Ensemble-Based Transfer Learning for Multi-Class Plant Disease Detection Using VGG16, ResNet50, and Xception Models

In the world, plant diseases pose a serious threat to agricultural productivity and food security. Early, accurate, and rapid identification of plant diseases is important for con-trolling loss of crops. In the following research, transfer learning models VGG16, ResNet50, and Xception are applied to attempt overcoming this challenge of multiclass plant disease detection. To improve classification accuracy, we propose an ensemble model that combines the strengths of these pre-trained networks. Multiple plant species and disease categories were experimented on extensively on publicly available plant disease datasets. The results show that ensemble model achieves better precision, precision and recall than individual models and therefore presents a robust solution for identifying several plant diseases together as a pack. Results from the experiment demonstrate that the proposed method could be deployed in real-time agricultural systems and have potential to provide a scalable and efficient diagnostic tool for farmers and agronomists to detect plant diseases and reduce their impact. This work is among the growing body of work in AI based agricultural solutions and indicates that transfer learning and ensemble techniques are promising in precision farming.