• A non-invasive Vis/NIR and ML method was developed for early detection and quantification of ZYMV in zucchini. • A novel spectral index (SZI) framework was developed for early and asymptomatic prediction of ZYMV in zucchini. • The asymptomatic three-band SZI in 400-920 nm predicted virus load, validated by RT-qPCR regression models. Timely and accurate detection of plant viral infections is crucial for effective crop health management. This study introduces a novel non-invasive Visible-Near Infrared (Vis/NIR) Spectroscopy approach for diagnosing Zucchini yellow mosaic virus (ZYMV) in zucchini plants at both symptomatic and asymptomatic stages. Spectral reflectance data (400–2500 nm) were collected from 507 inoculated plants, with PCR verification of infection by day 3 post-inoculation providing accurate labels for supervised model development. Utilizing Random Forest (RF) -based feature selection, eight relevant wavelengths were identified and subsequently employed to create a new three-band Spectral Zucchini Index (SZI). The SZI demonstrated remarkable performance, achieving 100% classification accuracy for symptomatic samples and 89% accuracy for asymptomatic detection. This performance significantly surpasses that of conventional indices such as Normalized Difference Vegetation Index (NDVI), Photochemical Reflectance Index (PRI), and Normalized Difference Water Index (NDWI), highlighting the potential of Vis/NIR spectroscopy for early ZYMV detection. Beyond classification, regression modeling was also applied to correlate SZI values with viral concentration levels, measured by Real-time quantitative Polymerase Chain Reaction (RTqPCR). Among six machine learning regressor algorithms, the Random Forest regressor exhibited the highest performance after tuning, yielding a Root Mean Squared Error (RMSE) of 0. 91, Mean Absolute Error (MAE) of 0. 35, and Mean Absolute Percentage Error (MAPE) of 8. 12%. These findings underscore the feasibility of integrating hyperspectral sensing, spectral feature engineering, and machine learning for scalable and timely ZYMV detection, offering a complementary, non-destructive screening approach to molecular diagnostics in controlled experimental settings. The SZI framework holds potential for adaptation to detect other plant viruses or abiotic stresses, facilitating real-time crop monitoring via edge-based AI systems.
Kazemi et al. (Sun,) studied this question.