Improving Plant Disease Detection Using Super-Resolution Generative Adversarial Networks and Enhanced Dataset Diversity

Abstract

The detection of plant diseases is critical for maintaining crop health and maximizing agricultural yields. This research proposes a comprehensive approach to improve plant disease detection by addressing challenges related to unbalanced datasets and leveraging generative adversarial networks (GANs). This research focuses on enhancing the accuracy and generalization capabilities of disease recognition models. To address dataset bias, a larger and more diverse dataset is collected, comprising unhealthy plant leaves from various plants, regions, and disease types. The expanded dataset enables comprehensive training and validation, ensuring a representative depiction of leaf variations and diseases. Domain-specific knowledge and expert guidance are incorporated to capture realistic and characteristic attributes of diseased leaves. To overcome overfitting, the regularization technique is applied during training. These techniques promote the learning of generalized representations and mitigate the generation of unrealistic or repetitive images. The proposed approach is extensively evaluated using Plant Village dataset encompassing various plant species, and disease types. By implementing these solutions, this research enhances the accuracy, robustness, and generalization capabilities of plant disease detection systems. It establishes a foundation for reliable and effective detection methods, contributing to the sustainable management of plant diseases and improved agricultural outcomes.

Plant phenotyping relevance

植物葉の病害状態を画像から検出する手法の改善が研究の中心であり、データセット拡張、GAN、正則化、評価を含むため、植物フェノタイピング手法として採用する。

Code and data availability

植物フェノタイピング解析を再現する公開資産であることを、入力本文と直接リンクから確認できなかったため保留しました。

No evidence-backed public reproduction asset is currently recorded.