64-3 Crop Classificatin With Spectral Angle Mapper.

See more from this Division: ASA Section: Climatology & Modeling
See more from this Session: Symposium--Use Of Remote Sensing For Crop and Pasture Statistics: I
Monday, November 4, 2013: 1:35 PM
Tampa Convention Center, Room 9
Share |

E. Raymond Hunt Jr., Hydrology and Remote Sensing Lab, USDA-ARS, Beltsville, MD
The Spectral Angle Mapper (SAM) is a very flexible method for classification of remotely sensed data.  Two reflectance spectra from either multispectral or hyperspectral sensors are defined as vectors in n-dimensional space, where n is the number of sensor bands.  The cosine of a spectral angle (Θ) is the dot product of the two vectors divided by the normalized vectors.  Small Θ represent similar spectra and large Θ represent different spectra, but the criteria for large and small depend on the application.  For hyperspectral target detection, spectral angles are calculated using a single target spectrum for every pixel in an image, and a threshold angle is used to determine matches to a target.  Increasing the threshold value (usually about 0.1 radians or 5.7 °) increases the number of false positives and decreasing the threshold value increases the number of false negatives.  For supervised image classification of a single pixel, spectral angles are calculated for each class signature, and the pixel is assigned to the class with the smallest angle.  A key advantage of SAM compared to other supervised methods is that simple differences in pixel brightness (caused by either a north or south facing slope) are eliminated by the vector normalization.  The key questions are whether each plant species has a unique reflectance spectrum and if all of the little bumps and wiggles in a reflectance spectrum can be used to distinguish among plant species.  Some cases yes, most cases no, because plant chemical composition is very similar among species.   There has been little systematic assessment on the strengths and weaknesses of SAM, but an array of spectral indices may lead to better classifications than an array of spectral reflectances.
See more from this Division: ASA Section: Climatology & Modeling
See more from this Session: Symposium--Use Of Remote Sensing For Crop and Pasture Statistics: I