Comparative Analysis of Cassava Leaf Disease Prediction Using the Deep Learning Approach
Pratham Kaushik, Eshika Jain, Kanwarpartap Singh Gill, Deepak Upadhyay, Swati Devliyal
Abstract
There are several diseases in the Philippines that significantly threaten the cultivation and output of cassava, the most important crop. With food security dependent on cassava to sustain, detecting these diseases before they become more severe is indispensable. The methodologies, which are founded on the fuzzy domain-related product expert knowledge and the experience of the framers, were used in disease detection. In this work, the automatic system based on pretrained models of neural networks and deep learning, inspired by convolutional neural networks approaches is proposed to handle the fuzzy data for identifying diseases. As a result, anthocyanin and red leach were identified automatically using the proposed neural networks systems. Here, while dealing with fuzzy data for the identification of cassava diseases, a model on the EfficientnetB3 configuration, which contained 10,791,220 learnable parameters, was trained using transfer learning. The accuracy achieved with the proposed data model, including the same, was 89.90%, which is above the baseline proportional chance consideration of 50.46%. The neural network classifier demonstrates impressive results, even when the data is imbalanced, by accurately identifying cassava leaf disease. The results obtained can also be used to benefit farmers and agricultural scientists in addition to improving crop output. The proposed approach to detect crop diseases can help automate data capture for disease identification to prevent crop loss.