Hi,
I'm working in R and I want to analyze some 8-bands resampled spectral signatures using SAM classification. I don't want to use an image to obtain a map of the surface, I have a dataset with two categories and I want to classify each sample into one of the two groups using vectorialization of the data in the multidimensional space and a threshold.
Dear Elena,
Hi, interesting question. You need to create a library of the spectral signatures. But before doing that, you need to re-sample the spectral signatures to fit the number of bands and the covered wavelengths by the used satellite images. After that you can provide the library as training data to the SAM supervised classification algorithm.
Best regards
N.B. Attached examples of original spectral signatures of wheat and potato graphs. Next you will see the same signatures re-sampled for Landsat and CHRIS
Dear Awad,
thank you for the answer.
I have already re-sampled into the 8 bands I'm interested in, and I created the library. It consists of a dataset of 70 spectral signatures representing four different objects. For each group, I have a signature corresponding to the standard endmember for that signature.
Now, I want to compare the vectorialized spectra in R , similarly to what SAM does comparing the library to the image, but the SAM function in R ( https://rdrr.io/cran/RStoolbox/man/sam.html ) requires an image (img). I want to use the library instead...
sam(img, em, angles = FALSE, ...)
So, how can I clusterise my 70 signatures to the four groups comparing the vectorialized spectra and a fixed threshold? Is there any tool in R able to do it? Basically instead to use pixels I want to give the algorithm directly the data organized as .csv or .txt
Thank you for your help
- Badges
- Science topic
- Similar topics
- Data Mining
- Classification