Automatic detection of fish scale circuli using deep learning.

Area of Science:

• Ichthyology
• Computational Biology
• Image Analysis

Background:

• Teleost fish scales exhibit growth rings reflecting somatic growth, crucial for fish ageing and growth analysis.
• Manual extraction of incremental growth data from scales is a time-consuming process.

Purpose of the Study:

• To develop a fully automated method for retrieving incremental growth data from fish scale images.
• To leverage Convolutional Neural Networks (CNNs) for accurate detection of scale centers and growth rings (circuli).

Main Methods:

• A pipeline of two CNNs was developed: one for scale center detection and another for circuli detection.
• The focus detector was trained on 725 scale images, achieving 99% average precision.
• The circuli detector was trained on 40,678 circuli annotations, reaching 95.1% average precision.

Main Results:

• The automated system accurately detects scale centers and growth rings across multiple radial transects.
• Circuli detection showed high accuracy, comparable to human labellers, despite challenges in densely spaced freshwater zones.
• The system enables rapid calculation of growth ring spacings for inferring growth in salmon.

Conclusions:

• The developed CNN-based pipeline offers a rapid and accurate solution for extracting fish scale growth data.
• This automated method significantly reduces the labor intensity of traditional scale analysis.
• The approach shows potential for application to a wider range of fish species.