Can explainable AI classify shrike (Laniidae) eggs by uncovering species-wide pigmentation patterns?

Abstract

The complex patterns on bird eggs, characterized by their replicability, distinctiveness, and intricacy, play significant roles in avian biology, including camouflage, protection from brood parasites, protecting embryos, nest identification, strengthening eggshells, and female sexual selection. The genus Lanius, known for its distinctive pigmentation patterns, shows considerable variability within species, making it an intriguing but poorly understood group. We applied Explainable AI (XAI) methods to uncover pigmentation patterns that represent species-wide identification signatures. To do this, we used Convolutional Neural Networks (CNNs) to classify shrike eggs and explore potential correlations between egg identification and eggshell patterns. Our CNN model achieved over 95% accuracy in predicting species, but identifying specific discriminative features proved difficult, as the model only highlighted general trends. This method could help organize collections and verify species affiliation in global ornithological collections, which often face challenges such as missing or illegible labels. CNNs can enhance species identification and improve the accuracy of ornithological studies. Despite some challenges, the potential applications of this research in avian biology and museum collections are promising. It offers new insights into the role of eggshell patterns in avian evolutionary strategies. This approach not only enriches our understanding of egg pigmentation but also contributes to advancements in studies spanning from ecology to biomedical research.