A Machine Learning-Based Approach for Billfish Identification and Morphometric Estimation in Sri Lankan Fisheries

Abstract

This research presents a machine learning-based approach for enhancing billfish species identification and size estimation in Sri Lanka's marine fisheries. The study involves developing two predictive models, where one model automatically identifies billfish species from images, while the other predicts the LJTL of Indo-Pacific Sailfish based on PDL measurements. The methodology incorporates YOLO for species identification, while the custom-trained regression model leverages localized morphometric datasets for size estimation. This dual-model approach addresses the challenges posed by incomplete fish specimens, a common constraint in fisheries data collection. The research contributes to sustainable fisheries management by providing a species-specific solution that enhances the accuracy of billfish population assessments. The proposed method aligns with the requirements of the National Aquatic Resources Research and Development Agency, supporting data-driven conservation and resource management initiatives in Sri Lanka.