Spectral Angle Mapper classification is a useful technique when you are attempting to find specific materials and need to account for potential variations in spectra due to topography. It is a powerful technique, but it is only as good as the data you give it to work with. Multispectral sensors like TM, SPOT, and ASTER probably do not have sufficient spectral resolution to separate table grapes from surrounding vegetation. With all three you should be able to distinguish vegetation from other ground cover types in a general sense, but species- and sub-species level identifications for plants is difficult--even with hyperspectral sensors like AVIRIS, AISA, HYDICE, and CASI. The key would be to work with data with very high spectral resolution (10 nanometers or less) in the visible and near infrared (VNIR) portion of the spectrum, where plants tend to have characteristic absorption features. The features could help identify grapes based on one of several vegetation indices (found under Spectral->Vegetation Analysis->Vegetation Index Calculator) or perhaps a classificaiton technique like SAM or an unmixing technique like MTMF.