AgroVision is an AI-IoT solution developed to transform greenhouse and agricultural monitoring. The system features an autonomous mobile platform (robotic vehicle) equipped with a camera that navigates through crop rows to perform real-time plant analysis and disease detection using a custom-trained Convolutional Neural Network (CNN).
This project was proudly represented by the Benha National University (BNU) team during the G-Force Round 2 Scientific Competition at Galala University.
During our extensive testing, we found that the ESP32-CAM sensor quality, high noise levels, and sensitivity to poor lighting conditions can significantly degrade the AI model's accuracy.
Recommendation: For production-grade or highly reliable results, we strongly recommend using a Raspberry Pi Camera (e.g., Camera Module v2) paired with a Raspberry Pi board. The superior image clarity and sensor stability of the Raspberry Pi Cam provide a much more consistent input for the CNN model compared to the ESP32-CAM.
Manual crop inspection is often slow and inaccurate. AgroVision solves this by integrating:
- Autonomous Navigation: A mobile chassis designed to traverse agricultural environments.
- AI Diagnostic Engine: A MobileNetV2-based CNN model capable of identifying various plant diseases with high precision.
- Actionable Agriculture: Instant identification of disease types to assist farmers in early intervention.
- Artificial Intelligence: TensorFlow, Keras, MobileNetV2 (Transfer Learning), Python.
- Embedded Systems: ESP32-CAM, Motor Drivers, Servo Motors.
- Data Analysis: Jupyter Notebooks (Google Colab).
arduino/: Firmware for the ESP32-CAM and vehicle motor control logic.notebooks/: IncludesAgroVision.ipynbcontaining the full training pipeline, data augmentation, and evaluation metrics.
- Event: G-Force Round 2 - Scientific Competition.
- Host: Galala University.
- Team Representation: Benha National University (BNU).
- Mahmoud Mohamed Mahmoud Abdeltawab
- Ahmed Mustafa Elbatanouny
- Ahmed Hassan Saleh
Supervised by:
- Prof. Mohamed Said
We would like to extend our deep gratitude to Mohamed Badawy for his invaluable technical guidance and support throughout the development of this project.
- GitHub: M7MEDpro
This repository provides the core training pipeline to generate the AgroVision.h5 model. Once you have generated the model file, deployment is flexible based on your needs:
- API Server Integration: You can host the
.h5model on a cloud server (e.g., using FastAPI and Ngrok) to perform remote inference from any device. - Local Usage: You can load the model directly into a local Python script or a custom GUI application for on-device processing.
Note: This repository currently includes the training and hardware logic. Custom integration scripts or GUI wrappers are left to the user's preference and specific use case.
Developed with passion by the Engineering students of Benha National University.