A Deep Ensemble Approach for Recognition of Papaya Diseases using EfficientNet Models

Diseases of papaya impeded quality production and caused severe financial damages to growers. An efficient diagnosis approach for papaya diseases is enormously desired to control and prevent the spread of diseases. At first, using a dataset of 138980 images of affected and healthy leaves and fruits of papaya which was generated with image augmentation techniques from 13898 collected images, eight models of EfficientNet between B0 and B7 were trained via transfer learning technique to recognize eight diseases. Afterward, fine-tuned versions of the three best-performing models were selected for ensemble learning such as EfficientNet B5, B7, and B6, which achieved 98.13%, 96.93%, and 96.87% accuracy under the test set of 6931 images, respectively. The deep ensemble model showed more effective recognition performance than single models, and test accuracy increased by 1.61%. The experimental result demonstrates that the proposed ensemble model can recognize papaya diseases more efficiently than single models of EfficientNet.