| Sumario: | The international exchange of rice germplasm is fundamental for crop improvement and production but poses risks of spreading seed-borne pathogens like Alternaria padwickii. Conventional detection methods, such as the blotter test, are effective but destructive and time-consuming. This study was conducted to investigate hyperspectral imaging (HSI) in the 400-1000 nm range as a rapid, non-destructive alternative in detecting A. padwickii in rice seeds. Initial classification models using whole-seed spectral data demonstrated high specificity but failed to achieve adequate sensitivity, as signals from localized infections were diluted by spectra of healthy tissue. Recognizing that A. padwickii infection originates at the seed tips, it was hypothesized that a targeted analysis would yield superior results. A second experiment isolated the seed tip as a specific region-of-interest (ROI). This biologically-informed strategy proved critical, filtering spectral noise and dramatically improving performance. To handle data dimensionality, spectral refinement was performed using Partial Least Squares Discriminant Analysis (PLS-DA). Based on Variable Importance in Projection (VIP) scores, the most informative wavelengths were identified for classification. This combined ROI and spectral refinement approach enabled multiple machine learning models to achieve perfect classification scores (100% sensitivity, specificity and accuracy). The models’ high accuracy was maintained even in the presence of other common seed-borne fungi, demonstrating robustness across diverse fungal conditions. HSI, coupled with a biologically-informed ROI and targeted spectral refinement, is therefore a powerful, non-destructive tool for detecting A. padwickii. An enhanced workflow is proposed, using HSI as a high-throughput screening tool to significantly improve the efficiency in seed health testing for germplasm transboundary movement.
|
|---|