Integrating Real-ESRGAN with CNN Models for UAV Image Based Plant Disease Detection

Abstract

The integration of deep learning models with UAV captured images for plant disease detection has been explored in many papers and has the potential to revolutionize commercial precision agriculture, by allowing for early and efficient detection and classification of crop disease stages.In order to address the limitations posed by low-resolution aerial imaging, this paper proposes the additional integration of an Enhanced Super Resolution Generative Adversarial Network (ESRGAN) with a Convolutional Neural Network model for field monitoring through UAV captured imagery.UAVs are a cost effective method of monitoring large swaths of agricultural land; however, it is difficult to capture images of a high enough quality and clarity to be adequately analyzed by a CNN.The images typically lack the necessary resolution for accurate classification, especially for diseases with smaller, less noticable symptoms.The Real-ESRGAN model is employed to generate a dataset of high-resolution images, from low-resolution inputs, allowing the disease detection CNN to more accurately and effectively identify and classify disease stages in Armillaria afflicted cherry trees.This solution offers a solution to the problem posed by traditional UAV based approaches that enhances classification accuracy even in suboptimal conditions.Through this integrated approach, the model was able to reach an increased validation accuracy, as well as significantly decreased loss values due to the ESRGAN enhanced imagery allowing for clearer detection of early stage Armillaria symptoms.This integrated system provides a practical scalable solution for commercial agriculture, allowing for more comprehensive and efficient crop disease monitoring.Future research can be explored to optimize the architecture of this model and expand its applicability to other crops and environmental conditions, allowing more efficient precision agriculture and paving the way for more sustainable farming practices.

Plant phenotyping relevance

UAV画像から植物病害の症状・病期を推定するReal-ESRGANとCNNの統合手法を開発・評価しており、植物フェノタイピング手法が研究の中心である。

Code and data availability

The paper describes a CNN and fine-tuned ESRGAN applied to the Cherry Tree Disease dataset, but provides no authors' data or code availability statement. The dataset is from a cited prior study (Chaschatzis et al., Remote Sens. 12(3):531), and the Real-ESRGAN model is a generic third-party GitHub model, not a paper-own

No evidence-backed public reproduction asset is currently recorded.