abhilash-neog / hyperspectral-image-learning Goto Github PK

Make 3 folders. One for the images[make sure to change the path], one for the graphs generated and the other for code. And also add a readMe file :)

To make a Depth-wise separable CNN- will drastically reduce number of multiplications without losing much accuracy.
Refer to this link: https://www.geeksforgeeks.org/depth-wise-separable-convolutional-neural-networks/

Made some changes to SOM_v1. Try running it once. Taking too much time on CPU

The current approach of reducing the dimensions to extract useful features is making things complicated.
Either we apply a different approach sticking to pca, or, go for autoencoders(not much of application till date in terms of HSI dimension reduction)

k-Fold cross validation might increase the accuracy to an extent. Required to be implemented