I have been working with a Landsat 8 image (reflectance values) to create a land cover map.I have used the Maximum likelihood algorithm and a set of ROIS (6 classes:forest, agriculture, water, shrubs, infrastructures, water). I have used the same parameters for the algorithm (Scale factor: 1.00 and ProbabilityThreshold: 0.85).
Using Envi5 SP3 (with the Landsat hotfix) or the Envi Classic I obtain the same result(with involves many unclassified pixels, as could be expected), but it is completely different when I use ENVI 4.7 (installed on a different computer). Using ENVI4.7 most of the pixels are classified, while with ENVI 5 or Classic even manypixels belonging to the class “water” are not classified. We have repeated theclassification several times and the results are consistent; Envi 5 and Classic provide the same results and Envi 4.7, different.We havedone it with other datasets/computers and we have had the same problem: using the same probability threshold ENVI 5/classic classifies less pixels than when we use ENVI 4.7.
I would behappy to know why this happens and if someone has had the same issue.
Thank you for your time!
|