Smart Agricultural Weed Management System

This repository contains the official code and resources for the research presented in the following papers:

1. "Quantized YOLO for Real-Time Weed Detection: An Affordable Embedded System for Precision Agriculture"
   Proceedings of SBIAGRO 2025

2. "Real-Time Weed Detection on Low-Cost Embedded Devices Using Quantized YOLO"
   2025 16th IEEE International Conference on Industry Applications (INDUSCON)

Abstract

This research addresses challenges in precision agriculture through the development of an affordable embedded system for real-time weed detection. By applying advanced deep learning optimization techniques to YOLOv5 neural networks deployed on edge computing platforms, we demonstrate how agricultural computer vision can be made accessible for producers of all sizes. Our methodology combines post-training quantization (16-bit floating point, 8-bit integer) and ONNX format conversion with systematic performance benchmarking on the Raspberry Pi 4B platform.

Key Features

• Affordable Hardware: Designed for Raspberry Pi 4B (~$60), democratizing precision agriculture.
• Deep Learning Optimization: Utilizes Quantization (FP16, INT8) and ONNX for real-time performance.
• Agricultural Relevance: Trained on the 4Weed Dataset covering Cocklebur, Foxtail, Redroot Pigweed, and Giant Ragweed.
• Real-Time Performance: Achieves up to 2.66 FPS (ONNX) on RPi 4B, suitable for tractor speeds up to 10 km/h.

Methodology

Dataset

We used the 4Weed Dataset containing 618 images of four weed species:

1. Cocklebur (Xanthium Strumarium)
2. Foxtail (Setaria Viridis)
3. Redroot Pigweed (Amaranthus Retroflexus)
4. Giant Ragweed (Ambrosia Trifida)

The dataset was split into training (518) and validation (100) sets.

YOLOv5 Architecture

We explored YOLOv5n, s, m, l, and x models. Training was conducted at 640x640 resolution for 150 epochs.

Optimization Techniques

1. FP16: 16-bit floating point quantization.
2. INT8: 8-bit integer quantization (Post-Training Quantization).
3. ONNX: Open Neural Network Exchange format.

Results

Accuracy (mAP@0.5) vs Speed (FPS) on Raspberry Pi 4B

Model	Format	mAP@0.5	FPS	Inference (s)
YOLOv5n	ONNX	0.727	2.66	0.376
YOLOv5s	ONNX	0.785	1.15	0.869
YOLOv5l	INT8	0.815	0.18	5.685

Conclusion:

• YOLOv5n-ONNX is best for real-time mobile applications (spraying at ~10 km/h).
• YOLOv5l-INT8 is best for high-accuracy stationary analysis.

Getting Started

Prerequisites

• Python 3.8+
• Git

Installation

1. Clone the repository:

   git clone https://github.com/LuizCMarquesJr/Smart-Agricultural-Weed-Management-Real-Time-Detection-System-Using-Quantized-YOLO.git
   cd Smart-Agricultural-Weed-Management-Real-Time-Detection-System-Using-Quantized-YOLO

2. Install dependencies:

   pip install -r requirements.txt

3. (Optional) Clone YOLOv5 submodule if you plan to retrain:

   git clone https://github.com/ultralytics/yolov5
   cd yolov5
   pip install -r requirements.txt
   cd ..

Usage

1. Dataset Preparation

Run the dataset preparation notebook to download the 4Weed dataset and organize it for YOLOv5:

• notebooks/01_Dataset_Preparation.ipynb

2. Training

To reproduce the training results:

• notebooks/02_YOLOv5_Training.ipynb

3. Quantization & Export

To convert trained models to ONNX and apply quantization:

• notebooks/03_Quantization_and_Export.ipynb

Citation

If you use this work, please cite our papers:

SBIAGRO 2025

@article{marques2025quantized,
  title={Quantized YOLO for Real-Time Weed Detection: An Affordable Embedded System for Precision Agriculture},
  author={Marques Jr, L. C. and Papa, J. P. and Ulson, J. A. C.},
  journal={Proceedings of SBIAGRO},
  year={2025}
}

IEEE INDUSCON 2025

@inproceedings{marques2025realtime,
  title={Real-Time Weed Detection on Low-Cost Embedded Devices Using Quantized YOLO},
  author={Marques Jr, L. C. and Ulson, J. A. C.},
  booktitle={2025 16th IEEE International Conference on Industry Applications (INDUSCON)},
  pages={1--6},
  year={2025},
  organization={IEEE},
  doi={10.1109/INDUSCON66435.2025.11241472}
}

Related Repositories

This project is part of a broader research initiative detailed in the PhD Thesis. For additional scripts, implementations, and the full thesis context, please refer to:

• Phd-Thesis-Embedded-Yolov5: Contains the comprehensive scripts and implementation of "Redes Neurais Profundas com Redução de Pesos para uso em Hardwares de Baixo Custo Aplicadas à Identificação de Espécies Invasoras em..."

License

This project is licensed under the GPL-3.0 License.