Experimenting Data Augmentation Using Image Data Generator and OpenCV for Melon Leaf Diseases Classification

Melon cultivation is significantly affected by various leaf diseases that hinder productivity and crop quality. Accurate identification of melon leaf conditions-such as healthy leaves, Gemini virus infection, Liriomyza infestation, and pesticide overdose-is challenging due to variations in imaging conditions including differences in distance, tilt angle, natural rotation, and lighting. Such variability can adversely affect the accuracy of image-based classification models. To address this, data augmentation techniques play a crucial role in increasing dataset diversity and enhancing the robustness of Convolutional Neural Network (CNN) models against visual inconsistencies. This study experimentally evaluates the effectiveness of three data augmentation techniques: TensorFlow ImageDataGenerator (IDG), OpenCV, and a hybrid combining both methods, in classifying melon leaf diseases. After preprocessing, resizing, and augmentation, datasets were utilized to train CNN models, evaluated using accuracy, precision, recall, and F1-score metrics. Experimental results demonstrate that augmentation with TensorFlow ImageDataGenerator (IDG) achieves the highest classification performance, yielding an accuracy of 95.31 %. These findings highlight the importance of selecting appropriate data augmentation techniques to optimize CNN performance, providing valuable insights for automated and precise disease diagnostics in precision agriculture.