Crop type discrimination using Geo-Stat Endmember Extraction and machine learning algorithms

Area of Science:

• Agricultural Science
• Remote Sensing
• Spectroscopy

Background:

• Crop diversity is vital for climate change adaptation, agricultural productivity, and food security.
• Hyperspectral Remote Sensing (HRS) offers a powerful method for discriminating crop types using detailed spectral information.
• Airborne Visible Infrared Imaging spectrometer- New Generation (AVIRIS-NG) data provides high-resolution spectral bands for advanced crop analysis.

Purpose of the Study:

• To explore and evaluate techniques for crop classification and identification using AVIRIS-NG data.
• To generate a spectral library of different crop types for accurate discrimination.
• To compare the performance of various supervised classifiers for crop identification.

Main Methods:

• Utilized AVIRIS-NG data for hyperspectral imaging spectroscopy.
• Employed the Geo-Stat Endmember Extraction (GSEE) algorithm to identify pure pixels and create a spectral library.
• Compared spectral features from AVIRIS-NG, ASD-Spectroradiometer, and Continuum Removed (CR) spectra.
• Applied ten supervised classifiers, including deep learning (2D-CNN) and ensemble methods, for crop discrimination.
• Evaluated classifier performance using Overall Accuracy, Kappa Coefficient, Precision, Recall, and F1 score.

Main Results:

• Successfully identified nine crop types: wheat, maize, tobacco, sorghum, linseed, castor, pigeon pea, fennel, and chickpea.
• The 2D-Convolutional Neural Network (2D-CNN) classifier demonstrated superior performance.
• Achieved high performance metrics with 2D-CNN: 89.065% Overall Accuracy, 0.871 Kappa Coefficient, 87.565% Precision, 89.541% Recall, and 88.678% F1 score.

Conclusions:

• Hyperspectral Remote Sensing with AVIRIS-NG data and advanced classifiers like 2D-CNN is highly effective for crop type identification.
• The developed methodology enables accurate, large-scale crop mapping at the species level within short timeframes.
• This approach supports agricultural monitoring, climate change adaptation strategies, and enhances food security efforts.