Preliminary study on disease severity classification of southern corn rust based on a novel vegetation index

Southern corn rust (SCR) is one of the major diseases threatening global corn production, and accurate monitoring of its disease progression is of great importance for effective disease management. Unmanned aerial vehicle (UAV) multispectral remote sensing presents an efficient approach for SCR monitoring,However, the spectral response mechanisms during SCR development remain insufficiently understood.To address this issue, this study systematically evaluated the applicability of selected vegetation indices for SCR severity classification using UAV multispectral imagery and plot-level disease severity index data from three corn cultivars. First, 34 vegetation indices (VIs) with a total of 5,850 band combinations were constructed. A multilayer perceptron (MLP) classification model combined with three-fold cross-validation was employed, and the Macro F1-score was used to identify the optimal single-feature indices. The results indicated that vegetation indices can effectively discriminate SCR severity levels. The best-performing two-, three-, and four-band indices were DVI (842, 668), BaI (475, 705, 842), and FCVI (668, 842, 740, 650), achieving Macro F1-scores of 0.709 ± 0.001, 0.730 ± 0.004, and 0.717 ± 0.003, respectively. Furthermore, recursive feature elimination (RFE) and minimum redundancy maximum relevance (mRMR) were introduced for multi-VIs selection, and six machine learning models were applied for performance evaluation. The results demonstrated that multi-VIs models outperformed single-VIs models, and satisfactory classification performance could be achieved with only 12 or even 6 features. In particular, using RFE on a mixed-band dataset in combination with the XGBoost model yielded strong classification performance, with a Macro F1-score of 0.913 ± 0.00117.