X
PrevPrev Go to previous topic
NextNext Go to next topic
Last Post 13 Nov 2019 05:26 AM by  MariM
Using own shapefile ground truth data for supervised OBIA
 3 Replies
Sort:
You are not authorized to post a reply.
Author Messages

Jodie Robertson



New Member


Posts:1
New Member


--
11 Nov 2019 08:50 PM
    I am trying to undertake a supervised OBIA of vegetation types on a sentinel 2 image. I am having trouble using a point shapefile as training data in the ENVI example based classification. My training data is based on a stratified random sample of a polygon layer containing 13 vegetation type classes. When I add it as ground truth data, it does not seem to choose the attributes in the field selected to assign classes and seems to assign them randomly. I am not sure how the training data is being utilised in this case.


    MariM



    Veteran Member


    Posts:2396
    Veteran Member


    --
    12 Nov 2019 08:49 AM
    If you are using example based classification, have you tried setting up the class_id as an attribute and assign each class to the class_id? From the documentation:

    From the Select Attribute drop-down list, select the shapefile attribute to group vector records into training classes. The default is CLASS_ID. You should choose unambiguous attribute fields for grouping records into training classes.

    Jodie Robertson



    New Member


    Posts:1
    New Member


    --
    12 Nov 2019 05:24 PM
    Thanks, I had string attributes which is I think why they were not working, I have since changed them to numerical, which I think will work, however, ENVI is having difficulty process them and consistently gets most of the way before crashing. Around what size training data set can ENVI handle, currently I have 13 categories, with between 29 and 4983 points per category. I am using single date sentinel 2 imagery of the Tiwi Islands, NT, Australia.

    MariM



    Veteran Member


    Posts:2396
    Veteran Member


    --
    13 Nov 2019 05:26 AM
    There shouldn't be a limit (or a limit you should hit) for training data size. However, if you are using an earlier version of ENVI, there were some issues with restoring large training data when it was loaded into the layer manager which I believe was resolved by ENVI 5.4.
    You are not authorized to post a reply.