AI-Driven Smart Farming for Automated Plant Health Monitoring and Nutrient Deficiency Detection

Abstract

Precision agriculture is transitioning to continuous, data-driven monitoring. Affordable sensors and edge intelligence enable near real-time crop oversight. Manual scouting is labor-intensive and inconsistent. Delayed detection of nutrient stress leads to increased yield loss and input waste. Legacy systems monitor soil or weather in isolation and depend on periodic human checks. Vision pipelines often use shallow models and are not connected to field actuators. Proposed work: We combine soil moisture, temperature, and humidity sensing with camera-based leaf analysis using a DenseNet 121 classifier. A microcontroller executes closed-loop irrigation and localized cooling, with a mobile app for telemetry and alerts. RGB leaf images were captured in field conditions and labeled by experts into healthy and deficiency classes. Images were resized to 224×224 and split into training, validation, and test sets by plot to avoid leakage. DenseNet 121 achieved 89.0% accuracy on a held-out test set and surpassed a MobileNet V2 baseline of 82.0% under identical training conditions. Prototype deployments reduced manual checks and improved response to moisture and heat stress. The integrated IoT and AI pipeline is practical for early detection of nutrient deficiencies and autonomous actuation in small plots and greenhouses.

Plant phenotyping relevance

葉画像から植物の健康状態・栄養欠乏を推定する画像解析手法を開発し、データ分割、ベースライン比較、独立テストで性能評価しているため、フェノタイピング手法が中心的です。

Code and data availability

The article describes a 1,200-image RGB leaf dataset, DenseNet-121/MobileNetV2 models, and an IoT pipeline, but contains no data or code availability statement, no public repository, and no author-provided URL for any dataset, images, code, or trained model. Only a vague note that a confusion matrix 'can be provided in

No evidence-backed public reproduction asset is currently recorded.