Hyperspectral Imaging for Crop Disease Detection: A Systematic Literature Review and Research Gap Analysis

Abstract

Crop diseases cause 20-40% of food losses every year and cause economic damage of more than USD 220 billion per year on a global scale. The basic problems in precision agriculture remain the same as early and accurate disease detection. Hyperspectral imaging (HSI) is a type of imaging technique that captures hundreds of contiguous wavelengths of the electromagnetic spectrum spanning from 400 to 2500 nm that has been found to be a useful non-destructive diagnostic tool that can detect the subtle biochemical differences that occur in plant tissue before symptoms are visible. The paper critically summarizes and reviews the literature from 2000 to 2024, especially focusing on the application of AI and machine learning (ML) for HSI-based crop disease detection. A total of 48 primary studies are reviewed and grouped into five thematic categories: (1) spectral vegetation index methods, (2) classic machine learning classifiers, (3) deep learning architectures, (4) attention and transformer mechanisms and (5) disease severity quantification. Based on this review, four gaps in the literature are identified: (1) lack of comparison of classical and deep learning models on the same splits of the same datasets, (2) underutilisation of the SWIR-2 spectral range (>2000 nm) for the discrimination of diseases, (3) lack of integrated spatial mapping of the disease severity from spectral index fusion, and (4) lack of lightweight deep learning spectral-only architectures for field deployment in resource-constrained environments. These gaps together form a promising research program based on this AI approach to automated crop disease detection, and the experimental research work reported in our companion paper is fueled by these gaps.

Plant phenotyping relevance

植物病害の症状・重症度をハイパースペクトル画像とAIで推定する手法を対象とした体系的レビューであり、植物フェノタイピング手法のレビューが中心。

Code and data availability

公開本文の所在を確認できませんでした。非公開または購読が必要な可能性があります。

No evidence-backed public reproduction asset is currently recorded.