Semantic segmentation for fully automated macrofouling analysis on coatings after field exposure

Area of Science:

• Marine Biology
• Materials Science
• Computer Vision

Background:

• Biofouling poses significant challenges to marine infrastructure, filtration systems, and medical devices.
• Manual assessment of biofouling is labor-intensive, subjective, and lacks detailed spatial and temporal data.
• Developing effective fouling-resistant coatings necessitates reliable and quantitative evaluation methods.

Purpose of the Study:

• To develop an automated, image-based approach for analyzing macrofouling.
• To create a robust system for assessing the effectiveness of anti-fouling coatings.
• To enable the generation of successional models for biofouling studies.

Main Methods:

• A dataset of field panel images with dense labels was curated.
• A convolutional neural network, specifically an adapted U-Net, was employed for semantic segmentation of fouling classes.
• The method focuses on image-based analysis for automated macrofouling quantification.

Main Results:

• The proposed convolutional network accurately segments different macrofouling classes from images.
• The automated analysis provides quantitative data on fouling progression and distribution.
• The approach facilitates the creation of successional models for biofouling dynamics.

Conclusions:

• Automated image analysis offers a more efficient and objective method for macrofouling assessment.
• This technology supports the development and validation of novel fouling-resistant coatings.
• The developed system advances in-depth epibiotic and surface attachment studies.