Novelty Recognition: Fish Species Classification via Open-Set Recognition.

Area of Science:

• Computer Vision
• Marine Biology
• Machine Learning

Background:

• Sustainable marine resource management requires reducing fish discards through accurate species registration.
• Current computer vision methods for fish classification are limited by their closed-set nature, failing to recognize unknown species.
• Real-world scenarios necessitate open-set recognition to handle novel fish species encountered in diverse environments.

Purpose of the Study:

• To evaluate open-set recognition techniques for automating fish registration.
• To compare the performance of Multiple Gaussian Prototype Learning (MGPL) against Open-Set Nearest Neighbor (OSNN) and Probability of Inclusion Support Vector Machine (PISVM).

Main Methods:

• Utilized the Fish Detection and Weight Estimation dataset comprising 2216 fish images across nine species.
• Assessed three open-set recognition algorithms: MGPL, OSNN, and PISVM.
• Evaluated algorithms based on their ability to classify both known and unknown fish species.

Main Results:

• OSNN and PISVM demonstrated superior performance over MGPL in recognizing both known and unknown fish species.
• OSNN achieved the highest F1-macro score of 0.79±0.05 and an AUROC score of 0.92±0.01.
• OSNN outperformed PISVM by 0.05 in F1-macro and 0.03 in AUROC for classifying known and unknown species.

Conclusions:

• Open-set recognition models, particularly OSNN, are effective for automating fish registration and supporting sustainable fisheries.
• OSNN provides a robust solution for handling the challenge of unknown species in real-world fish classification tasks.
• The study highlights the importance of open-set approaches for accurate species identification in marine resource management.