Assessment of EO-1 hyperion imagery for crop discrimination using spectral analysis

Abstract

This paper outlines the research objectives to discriminate crop species using pure spectral-spatial reflectance of EO-1 Hyperion imagery. Vigorous encroachment in remote sensing unlocks the new avenues to investigate the hyperspectral imagery for analysis and implication for crop-type classification and agricultural management. The investigated crop species were namely Sorghum, Wheat, and cotton located in West zone of Aurangabad, Maharashtra, India. The preprocessing algorithm namely quick atmospheric correction (QUAC) was applied to calibrate bad bands and construct precise data for crop discrimination. The machine learning classifiers applied to identify the pixels having a significant difference in pure spectral signatures based on Ground Control Point (GCP) and image spectral responses. The investigation was based on a binary encoding (BE) and support vector machine (SVM) learning approach in order to discriminate crop types. Crop discrimination followed land cover classes gives 73.35% accuracy using BE and SVM with polynomial third-degree order gives overall accuracy 90.44%. These results show that satellite data with 30 m spatial resolution (Hyperion) are able to identify crop species using Environment for Visualizing Images (ENVI) open source software.