X
PrevPrev Go to previous topic
NextNext Go to next topic
Last Post 12 Oct 2016 09:12 AM by  anon
Saving trained Nerual Net for later use
 3 Replies
Sort:
You are not authorized to post a reply.
Author Messages

anon



New Member


Posts:
New Member


--
12 Oct 2016 09:12 AM
    Is there any way to save out a neural net (Classification - Supervised - Neural Net) that has been trained on one image, so that it can be used to classify another similar image? It looks like any time I want to run the Neural Net classification on an image I need to select ROIs from that particular image to be used for the training. If there was some way to train a net using one image, then be able to apply it to other images that would be helpful. Is there any way to do this?

    MariM



    Veteran Member


    Posts:2396
    Veteran Member


    --
    12 Oct 2016 09:58 AM
    Currently, ENVI's Neural net does not allow you to save and apply the neural net to other datasets. However, in the next release, there is a new Classification Framework API that will allow this and it will be much more flexible. Here is a short description: The ENVI classification framework provides a powerful and flexible way to classify data. You can create and train a classification algorithm (or classifier) on one dataset and apply it to a similar dataset. Traditionally, classifiers are tied to a specific dataset. With the ENVI classification framework, you can train and evaluate the performance of a classifier once and apply it anywhere, including desktop or enterprise environments. You will be able to use Softmax logistic regression or the SVM classifier.

    Deleted User



    New Member


    Posts:
    New Member


    --
    12 Oct 2016 11:35 AM
    Thanks for this information, that sounds like what I'm looking for. When is the next release scheduled to come out?

    Deleted User



    Basic Member


    Posts:228
    Basic Member


    --
    17 Oct 2016 01:52 PM
    We are hoping to have the next release of ENVI out before the end of this year. The exact date is not yet clear. - Peg
    You are not authorized to post a reply.