Table 1. Convolutional neural network architecture.

Layer	Type	Maps and neurones	Kernel size
0	input	3 maps of 37 × 37 neurons
1	convolution	100 maps of 35 × 35 neurons	3 × 3
2	pooling	100 maps of 18 × 18 neurons	2 × 2
3	convolution	150 maps of 16 × 16 neurons	3 × 3
4	pooling	150 maps of 8 × 8 neurons	2 × 2
5	convolution	150 maps of 6 × 6 neurons	3 × 3
6	pooling	150 maps of 3 × 3 neurons	2 × 2
7	fully connected	300 neurons	1 × 1
8	fully connected	2 neurons	1 × 1

The input is processed from the top to the bottom, where the two output neurons each represent one class. Rectified linear activation is used after each convolution and the first fully connected layer. The two final neurons are activated by a softmax function and can be interpreted as the probability of a particular input to belong to the respective class.