I have High-resolution multispectral image and I need to classify it taking input from GLCM texture analysis. I got 10 feature extraction using SNAP and now need to input all these bands to classify using supervised classification. May I know how can I take this as an input?
if you want to have the multispectral bands as well, you have to create a stack with the original product and the textures and then use this stack as input for the classification.
It is worth mentioning that many textures are quite redundant and so are the multispectral bands. So I would recommend a random forest classifier with a large number of trees (> 250) to overcome this redundancy and the identification of the most suitable features.
I got 9 texture features. Does adding multispectral bands(3) will drop the accuracy of classsification results? Also, may I know how can I stack both 3 multispectral bands and the 9 textured features.
Is it possible to perform PCA to reduce this redundancy and then stack textured measures to multispectral bands?
No, the accuracy will not drop with more bands, but at a certain point more bands do not increase it either. 3 bands and 9 textures is completely fine, you don’t need a PCA for this.
You can combine all products using the colocation tool.
May I know how to access the colocation tool? I did as suggested in the help contents but couldn’t able to find it yet. Could you please guide me to access it?