# Difference between convolution structures

I am having a hard time understanding the difference what is a multichannel CNN: In the paper titled, "A Multichannel 2D Convolutional Neural Network Model for Task-Evoked fMRI Data Classification" (https://www.hindawi.com/journals/cin/2019/5065214/#sec2.1) Figure 1 is a multichannel CNN where the authors are using a channel to represent an input image type. Image is a 2D grayscale matrix. According to the diagram, it appears the number of images = number of channels = number of different CNNs. Is that true? I was under the impression that irrespective of the number of channels, we have one CNN that applies filters to each input kind separately. However, this is not the same as having multiple CNNs for each input type.

I am confused -- does multi channel CNN mean that there is a separate CNN for each input data Or a single CNN that applies filters to each input separately.

## 1 Answer

In absolute, it is one CNN wich takes 3 inputs images. You could see it as 3 separate features extractors (CNN) which merge their results while trained together.

The author obtain 3 2D input from a 3D images by keeping 3 2D images; one in each plane.

Each of these images has multiple channel because they slices the input among the respective axis. It is called a "multichannel" CNN because it uses differente representation (3 2D images) of the input (3D image) and therefore extract 3 times the features in order to merge then. (like looking an elephant from each side and above)

• Thank you for your answer. So does the fully connected layer do the merge of the features by combining all the features from the preceding layers for classifying? – Sm1 Jul 20 '20 at 17:33
• Between the last conv layer and the first fully connected layer, the features are flattened and concatened. The fully connected layer use these merged features and decide how to use them. – Chopin Jul 20 '20 at 20:14