Fig. 4: Deep learning model layout.

The boxes show the filters used in each layer. A total of 500 epochs were run in the training stage, with a learning rate of 0.0001 and Nesterov Adam as an optimizer. This model is a combination of a ResNet CNN model and an RNN model. The output is from the final fully connected layer with 1000 nodes, where each node is a frequency point in the absorption curve