Enhanced Orange Disease Detection and Classification with DenseNet121: Advancing Deep Learning for Precision Agriculture

This research aims to enhance orange disease detection through a deep learning-based approach for the classification of orange diseases into four classes: Blackspot, Canker, Fresh, and Greening Citrus. By using the DenseNet121 architecture, this work tackles challenges of proper identification and efficiency in the process, which is one of the most crucial factors to improve agricultural productivity and management. A total of 1,164 images were drawn from Kaggle, divided 80–20 to ensure an optimal performance evaluation. Further preprocessing involves normalizing these images, which will prevent class imbalance and improve the generalizability of the network. Early stopping was utilized during training to avoid overfitting; hence, the model is robust. It was with a 97% accuracy rate, which gave high precision and recall over all classes but especially performed very well on Blackspot and Greening Citrus. However, slight misclassifications occurred in the Canker class; this showed that further work was needed with respect to data augmentation or refining the feature extraction technique. The confusion matrix and classification report provided showed the model performance of being reliable and balanced across all metrics for each category. This research demonstrates that deep learning can be used as a viable method for automatic orange disease detection and thus can serve as a scalable solution for real-time diagnostics in agricultural settings. Further studies could be conducted on more sophisticated architectures and data augmentation strategies to achieve higher classification accuracy for supporting precision agriculture practices. These findings pave the way for the development of practical, user-friendly applications for farmers, enabling timely disease interventions and sustainable crop management.