Performance of Pure Pixel Extraction Algorithms on Hyperspectral Data for Species Level Classification of Mangroves

This study attempts to apply, compare and analyze the performance of automated target detection algorithms on hyper spectral data with the aim to identify mangroves species in the Sunderban Delta of West Bengal. The performance of algorithms such as Pixel Purity Index (PPI) and NFINDR has been evaluated on the basis of spectral difference between pure pixels extracted from image data and ground measured values. The accuracy of each algorithm in identification of pure mangrove patches has been assessed from the classification results obtained after spectral unmixing of the Hyperion data. Linear Mixing Model (LMM) has been applied on the hyper spectral imagery for calculation of abundance values of each sub-pixel existent within each pixel using the pure spectra derived from the target detection algorithms. It has been observed that NFINDR shows higher accuracy in identification of pure spectra of mangrove species as compared with PPI. The algorithm has been successful in identifying dominant species namely Avicennia Marina, Avicennia Alba, Avicennia Officinallis, Excoecaria Agallocha, Ceriops Decandra, Phoenix Paludosa and Aegialitis. The accuracy assessment of identified endmembers is achieved by calculating the Root Mean Square Error (RMSE) of image derived data and field measured data.