Problem

Elasmobranch populations, which include sharks and rays, are experiencing significant global declines, with many species currently classified as threatened. Effective conservation efforts and spatial planning initiatives, such as Important Shark and Ray Areas (ISRAs), rely heavily on accurate and reliable species-level identification. However, existing visual datasets for elasmobranchs predominantly focus on detection rather than fine-grained classification, often feature underwater images with suboptimal visibility due to environmental factors, or are limited to coarse-grained categories. This creates a critical gap for homogeneous, species-centric visual datasets collected under standardized protocols that can support precise morphological classification.

Method

The authors present the eLasmobranc Dataset, a curated and publicly available image collection of seven ecologically relevant elasmobranch species found along the eastern Spanish Mediterranean coast. Data for the dataset was obtained through dedicated field campaigns, collaborations with local fish markets, and open-access public sources like GBIF, iNaturalist, and DeepFish*. A key methodological aspect was the predominant acquisition of images outside of the aquatic environment, following standardized protocols, to ensure clear visualization of diagnostic morphological traits and minimize visual distortions. The dataset integrates expert-validated species annotations, structured spatial and temporal metadata, and additional species-level information. A rigorous filtering strategy was applied to ensure quality control, correct species labeling, and adequate visibility of anatomical features, with all external sources verified for appropriate Creative Commons licenses allowing reuse.

Results

The eLasmobranc Dataset comprises a total of 1,117 images, corresponding to 807 distinct individual specimens across seven specific elasmobranch species: Galeorhinus galeus, Galeus melastomus, Leucoraja naevus, Mustelus mustelus, Oxynotus centrina, Scyliorhinus canicula, and Torpedo marmorata. The dataset is organized into a main directory containing species-specific image subfolders, an attribution CSV file, a citations text file, and an Excel document detailing the included species. Each image is assigned a unique identifier and includes spatial (country, specific area) and temporal (capture date) metadata, which were meticulously reviewed, normalized, and standardized. Expert marine scientists, with extensive experience in elasmobranch research, performed all annotations and image acquisition, ensuring high confidence in species identification. The eLasmobranc Dataset is publicly available on Zenodo at https://zenodo.org/records/18549737, with supplementary resources accessible via GitHub at https://github.com/Tech4DLab/eLasmobranc-Dataset.

Implications

The eLasmobranc Dataset is specifically designed to support multidisciplinary research, including supervised species-level classification, population studies, spatio-temporal analyses, and the development and validation of artificial intelligence systems. It aims to improve monitoring, automated identification, and decision support within marine conservation contexts. By combining regional ecological relevance, morphological clarity achieved through standardized acquisition, taxonomic reliability via expert validation, and public accessibility, the dataset directly addresses a critical gap in resources for fine-grained elasmobranch identification. Furthermore, it actively promotes reproducible research in conservation-oriented computer vision, facilitating advancements in marine management tools and strategies.