Leveraging Feature Extraction and Machine Learning for Effective Plant Disease Classification

Disease prediction in crops is crucial for several reasons, including its impact on farming practices, food availability, and the global economy. This study focuses on predicting severe diseases in pear and watermelon leaves by applying feature extraction (FE) techniques to small image datasets and then using machine learning algorithms for classification. Deep learning models typically require large datasets and substantial computational resources, which makes them less suitable for small image datasets. In contrast, feature extraction combined with machine learning offers a more efficient solution, requiring less computation time and power. The study extracts various color, HOG and texture features such as dissimilarity, energy, contrast, angular second moment, correlation, and homogeneity from pixel pairs at different angles (0°, 45°, 90°, and 135°) across all three color channels, resulting in a total of 2994 features. The feature dataset is then balanced using the SMOTE technique and processed with five machine learning algorithms: Decision Tree, k-Nearest Neighbor, Gaussian Naïve Bayes, Random Forest, and XGBoost ensemble. The results demonstrate that the proposed model FE+SMOTE+XGB achieves optimal accuracy, with 93.4% for the pear dataset and 96.8% for the watermelon dataset, outperforming the state-of-the-art. Early disease prediction enables timely interventions, helping to reduce significant yield losses and ensuring adequate crop quantities.