Estimation of SPAD in corn leaves affected by Physoderma Brown Spot based on optimized vegetation indices combined with texture features

Chlorophyll content is a key physiological indicator for maintaining leaf health and characterising disease progression. Accurate quantification of leaf chlorophyll content provides reliable evidence for assessing the severity of Physoderma Brown Spot (PBS). Therefore, it is essential to develop SPAD estimation models specifically designed for disease conditions. Based on the spectral response mechanisms under PBS, this study proposes a novel optimisation method for hyperspectral vegetation indices (VIs), integrating texture features through multi-feature fusion to improve SPAD estimation accuracy under disease stress. First, an exhaustive band-by-band traversal was applied, and all wavelength combinations were assembled using candidate index formulations. While optimising VIs, bands from the optimal combinations were selected to extract corresponding texture features. Sensitive texture features were then identified using supervised dimensionality reduction. Finally, the stacking strategy integrates the predictions from the VI-based model and the texture-based model, rather than directly combining them as input features.. The results show that a ratio-type three-band VI constructed from 646, 694, and 966 nm achieved the best performance for SPAD estimation under PBS, with a correlation coefficient of 0.7059 and achieved an R² of 0.5430 with an RMSE of 10.7242. The sensitive texture features selected via supervised dimensionality reduction effectively captured the distribution patterns of SPAD values and exhibited reasonable estimation capability, achieved an R² of 0.4453 with an RMSE of 11.8155. By further integrating the optimised VI and sensitive texture features using ensemble learning, model performance was substantially improved, achieving an R² of 0.8276 and an RMSE of 6.6549. This study successfully developed a high-accuracy SPAD inversion model under PBS conditions, demonstrating great potential for disease diagnosis and management in precision agriculture.