10 3rd-level convolution 384 channels, 3 × 3 × 256 convolutions
11 ReLU Rectified linear units
12 4th-level convolution 384 channels, 3 × 3 × 192 convolutions
13 ReLU Rectified linear units
14 5th-level convolution 256 channels, 3 × 3 × 192 convolutions
15 ReLU Rectified linear units
16 Max pooling 3 × 3 max pooling
17 6th-level fully connected layer 4096 fully connected layer
18 ReLU Rectified linear units
19 Dropout 50% of dropout
20 7th-level fully connected layer 4096 fully connected layer
21 ReLU Rectified linear units
22 Dropout 50% of dropout
23 8th-level fully connected layer 4 fully connected layer
24 Softmax Softmax
25 Comprehension output Crossentropyex with Androsace umbellata (Lour.) Merr., Bidens pilosa L., Trifolium repens L. and Fragaria × ananassa