I am trying to classify the Sentinel 1A imagery using supervised classification methods. I have performed the following:
Calibration
GLCM Addition
Speckle Correction
Terrain Correction
After the classification is complete. the classified product formed has two bands: Labelled Classes and Confidence. However, none of those bands are being displayed.
Could anyone explain why this is happening? Also, is there any other method to view classification results?
Hi,
As long as the steps were carried out correctly, I do not find a reason for the results not to be displayed.
Can you please provide a print screen of your results? Is the classified image not displayed at all?
The steps that should be followed for classification are the following:
Image pre - processing (as you described above)
training data selection
on the random forest window, we select ‘train on vectors’ , we select our training vectors we created in the window below ‘vector training’ and finally we select our features (image bands) in the window on the bottom.
I have never used train on raster. The most straightforward way for classification is to use train on vectors. For this option, vector data is required. You can create your training datasets by using the Rectangular drawing tool provided in SNAP. Once you have your vector data for each class, then you select the option train on vectors and select your training vector data on the window below.
because you derived image textures one step before. Speckle removal usually is applied on the images in order to get more homogenous areas. In terms of textures you might destroy texture information which was just generated and could highlight some features which were not visible before. But, as always, this depends on the types of surfaces and the resolution of your data. If you checked the output of the speckle filtering and it looks okay, go on.
If you are using “Train on raster” and your “Training band” has range say in [0.0, 1.0], then you have to check the “Quantize class value” checkbox and specify the “Min class value” to 0, "Class value step size"to 0.1 and "Class levels"to say 10 to quantize the values to 10 levels: 0.0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9.
Note that you may have to click another tab to force “Max class value” to refresh to show 0.9.
Could you tell me how to train data in Random Forest classification in SNAP. I am trying to run RF classification on a non satellite image just for convenience. There is no provision of assigning labels to training pixels. The classification still runs and a band named as labelled classes is created in the target product. I am attaching the original image and output.