Quantitative evaluation of polarimetric classification for agricultural crop mapping

Agricultural crops classification capability of single band full polarization SAR data with different classification methods was evaluated using AIRSAR L-band polarimetric SAR data. It has been found that if only maximum likelihood (ML) classifiers, such as Wishart-maximum likelihood (WML) and normal distribution probability density functions (PDF)-based Maximum Likelihood (NML) classifier can be utilized, it is better to choose WML directly applied to complex coherency or covariance matrix. NML cannot achieve acceptable classification result if intensity and phase images derived from coherency matrix are directly used for training the classifier. But if these images were supplied to the spatial-spectral based classifier, Extraction and Classification of Homogenous Objects (ECHO), higher classification accuracy can be obtained. Very low crop types discrimination accuracy has been observed when only H-Alpha polarimetric decomposition resultant images, such as entropy, alpha and anisotropy, were supplied to NML or a spatial-spectral based classifier such as ECHO.