Help creating a Multi-Input CNN with 4D array images for Multi-Channel EEG analysis ?
显示 更早的评论
I have scalograms of EEG signals of 16 different channels stored in 4D array .tiff files of size 299x299x3x16 (which are 16 images of 299x299x3) I'm trying to pass this files through a NN to classify the images out of three different classes. I have tried working with GoogleNet and Inception-ResNet-v2 as I'm not familiarized with building NN i decided to modify the input and output of this NN for my task but I found issues trying to modify the input layer and when i manage to make the input the size of the file the conv layers below give me issues like the dimensions of the input are not valid like:
Error using trainNetwork
Invalid network.
Caused by:
Layer 'conv2d_2': Input data must have two spatial dimensions only, one spatial dimension and one temporal dimension, or two spatial dimensions and one temporal
dimension. Instead, it has 3 spatial dimensions and 0 temporal dimensions.
My question is how can i modify Inception-ResNet-v2 to achieve this task or how can i build a NN myself for this (it doesn't have to be a complex one)?
I will attach one of the images i generated, to visualized just use the code below and change n to visualize 1 of the 16 images.
imshow(imstack(:,:,:,n))
Thanks for the help
回答（1 个）
I know nothing about EEG signal processing, but it seems to me that you should just view the inputs as 299x299 images with 48 channels. In other words, reshape the imstack as follows before giving it to the network:
imstack=imstack(:,:,:);
类别
在 帮助中心 和 File Exchange 中查找有关 EEG/MEG/ECoG 的更多信息
2023-2-19
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!
Translated by 
