RGB image-based drought stress classification of garden plants using SVM model

Water management in urban gardens is increasingly complex due to diverse plant species and growing drought stress under climate change. This study proposes a non-destructive method to classify drought responses of mixed garden plant species using RGB image indices and a support vector machine (SVM) model. Chlorophyll fluorescence responses were used to evaluate photosynthetic stress, while image-derived indices—green leaf index (GLI), normalized green-red difference index (NGRDI), and blue-green pigment index (BGI)—were analyzed to assess drought responses. Hierarchical clustering grouped species into three response clusters based on fluorescence and image patterns. Principal component analysis (PCA) identified NGRDI, GLI, and BGI as key variables, with NGRDI and GLI showing strong correlations with soil moisture content and BGI distinguishing cluster-specific responses. An SVM model was constructed using RGB indices and soil moisture content as input features, achieving a classification accuracy of 88.3% and an F1 score of 0.85 through five-fold cross-validation. This approach supports efficient water management and plant selection, and can be extended to precision irrigation strategies using machine learning.