Deep-learning technology provides insights into the morphological evolution of birds

The evolution of biological morphology is critical for understanding the diversity of the natural world, yet traditional analyses often involve subjective biases in the selection and coding of morphological traits. This study employs deep learning techniques, utilizing a pretrained ResNet34 model capable of recognizing over 10,000 bird species, to explore avian morphological evolution. We extracted weights from the model's final fully connected (fc) layer to create vector representations of avian species and assessed their similarities using cosine similarity metrics. The results demonstrated multiple clustering patterns with or without biological meaning. Some clustering results are consistent with traditional classifications based on morphology, some are consistent with modern cladistic classifications, and some show behavioural and ecological similarities. Despite these insights, some clusters indicated the influence of non-biological image features on clustering outcomes. This study underscores the potential and limitations of using deep learning approaches in morphological evolution studies.