AI based plant stress detector

AI-POWERED PLANT STRESS DETECTION SYSTEM

★Project Domain: Artificial Intelligence, Machine Learning & IoT ★Project Type: Engineering / Student Project ★Project Status: Prototype Development ★Month: September 2026

1. COMPREHENSIVE DOCUMENTATION ★ Introduction Plants are essential for agriculture, food production, and maintaining the environment. However, plants can be affected by different types of stress such as water deficiency, nutrient deficiency, diseases, and pest attacks.

These problems often produce visible changes in leaves, including yellowing, spots, discoloration, drying, and damaged regions. Identifying the exact cause of these symptoms through manual observation can be difficult and time-consuming.

To address this problem, our team proposed LeafScan, an AI-assisted portable plant stress detection system.

The system uses an ESP32-CAM to capture images of plant leaves. A WS2812B LED ring is proposed to provide controlled and uniform lighting, while an LDR sensor helps monitor the lighting conditions. The captured image is transmitted through Wi-Fi to a processing system, where an AI model can analyze the image.

The proposed system aims to classify the plant condition into five categories: •Healthy •Water Stress •Nutrient Deficiency •Disease •Pest Damage

If the AI model is not sufficiently confident in its prediction, the system can provide an “Unclear – Retake Photo” response.

★ Problem Statement Early identification of plant stress is difficult because different stress conditions can produce similar visual symptoms. Manual inspection can also be time-consuming and may require agricultural knowledge. Therefore, there is a need for a portable and easy-to-use AI-assisted system that can capture plant leaf images and provide a preliminary indication of possible plant stress. Problem Statement

To develop a portable AI-assisted plant stress detection system that captures leaf images and analyzes them to identify possible water stress, nutrient deficiency, disease, pest damage, or healthy conditions while providing the result in a simple and understandable manner.

★Objectives The main objectives of our project are: •To develop a portable plant stress detection prototype. •To capture plant leaf images using an ESP32-CAM. •To provide controlled illumination using an LED ring. •To monitor lighting conditions using an LDR sensor. •To transmit captured images through Wi-Fi. •To explore CNN/MobileNetV2-based image classification. •To classify leaves into five proposed categories. •To display system status and results using an OLED display. •To provide visual and audio feedback using LEDs and a buzzer. •To explore Grad-CAM for explaining which regions of a leaf influenced the AI prediction.

★ Proposed System The proposed LeafScan system consists of two major sections:

#Hardware Section •ESP32-CAM •ESP32 DevKit •OV2640 camera •WS2812B LED ring •LDR sensor •OLED display •Push button •Status LEDs •Buzzer •Portable power supply

#Software and AI Section •Arduino IDE •ESP32 firmware •Wi-Fi communication •Python-based processing •CNN / MobileNetV2 •Image classification •Confidence threshold •Grad-CAM visualization •GitHub for code and documentation

★ Working Methodology The proposed workflow is: Leaf → Image Capture → Lighting Check → Wi-Fi → AI Analysis → Classification → Result

#Step 1: System Initialization When the device is powered on, the ESP32 initializes the connected components such as the camera, OLED, sensors, LEDs, and buzzer.

#Step 2 Lighting Check The LDR sensor checks the surrounding lighting condition.

#Step 3 Controlled Illumination The WS2812B LED ring provides consistent illumination around the leaf to reduce variations caused by environmental lighting. #Step 4 Leaf Image Capture The ESP32-CAM captures an image of the plant leaf using the OV2640 camera.

#Step 5 Wireless Transmission The captured image is transmitted through Wi-Fi to the processing device. #Step 6 AI Analysis The image is processed using the proposed AI classification pipeline. A CNN or MobileNetV2-based model can be trained to recognize different visual patterns associated with the selected plant-stress categories.

#Step 7 Classification The system attempts to classify the leaf as: Healthy / Water Stress / Nutrient Deficiency / Disease / Pest Damage

#Step 8 Confidence Check If the model confidence is below the selected threshold, the system can display: UNCLEAR – RETAKE PHOTO This helps avoid presenting an uncertain prediction as a definite result.

#Step 9 Result Display The result can be communicated using the OLED display, LEDs, buzzer, and software interface.