Number of training samples at Random forest classifier

like i have three images of sentinel 1 grd one image of pre-flood and two images of post-flood. after pre-processing please tell me the steps for RFC.

change detection is a different approach actually.

Have you seen these tutorials:

However, your approach with RF stays the same. Calculate image textures and classify your image(s).

yes i have attented that course but i was trying to mapped it with RFC. so i will try your methods
thank you

I have 4 training sets (and each set has multiple polygons). Also I will use 128 bands or more. So, what should be the optimum “number training samples” and “number of trees” for my RF classification? (Number of pixels of the smallest training set are 291 pixels and number of pixels of the biggest training set are 1850 pixels, and total number of pixels are 3489 pixels)


1 Like

Dear @ABraun,

So is there any estimated number or relation between number of classes (need to classify), number of training samples, number of trees and number of input bands (variables)?

Thank you,

training/validation are often split 2/3 to 1/3, to test the accuracy of the classifier (not the classified map), but the number of trees and the number of input bands are not linearly linked to the other parameters.