Effectiveness of SIFT Features in Enhancing Watermelon Leaf Disease Classification Accuracy

The prompt and precise identification of plant diseases is essential for mitigating yield losses and safeguarding global food security. This study examines the categorization of watermelon leaf diseases utilizing both unprocessed grayscale pixel values and manually derived visual features obtained through the Scale-Invariant Feature Transform (SIFT). Two categorization models, Support Vector Machines (SVM) and Artificial Neural Networks (ANN), were developed and assessed. A bespoke picture collection was utilized, comprising healthy watermelon leaves and those afflicted with prevalent illnesses including mosaic virus, downy mildew, and anthracnose. The preprocessing stages encompassed scaling, conversion to grayscale, and normalizing. In the initial method, the flattened pixel values were directly fed into the models. SIFT descriptors were calculated and averaged to create feature vectors in the second instance. Experimental findings indicate that SIFT-based feature extraction markedly improves classification efficacy. The ANN-SIFT model achieved superior results, attaining an accuracy of 0.9960 and an F1-score of 0.9957. The SVM-SIFT model exhibited robust classification proficiency. Conversely, models exclusively trained on raw pixel inputs yielded less accuracy and inferior class separability. The model's performance was evaluated using metrics such as accuracy, precision, recall, specificity, and F1-score, supplemented with visual examination via confusion matrices. These findings underscore the efficacy of advanced feature extraction techniques, especially in differentiating between diseases with visually analogous symptoms. Future research may explore the integration of deep learning architectures with handcrafted features in hybrid models to enhance accuracy and improve adaptability in practical agricultural applications.