MLVI-CNN: a hyperspectral stress detection framework using machine learning-optimized indices and deep learning for precision agriculture.

Abstract

Introduction Early and accurate detection of crop stress is vital for sustainable agriculture and food security. Traditional vegetation indices such as NDVI and NDWI often fail to detect early-stage water and structural stress due to their limited spectral sensitivity. Method This study introduces two novel hyperspectral indices - Machine Learning-Based Vegetation Index (MLVI) and Hyperspectral Vegetation Stress Index (H_VSI) - which leverage critical spectral bands in the Near-Infrared (NIR), Shortwave Infrared 1 (SWIR1), and Shortwave Infrared 2 (SWIR2) regions. These indices are optimized using Recursive Feature Elimination (RFE) and serve as inputs to a Convolutional Neural Network (CNN) model for stress classification. Results The proposed CNN model achieved a classification accuracy of 83.40%, effectively distinguishing six levels of crop stress severity. Compared to conventional indices, MLVI and H_VSI enable detection of stress 10-15 days earlier and exhibit a strong correlation with ground-truth stress markers (r = 0.98). Discussion This framework is suitable for deployment with UAVs, satellite platforms, and precision agriculture systems.

Code and data availability

The supplied blocks describe the MLVI-CNN methodology using the GHISACONUS Hyperspectral Spectral Library (a pre-existing public dataset from NASA's EO-1 Hyperion, i.e., cited prior work rather than a paper-specific deposit). No data availability statement, code repository, trained model checkpoint, or author-deposited

No evidence-backed public reproduction asset is currently recorded.