Architecture of the CNN used to identify major mergers and non-mergers. The pre-trained VGG network (shaded area) is used with the top three fully connected layers removed. There are five blocks containing 2, 4, 4, 4, and 4 convolutional layers with 64, 128, 256, 512, and 512 kernels respectively. All kernels are 3 × 3 pixels and each block is connected with a 2 × 2 pixel max pooling layer (blue arrows). We add a single fully connected layer with 512 neurons and a fully connected output layer with 2 neurons (unshaded area). The input to the network is a 88 × 88 pixel image with three identical channels.