Sequence Segmentation of Nematodes in Atlantic Cod with Multispectral Imaging Data

Area of Science:

• Food Science and Technology
• Image Processing
• Machine Learning

Background:

• Nematode contamination is a significant issue in the fish processing industry, especially for white fish.
• Current manual inspection methods for nematodes are labor-intensive and prone to errors.
• Technological advancements are needed to improve the efficiency and accuracy of nematode detection in fish fillets.

Purpose of the Study:

• To develop and evaluate an automated system for detecting and differentiating nematodes from other defects in fish fillets.
• To investigate the potential of multispectral imaging (MSI) and machine learning for identifying nematodes based on their spectral signatures.
• To improve process control and product safety in the fish processing industry.

Main Methods:

• Acquisition of 270 multispectral images of Atlantic cod fillets using VideometerLab 4.
• Automated segmentation of nematodes using the Segment Anything Model (SAM) with minimal labeled data.
• Development of segmentation models using normalized Canonical Discriminant Analysis (nCDA) to distinguish nematodes from skin remnants and blood spots.
• Evaluation of model performance using precision and recall metrics.

Main Results:

• Successfully labeled 173 nematodes using the Segment Anything Model (SAM).
• Achieved 88% precision and 79% recall in identifying nematodes in the annotated test data.
• Demonstrated the potential of MSI and nCDA for accurate nematode differentiation in fish fillets.

Conclusions:

• The developed automated system using MSI and nCDA shows promise for improving nematode detection in fish processing.
• This approach can enhance product safety and quality by accurately identifying contaminated fillets.
• The system can reduce unnecessary inspections, optimizing the processing workflow.