metadata

license: cc0-1.0

River Bollin Pollution Detection Dataset

Overview

This dataset contains images captured from the River Bollin in Macclesfield, UK. The images are to provided for anyone wanting to practice CV ML development using a typical real world dataset. The plan is to deploy your model to identify pollution incidents in real-time. The dataset is highly imbalanced.

Dataset Description

Total Images: Approximately 60,000
Pollution Incident Images (Positive Class): Approximately 50
Image Dimensions: 224x224 pixels
File Format: JPG
Filename Convention: Filenames are based on Unix timestamps indicating when each image was captured.

File Structure

imgs/ - Directory containing all the image files.
pollution_incidents.txt - A text file listing the filenames of images that contain observed pollution incidents (positive class).

Purpose

The primary goal of this dataset is to train a machine learning model that can identify pollution incidents in real-time. Given the imbalanced nature of the dataset, anomaly detection techniques are recommended.

Getting Started

Understand the Dataset

Images: Located in the imgs/ directory. Each image is a 224x224 pixel JPG file.
Pollution Incident List: The pollution_incidents.txt file contains filenames of images that have observed pollution incidents.

Model Training

While this README does not provide specific code, it is recommended to:

Explore convolutional neural networks (CNNs) for image classification.
Use anomaly detection techniques to handle the imbalance.

Evaluation

Evaluate your model's performance using appropriate metrics. Due to the imbalance, consider metrics beyond accuracy, such as precision, recall, and F1-score.

License

This dataset is provided under an open license. You are free to use, modify, and distribute it. For more details, visit Open License.

Acknowledgments

Special thanks to the contributors and the Friends of the River Bollin for providing the data.