Figure 1.

The architecture of the Convolutional Neural Network (CNN). Input are the spectra with 3438 frequencies, which are subjected to the following operations in three iterations: pattern detection (1D convolution), activation function (leaky ReLu), and data reduction (maximum pooling). The pattern detection window (kernel) size is 3 in each layer, and the convolution layers have 8, 16, and 32 descriptors (patterns), respectively. Finally, the output is flattened and input to the MLP (light blue) with 3 hidden layers of sizes 256, 128, 56 and as many outputs as classes to be predicted.