Pre-Symptomatic Identification of CMV and PVX Infection in Nicotiana benthamiana Using Spectral–Spatial Fusion of Hyperspectral Imaging

Early detection of cucumber mosaic virus (CMV) and potato virus X (PVX) infection at the pre-symptomatic stage is essential for timely disease management and for limiting viral spread. Conventional molecular assays are accurate, but they generally require destructive sampling and are time-consuming. To enable rapid and non-destructive identification of these two viruses before visible symptom development, Nicotiana benthamiana seedling leaves were used as experimental materials, and CMV and PVX were selected as target viruses. A hyperspectral dataset of healthy, CMV-infected, and PVX-infected pre-symptomatic samples was constructed and validated by RT-PCR. After spectral preprocessing, eight selected wavelengths were identified and used to develop a 1D-CNN model, a MobileNetV3 model, and a spectral–spatial dual-branch fusion model. The 1D-CNN achieved an accuracy of 0.9074, whereas MobileNetV3 achieved 0.6835. The feature-level fusion model performed best, with an accuracy of 0.944, a precision of 0.945, a recall of 0.944, and an F1-score of 0.944. These results suggest that spectral information provides the primary discriminative basis, while image information offers complementary spatial and textural features for early non-destructive detection of plant viruses.