EfficientNet-Based Cassava Leaf Disease Classification with Advanced Augmentation for Enhanced Accuracy

In this paper, a deep learning approach for detecting cassava diseases from high resolution images is presented. The dataset, sourced from the Cassava Leaf Disease Classification competition on Kaggle, includes five disease categories: Cassava Mosaic Disease, Cassava Brown Streak Disease, Cassava Green Mottle Disease, Bacterial Blight and Healthy leaves. EfficientNetB3 architecture and sophisticated data augmentation (random cropping, flipping, rotation and contrast adjustment) is utilized to increase robustness and generalization. With stratification, the dataset is split into training, validation, and test subsets to guarantee balanced class representation. The model trained for eight epochs and achieved 88.5% validation accuracy, which proves it could be used for disease classification. Performance metrics like precision, recall and F1 score are analyzed, the model shows high performance in classifying healthy leaves. The results of this research demonstrate the ability of AI to support precision agriculture and provide a foundation for scalable cassava disease detection solutions.