Towards smart farming: a real-time diagnosis system for strawberry foliar diseases using deep learning.

Abstract

Background Developing an effective machine vision system is crucial to successfully deploying robotic inspection in open field conditions and controlled environments like greenhouses. Robotic arms with vision-based deep learning models offer an efficient, real-time, non-invasive crop monitoring solution. In agricultural settings, they enable consistent, automated inspection under varying conditions, reduce labor dependency, and support early disease detection, enhancing productivity and sustainability in precision farming. Although considerable progress has been made in computer vision-based approaches, significant challenges persist in developing models that reliably perform under the diverse and variable conditions encountered in real-world agricultural settings. Method Within the domain of precision agriculture, we introduce an advanced robotic system for the detection of plant diseases, utilizing an innovative model based on deep learning principles. This system introduces an algorithm for real-time analysis, called as Strawberry Leaf Disease Inspection (SLDI). The algorithm integrates the use of Receptive Guided Channel Attention (RGCA) alongside a Deep Context Aggregator (DCA), designed to significantly improve the characterization and representation of feature sets, thereby enhancing the overall accuracy and efficiency of disease identification. To optimize the system performance and preserve real-time performance, a Multi-Scale Feature Fusion Module (MSFF) is proposed that facilitates a comprehensive multi-level representation, enabling the model to capture disease symptoms promptly. The SLDI algorithm is deployed on a robotic platform equipped with an RGB camera, enabling real-time, in-field inspection of strawberry crops. Results The proposed system is trained on two publicly available datasets, PlantDoc and PlantVillage. It attains a precision of 91.10% and a recall of 88.50%, while maintaining a real-time processing speed of 76.50 frames per second (fps). Experimental field inspection of strawberry studies demonstrates that the proposed model significantly outperforms existing approaches in accuracy and efficiency.

Plant phenotyping relevance

イチゴ葉の病徴をRGB画像と深層学習でリアルタイム検出するアルゴリズムおよびロボットプラットフォームを開発・評価しており、植物病害状態の表現型取得が中心である。

Code and data availability

保存済みの本文根拠を更新済みルールで再検証し、公開資産1件を確認しました。