Heliyon (Oct 2024)
Early detection and spectral signature identification of tomato fungal diseases (Alternaria alternata, Alternaria solani, Botrytis cinerea, and Fusarium oxysporum) by RGB and hyperspectral image analysis and machine learning
Abstract
Early identification of plant fungal diseases is critical for timely treatment, which can prevent significant agricultural losses. While molecular analysis offers high accuracy, it is often expensive and time-consuming. In contrast, image processing combined with machine learning provides a rapid and cost-effective alternative for disease diagnosis. This study presents a novel approach for detecting four common fungal diseases in tomatoes, Botrytis cinerea, Fusarium oxysporum, Alternaria alternata, and Alternaria solani, using both RGB (visible) and hyperspectral (400–950 nm) imaging of plant leaves over the first 11 days post-infection. Data sets were generated from leaf samples, and a range of statistical, texture, and shape features were extracted to train machine learning models. The spectral signatures of each disease were also developed for improved classification. The random forest model achieved the highest accuracy, with classification rates for RGB images of 65%, 71%, 75%, 77%, 83%, and 87% on days 1, 3, 5, 7, 9, and 11, respectively. For hyperspectral images, the classification accuracy increased from 86% on day 1 to 98% by day 11. Two- and three-dimensional spectral analyses clearly differentiated healthy plants from infected ones as early as day 3 for Botrytis cinerea. The Laplacian score method further highlighted key texture features, such as energy at 550 and 841 nm, entropy at 600 nm, correlation at 746 nm, and standard deviation at 905 nm, that contributed most significantly to disease detection. The method developed in this study offers a valuable and efficient tool for accelerating plant disease diagnosis and classification, providing a practical alternative to molecular techniques. .
