// IEEE Research · ICFCR 2026

YOLOv8
Micro
plastics

Deep learning framework for automated detection, localization, and quantification of microplastics in microscopic freshwater images — submitted to IEEE ICFCR 2026.

// mAP@0.5

0.743

Mean Avg. Precision

// Precision

0.761

Recall: 0.685

// Inference Time

47.4ms

25 MB model size

// ROC AUC

0.85

751 images · 50 epochs

// Overview

Overview

Developed a YOLOv8-based deep learning pipeline for automated detection and quantification of microplastics in freshwater microscopic images. Trained on 751 labeled samples with an 80:20 train-test split, the model outperforms SSD, RetinaNet, Faster R-CNN, and prior YOLO versions in inference speed and model compactness while maintaining competitive detection accuracy. An interactive Streamlit dashboard provides real-time visualization and quantification. Co-authored with S Sakthi Nikkithan, supervised by Dr. Hariharan R at SRM Institute.

// Specs

Specifications

Model

YOLOv8n (Nano variant)

Dataset

751 labeled microscopic freshwater images

Split

80% Train / 20% Test

Training

50 epochs · CIoU Loss · NMS post-processing

Model Size

25 MB

Conference

IEEE ICFCR 2026 (Submitted)

Plagiarism

~6% similarity — within IEEE norms

// Benchmarks

Model Comparison

Model	mAP@0.5	Precision	Inference (ms)	Size (MB)
YOLOv8n Ours	0.743	0.761	47.4	25
YOLOv5	0.698	0.712	61.2	42
SSD	0.621	0.648	89.3	98
RetinaNet	0.657	0.671	112.4	145
Faster R-CNN	0.714	0.738	203.7	162

// Features

Features

Real-time microplastic detection and bounding-box localization in microscopic images

CIoU loss function for precise bounding box regression and shape-aware detection

NMS post-processing to suppress overlapping detections and improve precision

Interactive Streamlit dashboard for real-time visualization and particle quantification

Custom LabelImg-annotated dataset of 751 freshwater microscopic images

Training loss curve analysis and model evaluation with ROC-AUC, mAP, precision/recall metrics

// Tech Stack

Tech Stack

YOLOv8n Python Ultralytics Streamlit LabelImg CIoU Loss NMS Google Colab OpenCV NumPy Matplotlib IEEE LaTeX

Interested
in this
Research?

Want to discuss the methodology, explore the dataset, or collaborate on environmental computer vision research? Let's connect.

Get In Touch →

YOLOv8Microplastics

YOLOv8
Micro
plastics