Assessing the Effectiveness of Deep Learning Approaches in Forecasting Cassava Leaf Diseases

In the Philippines, cassava-one of the most crucial crops-faces significant threats from various diseases, which can severely impact its cultivation and yield. To ensure food security, it is vital to detect these diseases early before they escalate. Disease detection methodologies are currently based on fuzzy logic and the expertise of farmers. This study proposes an automated system utilizing pretrained neural network models, specifically inspired by convolutional neural network techniques, to handle fuzzy data and identify diseases. The proposed system successfully identified anthocyanin and red leach using neural network models. An EfficientNetB3 model, featuring 10,791,220 learnable parameters, was trained with transfer learning to manage fuzzy data for cassava disease identification. The model achieved an accuracy of 89.90%, surpassing the baseline accuracy of 50.46%. This neural network classifier shows strong performance in identifying cassava leaf diseases, even with imbalanced data. The findings are beneficial for farmers and agricultural scientists, potentially enhancing crop yield and aiding in disease prevention. Our approach could automate disease identification processes, thereby reducing crop loss. We recommend further research projects aimed at scaling up plant disease detection efforts.