Dataset Card for microfibres_Glass_filter

A dataset of annotated images derived from wastewater sludge samples collected using fibreglass filters ("Glass dataset"), supporting microfibre detection and segmentation through deep learning. Each image is manually annotated to identify microfibres, their location, and area, and is designed for the development and benchmarking of computer vision models, especially for environmental monitoring applications.

Dataset Details

The wastewater sludge samples were collected from a wastewater treatment plant and subjected to a pretreatment process to eliminate the organic matter from the sludge matrix so that MFi could be better distinguished. The pretreatment involved the following steps: initially, total solids were measured, diluting the sample with osmotic water if this value exceeded 14 g/L; after that, the organic matter was oxidised with 30% hydrogen peroxide (H2O2) by mixing equal volumes of the sample and H2O2 for 2 hours at 60◦C. These operation conditions ensured that MFi do not experience any degradation, meanwhile around 98% of organic matter was oxidised. Temperatures between 60 − 70◦C improve the organic matter removal; however, it should not be exceeded 70◦C, as some plastics such as polyamide begin to degrade. Subsequently, the sample was filtered at 5 μm. Filtration can be carried out using fibreglass filter. Finally, the filters were placed in an oven-drying process for a period of two and a half hours. With this procedure, the MFi was retained above the filter, thus allowing them to be viewed using a Leica S APO Zoom 8:1x stereomicroscope. The stereomicroscope was connected to a computer running the LAS X1 application suite software. Upon the detection of a region of fibres, animage was captured and stored in a user-created folder with the selected magnification. The images may be captured at magnifications ranging from 10x to 80x.

Dataset Description

The dataset contains 1203 original high-resolution images (4000x3000 px), evenly split between light and dark backgrounds, reflecting different levels of residual solids. Images have various magnification markers (16x, 20x, 32x, 42x), indicated with scale bars. Microfibre annotations include the count and precise segmentation (polygonal masks), following guidelines similar to the COCO dataset. Each image is annotated with a JSON file listing bounding boxes and polygons for each fibre.

• Curated by: Universitat Politècnica de València, Spain

Dataset Sources [optional]

• Repository: https://github.com/femartip/Detection-Microfibres-in-Sludge
• Paper:
• Demo: https://ojs.aaai.org/index.php/AAAI/article/view/35366 , https://github.com/femartip/Detection_Microfibers_APP

Uses

Direct Use

Training and evaluating deep neural networks for microfibre detection, segmentation, and counting in microscopy images. Benchmarking of computer vision techniques (including Mask R-CNN, UNet) for segmentation accuracy and precision. Research into sources, prevalence, and morphology of microfibres in wastewater sludge. Academic use for environmental monitoring, plastics pollution studies, and machine learning applications in environmental sciences.

Out-of-Scope Use

Use cases requiring detection of objects not classified as microfibres or from unrelated imaging modalities. Applications needing real-time or field-deployable fiber identification without validation.

Dataset Structure

Fields:
    microfibres_Glass_filter (int8): Number of microfibres annotated per image.
    mask: Polygonal segmentation(s) per fibre.
        polygons: List of polygons, each as list of [x, y] points (float32).
        label: Class label, always “microfibre”.
Splits: Typically, training and validation performed with cross-validation. Provided split: "train" (1661 examples).

Annotation files are per-image JSON, each containing bounding box coordinates and masks. Most images contain 1–2 fibres, with a maximum of 7 per image.

Dataset Creation

Curation Rationale

Source Data

Created to automate and improve the detection and quantification of microfibres in wastewater sludge, addressing the laborious nature and scalability limits of manual counting.

Data Collection and Processing

Wastewater samples from various treatment plant processing stages.
Pretreatment to remove organic matter, filter on fibreglass, and microscope imaging.
Manual annotation of fibres in each image for both count and segmentation mask.

Who are the source data producers?

Researchers and laboratory technicians at Universitat Politècnica de València (UPV), Spain.

Annotations

Annotation process

Images manually reviewed and annotated for microfibre presence, area, and location by expert researchers. Annotations are at the instance level (polygon mask and bounding box) and stored in per-image JSON files.

Who are the annotators?

Laboratory and research staff experienced in fibre identification under the microscope.

Personal and Sensitive Information

The dataset contains only microscopy images of environmental samples; no personal, private, or sensitive information is present.

Bias, Risks, and Limitations

The dataset is specific to samples at the UPV, Valencia, Spain. Morphology, concentration, and imaging conditions may not generalize to other locales or treatment plants.
Images only represent microfibres retained on fibreglass filters, and the methodological pipeline may influence visibility of certain fibre types, especially fine or transparent fibres.
Manual annotation may introduce subjective bias and labeling inconsistencies.

Recommendations

Users should consider potential distributional shifts when applying models trained on this dataset to new environments (e.g., other waste streams or different filter types).
Models may underperform for very small fibres or those with morphology similar to the matrix background.

Citation [optional]

BibTeX:

@article{Pérez_Domínguez-Rodríguez_Ferri_Monserrat_2025, title={MicroFiberDetect: An Application for the Detection of Microfibres in Wastewater Sludge Based on CNNs}, volume={39}, url={https://ojs.aaai.org/index.php/AAAI/article/view/35366}, DOI={10.1609/aaai.v39i28.35366}, abstractNote={Microplastics and microfibres are now widespread in aquatic ecosystems, as oceans and rivers. A serious portion of these microplastics come from urban wastewater treatment plants. Traditional methods for detecting and quantifying them are labour-intensive and time-consuming. This paper introduces MicroFiberDetect, a novel application designed to enhance the detection and quantification of microfibres within sludge samples. Leveraging the power of deep learning, this innovative tool provides detection accuracy and insights into the size and colour of each identified fibre. Reducing time and manpower required for analysis while increasing accuracy and throughput. The application has been deployed as a desktop application that allows field experts to quantify and analyse microfibres in sludge samples.}, number={28}, journal={Proceedings of the AAAI Conference on Artificial Intelligence}, author={Pérez, Félix Martí and Domínguez-Rodríguez, Ana and Ferri, Cèsar and Monserrat, Carlos}, year={2025}, month={Apr.}, pages={29682-29684} }

Dataset Card Contact

Felix Marti Perez - fmarper@upv.edu.es