Unmasking Cryptic Diversity: Automated Species Discrimination in the Genus Raja using Deep Learning and Geometric Morphometrics

Abstract

The genus Raja (Family: Rajidae) presents a significant challenge to marine conservation and fisheries management due to profound morphological stasis and the prevalence of cryptic species complexes. Traditional alpha-taxonomic methods frequently fail to distinguish closely related species, such as the phenotypic variants of Raja clavata and Raja montagui, leading to aggregated catch data and the potential overexploitation of vulnerable stocks. This study proposes a transformative "Digital Taxonomy" framework to resolve this taxonomic impediment by synergizing Geometric Morphometrics (GMM) with Deep Learning (DL). We outline a methodology utilizing Generalized Procrustes Analysis (GPA) on 16 anatomical landmarks to quantify shape variation independent of size, combined with Convolutional Neural Networks (CNNs) to extract high-dimensional features from complex dermal textures and chromatic patterns. Furthermore, we propose a Feature-Level Fusion Network architecture that integrates geometric and visual data streams to achieve superior classification accuracy. This automated, reproducible approach offers a robust solution for unmasking cryptic diversity, facilitating accurate stock assessments, and ensuring the long-term resilience of Northeast Atlantic skate populations.