Detection and Classification of Diseases of Cassava Leaves Using Deep Neural Networks and Various Optimization Algorithms on Large Image Datasets

Cassava is a critical staple crop for millions globally and the basis of food security for many countries, but its production is severely constrained by viral and bacterial diseases. The accurate and on-time leaf disease detection allows us to deal with this issue quickly and efficiently. In this paper, we propose a new framework for detecting and classifying leaf diseases using advanced deep neural networks: YOLOv9 for leaf detection and EfficientNet V2 for disease classification. The Adam W optimizer is used to enhance training efficiency and speed. Our method is tested on a large public dataset of 20,000 cassava leaf images and a new dataset of 9000 images carefully collected in Vietnam. Experimental results demonstrate that our model achieves an [Formula: see text] score of 96.15% on the public dataset, outperforming MobileNet V3 (92.15%) and DenseNet-121 (93.5%). On the Vietnamese dataset, our algorithm still achieves a higher [Formula: see text] score than others. The novelty of this work is the first application of these emerging models to classify cassava leaf diseases, integrating the AdamW optimization algorithm in the training process and providing a new dataset for comprehensive evaluation. The performance of our proposed method is compared to various mainstream methods to demonstrate their practical applicability.