Identifying escaped farmed salmon from fish scales using deep learning.

Area of Science:

• Aquaculture
• Conservation Biology
• Machine Learning

Background:

• Escaped farmed salmon pose a significant threat to wild Atlantic salmon (Salmo salar) populations in Norway.
• Traditional fish scale analysis is effective but labor-intensive and time-consuming for distinguishing farmed from wild fish.

Purpose of the Study:

• To develop and validate a deep learning model for automated identification of escaped farmed salmon using fish scale images.
• To assess the model's generalizability across diverse datasets and imaging conditions.

Main Methods:

• A convolutional neural network was trained and validated on approximately 90,000 fish scale images from national archives.
• The dataset included images with varying imaging protocols, from numerous rivers, and spanning back to the 1930s.
• Model performance was evaluated on a large, independent test set.

Main Results:

• The deep learning model achieved a high F1 score of 0.95 on the independent test set.
• Predictions from the model demonstrated strong agreement with genetic reference samples and known farmed-origin scales.
• The model showed robust generalization across different ecological and methodological contexts.

Conclusions:

• Deep learning provides a highly accurate and efficient automated method for identifying escaped farmed salmon.
• The validated model can serve as a powerful, large-scale tool for monitoring the impact of escaped farmed salmon on wild populations.
• This approach supports conservation efforts for wild Atlantic salmon by enabling more effective population management.