Intelligent Drone Assisted Plant Disease Detectionand Precision Pesticide Spraying Using a Residual-Inception Deep Learning Framework

Abstract

Abstract Deep learning techniques combined with unmanned aerial vehicles (UAVs) create new possibilities which will transform smartagricultural practices. This project introduces an innovative system which merges a Residual-Inception convolutional neuralnetwork with an intelligent drone spraying system to enable early detection and targeted treatment of plant diseases. Theresearch used a dataset which contained 38 crop disease classes and achieved a 94.5% validation accuracy after 10 trainingepochs. The proposed model combines multi-scale convolutional branches with residual shortcuts to improve feature learningwhile decreasing classification errors. The researchers created a mathematical waypoint generation algorithm which enablesUAVs to navigate through fields by creating efficient field coverage paths with optimal spray distribution. The researchersdeveloped a pesticide prioritization system which maps detected diseases to customized treatment plans based on diseaseseverity. The simulated results show precise classification results which successfully plan waypoints and create clear visualdisplays of the spraying operations. This Study demonstrates a scalable and economical method for executing precisionagriculture because it decreases chemical application while enhancing crop health management.

Plant phenotyping relevance

植物病害を対象とする深層学習による検出・分類手法が中心的に開発されており、植物の病害状態を直接推定するため、病害フェノタイピング手法として収録する。

Code and data availability

The paper uses the public New Plant Diseases Dataset (Kaggle) as its phenotyping image input, but the only URL provided for it is a Kaggle link not present in the allowed_urls list, so it cannot be listed as an actionable public asset. The authors' trained model and implementation scripts are available only uponrequest

No evidence-backed public reproduction asset is currently recorded.