Hello, I am processing Landsat images. I would like to classify them using FCM clustering. In original Landsat GeoTIFF's, 0 means there is no data. I replaced them with NaN, this makes better results in some cases, in my application. Problem is - it seems, that FCM cannot work well with NaN's. If image contains them, resulting clusters all are NaN's. What I did - replaced them back to 0. Then it works. But question - is there a better solution? Cause I think, that those zeros might push my resulting clusters in the wrong direction.
