Landuse classification of hyperspectral data by spectral angle mapper and support vector machine in humid tropical region of India

Hyperspectral images are being used in various fields. The main objective of the present study was to use hyperspectral imagery from Hyperion with Spectral Angle Mapper (SAM) and Support Vector Machine (SVM) for discriminating the landuse/landcover classes in Kozhikode district, Kerala which constitutes a combination of different physiographic land features. Hyperion functions from a space platform with modest surface signal levels and a full column of atmosphere persuading the signal, hence, the data derived from this demand careful pre-processing to minimize sensor and atmospheric noise. The atmospheric correction using MODTRAN based FLAASH module as well as the data dimensionality reduction by Principal Component Analysis (PCA) made the Hyperion to allow discrete reflectance values. Advanced classifiers like SAM and SVM could describe the pattern and spatial distribution of landcover. From the accuracy assessments, SVM showed better classified result than SAM with overall accuracy 85.6% and kappa coefficient 0.89. This study suggests that SVM can be used for landuse/landcover classification of hyperspectral data with high accuracy.