amateurmind/microplastic-detection

Edge AI-Powered Smart Microscope for Microplastic Detection in Water

This repository provides the dataset and resources for an AI-enabled Two-Stage Microplastic Detection and Monitoring System designed for optical smart microscopy and optimized for Edge AI platforms (NVIDIA Jetson Nano / Raspberry Pi).

🎓 Academic Project: Dataset component for the Bachelor of Engineering (B.E.) Final Year Capstone Project:
"Edge AI-Powered Smart Microscope for Microplastic Detection in Water".

GitHub Repository: 🔗 AmateurMind/Microplastic_rasberry_jetsonnano

---

🎯 Two-Stage Pipeline Architecture

To eliminate label conflicts (where some datasets label generic microplastic while others annotate specific classes), the system decouples detection from classification:

1. Stage 1 (Detector): YOLO11n trained on 1 unified class (microplastic) for fast, high-recall particle localization (82.4% mAP@50, 4.7ms latency).
2. Stage 2 (Classifier): High-resolution crop classified into specific morphologies:
   • fiber
   • film
   • fragment
   • pallet
   • (99.79% validation accuracy via SVM / Random Forest)

---

📦 Dataset Package Structure

The archive merged_dataset.zip (~75 MB) contains standard YOLO format splits:

• train/images/ & train/labels/ (1,725 images)
• valid/images/ & valid/labels/ (322 images)
• test/images/ & test/labels/ (55 images)
• data_1class.yaml (Recommended 1-class config)
• data.yaml (Multi-class config)

---

🚀 Quick Download & Usage

1. Download & Extract using Python

from huggingface_hub import hf_hub_download
import zipfile

# Download dataset archive
zip_file = hf_hub_download(
    repo_id="amateurmind/microplastic-detection",
    filename="merged_dataset.zip",
    repo_type="dataset"
)

# Extract
with zipfile.ZipFile(zip_file, 'r') as zip_ref:
    zip_ref.extractall("microplastic_dataset")

2. Run Pre-trained Detection & Classification

See full code and instructions on GitHub:
👉 GitHub Repo