Data augmentation and transfer learning efficientnetv2-s on rice leaf disease classification

Abstract

Abstract Rice is a major food commodity in Indonesia, including in Merauke Regency, but its productivity is often threatened by leaf diseases such as bacterial leaf blight, brown spot, and leaf smut. This study aims to evaluate the effectiveness of combining data augmentation and transfer learning techniques using EfficientNetV2-S in classifying rice leaf diseases. The dataset consists of 120 images of infected rice leaves divided into 3 classes, which are expanded and enriched through augmentation techniques such as rotation and flipping. The EfficientNetV2-S model trained on the ImageNet dataset was used as the basis and fine-tuned. The training was performed using the Adam optimizer with an adjusted learning rate 0.0001. The evaluation results show that data augmentation can reduce overfitting and improve the model’s capability to an accuracy of 95% ± 2, which is an improvement of 29% ± 2 from the accuracy of the model trained without data augmentation techniques. In addition, using the right learning rate can also help improve the performance of the model in the training process.

Access to Document

Other files and links

• Link to publication in Scopus
• Open Access Version Available

Fingerprint