Research on Intelligent Recognition and Location Method of Crop Diseases Based on Multi-spectral Images of Unmanned Aerial

Abstract

This paper studies the intelligent identification and location method of crop diseases based on multispectral images of unmanned aerial vehicles. With the development of precision agriculture, traditional crop disease monitoring methods have become difficult to meet the demands of large-scale, high-efficiency and early warning. The article first constructs a multispectral image dataset including visible light, near-infrared and red-edge bands, covering common types of crop diseases. Subsequently, an improved deep learning network architecture was proposed. The attention mechanism was adopted to enhance the model's ability to extract disease features, and a multi-scale feature fusion strategy was introduced to handle disease spots of different sizes. The research designed a data augmentation method based on spectral-spatial joint optimization, which effectively solved the problem of unbalanced samples of crop diseases. To improve positioning accuracy, this paper proposes a disease area positioning algorithm combined with geographic information system, achieving centimeter-level positioning accuracy. The experimental results show that the proposed method improves the accuracy of disease identification by 15.3% compared with the traditional methods, reduces the positioning error to an average of 3.2 centimeters, and can maintain high stability in complex field environments. In addition, this paper has established a complete technical system covering data collection, disease identification and information visualization, and has conducted application verification on crops such as wheat and rice. It has been confirmed that this method can effectively support precise pesticide application decisions in agricultural production and has significant economic and ecological benefits

Plant phenotyping relevance

マルチスペクトル画像から作物病害の特徴・病斑領域を抽出する認識および位置推定手法の開発、検証、実地適用が研究の中心であり、植物の病害状態を直接評価している。

Code and data availability

The supplied blocks describe a UAV multispectral crop-disease dataset (~10,000 images) and models (MS-ResNet, SA-UNET, FS-ProtoNet), but contain no data or code availability statement, no public repository, and no authors' URL for the dataset, images, models, or analysis code. All URLs present are references to cited,

No evidence-backed public reproduction asset is currently recorded.