Rapid detection of microfibres in environmental samples using open-source visual recognition models.

Area of Science:

• Environmental Science
• Computer Vision
• Analytical Chemistry

Background:

• Microplastics, especially microfibres, are pervasive environmental pollutants.
• Accurate detection and quantification in complex samples are challenging and labor-intensive.
• Existing methods for microplastic analysis often lack speed and efficiency.

Purpose of the Study:

• To introduce and evaluate open-source visual recognition models for efficient microfibre identification.
• To compare the performance of YOLOv7 and Mask R-CNN in microplastic quantification.
• To provide accessible tools for advancing microplastic contamination assessment.

Main Methods:

• Training and application of two deep learning models: YOLOv7 and Mask R-CNN.
• Utilizing extensive datasets for model training on microfibre recognition.
• Testing model performance on real-world aquatic samples from Seyðisfjörður, Iceland.

Main Results:

• YOLOv7 achieved 71.4% accuracy, while Mask R-CNN achieved 49.9% accuracy in microfibre detection.
• YOLOv7 demonstrated significantly faster identification of microfibres compared to manual methods.
• The models provided rapid results, processing samples in seconds.

Conclusions:

• Open-source visual recognition models, particularly YOLOv7, offer efficient and rapid solutions for microplastic quantification.
• These user-friendly models enhance the speed and accessibility of microplastic pollution research.
• The study advances environmental science by providing tools to better assess microplastic contamination and its impacts.