Automated Classification
Hello everyone!
In the preparation of a project i want to make an automated classification of an area in greenland. for this i have calculated spectral indices (NDVI,NDWI,..) for a tile. Now i want to find thresholds for the classes and make them transferable to other tiles. Would the best solution be to find the thresholds myself and specify them via assing class or is there a better solution?
Thanks a lot!
Kind regards
Simba
Was this article helpful?
4 comments
First of all, please try the 'automatic threshold' algorithm: https://support.ecognition.com/hc/en-us/articles/360017657260-Automatic-Threshold
Combine it with unsupervised learning (if the results makes sense): https://support.ecognition.com/hc/en-us/articles/360016655299-Unsupervised-Classification-Segmentation-
And/or supervised learning: https://support.ecognition.com/hc/en-us/articles/360016174419-Supervised-Classification
At the end, please optimize the combination with manual thresholds based classification (assign class) or fuzzy logic based classification (membership functions).
i have used the automatic threshold method via video ;) before. Unfortunately, the result was not so good.
My first thought was to do a multiresolution segmentation and determine manual tresholds.
Does anyone know if/ and how well this process tree can be transferred to other tiles?
Greets
First you should determine what characteristics enable you to extract features of interest. Is it NDVI? Are the values stable over the entire image? Most likely you will need to make use of contextual information (e.g., relative border to).
Thanks for your advice Jarlath!