Abstract
Illegal souvenir collection strips an estimated five tonnes of seashells fromCosta Rica's beaches each year. Yet, once these specimens are seized, theircoastal origin -- Pacific or Caribbean -- cannot be verified easily due to thelack of information, preventing their return when confiscated by localauthorities. To solve this issue, we introduce BackHome19K, the firstlarge-scale image corpus (19{,}058 photographs, 516 species) annotated withcoast-level labels, and propose a lightweight pipeline that infers provenancein real time on a mobile-grade CPU. A trained anomaly filter pre-screensuploads, increasing robustness to user-generated noise. On a held-out test set,the classifier attains 86.3\% balanced accuracy, while the filter rejects 93\%of 180 out-of-domain objects with zero false negatives. Deployed as a webapplication, the system has already processed 70{,}000 shells for wildlifeofficers in under three seconds per image, enabling confiscated specimens to besafely repatriated to their native ecosystems. The dataset is available athttps://huggingface.co/datasets/FIFCO/BackHome19K