Structure 2: For this structure the input is convolved with 128 filters of size (5 × 5) before a multiscale filter bank is applied. Next, three convolutionalized blocks are placed, where the residual blocks are used with three convolutional layers. Similar to structure 1, a dropout layer and two fully connected layers are implemented to extract the output of the network. The total number of trainable parameters in this structure is 4538 497.