Citrus diseases detection and classification based on efficientnet-B5

Abstract

The detection and classification of citrus diseases are critical for ensuring the health and productivity of citrus fruits. This research focuses on enhancing the accuracy and effectiveness of citrus disease detection and classification using the EfficientNet-B5 model. The dataset, sourced from Kaggle, includes images of various citrus diseases: black spot, canker, huanglongbing (greening), and healthy instances. To address class imbalance and improve data diversity, synthetic minority oversampling technique (SMOTE) and augmentation fusion were employed, resulting in 970 images per class. The preprocessed data were partitioned into training, validation, and test sets. The efficientNet-B5 model was trained and validated, achieving a remarkable accuracy of 99.22 %. The study also includes a comprehensive comparison with existing systems based on accuracy and loss curves, confusion matrices, and classification reports. The proposed system demonstrated superior performance, outperforming other models in terms of both accuracy and robustness, highlighting its potential for practical applications in citrus disease management.