AI-Enabled Sensor-Based Crop Stress Early Prediction and Detection System

Abstract: Timely identification of crop stress is critical to reducing agricultural losses and supporting sustainable food production. This paper presents an AI-enabled sensor-based platform that integrates Internet of Things (IoT) sensing, cloud-based data management, and deep learning inference to predict and detect crop stress at an early stage. Temperature, humidity, and soil moisture data are continuously acquired using a DHT11 sensor and a capacitive soil moisture sensor interfaced with an ESP32 microcontroller, and the readings are transmitted to a Firebase Realtime Database. A Flask-based backend retrieves the stored time-series records and feeds them to a trained Long Short-Term Memory (LSTM) network, which forecasts future soil moisture values. The system concurrently analyses moisture trend behaviour—classifying conditions as stable, gradually decreasing, or critically decreasing—and cross-references actual and predicted values against crop-specific thresholds to generate three-level stress alerts: Low, Medium, and High. Early- warning notifications are raised when predicted values approach critical limits, while High-stress alerts are triggered when readings fall below safety thresholds. An interactive web dashboard delivers live sensor telemetry, LSTM forecast outputs, trend indicators, and stress alert status in real time. Experimental evaluation confirms accurate prediction and reliable alert generation, demonstrating that the proposed system provides an affordable, scalable, and intelligent solution for modern precision agriculture.

Keywords: crop stress detection; LSTM time-series forecasting; IoT smart agriculture; soil moisture prediction; ESP32; Firebase; precision farming; early warning system