COMBINED SYSTEM FOR PLANT DISEASE MONITORING USING UAVS, GROUND ROBOTIC PLATFORMS, AND NEURAL NETWORK-BASED IMAGE ANALYSIS

Abstract

This paper addresses the problem of automated monitoring of agricultural crop diseases under field conditions using unmanned aerial vehicles. It is shown that most existing solutions are primarily focused on disease detection from individual images. In contrast, issues such as further diagnosis refinement, repeated inspection of problematic areas, and subsequent action determination after disease detection are considered much less frequently. Stem-type diseases, in particular sclerotinia, pose an additional challenge, as top-view imaging alone may not detect early infection signs promptly. On this basis, a proposed automated system combines a UAV for primary inspection, subsequent georeferencing of the point of interest, a ground module for additional follow-up inspection in the case of stem-type diseases, and a central computing module for data processing and decision-making regarding the treatment of infected areas. This approach enables combining rapid aerial inspection of large areas with more accurate follow-up inspection of plants from a side view, which is especially important for diagnosing lesions that are poorly visualized from above. As part of the experimental study, a prototype classifier based on the EfficientNetV2S convolutional neural network and the transfer learning approach was implemented. To improve training quality, image augmentation and a pseudo-negative sample generation method were applied. The obtained results confirmed the potential of convolutional neural networks for automated plant condition classification, as well as the feasibility of combining aerial imaging and ground-based follow-up inspection within a unified monitoring system. The proposed approach can serve as a basis for the further development of an intelligent system for the detection and localized treatment of disease foci under field conditions within the framework of precision agriculture.

Plant phenotyping relevance

UAV・地上ロボット・画像解析を統合し、植物病害の病変・状態を画像から分類するシステムの開発が中心であり、植物の疾病状態を対象とする実質的なフェノタイピング手法である。

Code and data availability

The paper describes a UAV+ground-robot disease monitoring concept and an EfficientNetV2S classifier trained on a 350-image sunflower/sclerotinia dataset, but no public dataset, image repository, code deposit, or model checkpoint is mentioned anywhere in the supplied blocks. No availability statement or authors' URL for

No evidence-backed public reproduction asset is currently recorded.