Hybrid Deep Learning for Morphological Classification of Planetary Nebulae

This record provides the preprint manuscript “Hybrid Deep Learning for Morphological Classification of Planetary Nebulae.” The paper presents a two‑stream hybrid model that combines deep convolutional image embeddings with physics‑informed shape and texture descriptors to classify planetary nebulae into bipolar, elliptical, and spherical morphologies. Starting from a curated sample of 171 well‑resolved nebulae (74 bipolar, 38 elliptical, 59 spherical) collected from public astronomical archives, the study applies class‑dependent, physics‑aware data augmentation to expand the effective training set to 1,477 images while preserving morphological realism. The final hybrid network achieves around 94% mean accuracy in repeated 5‑fold cross‑validation and about 92% accuracy on a held‑out test set, clearly outperforming CNN‑only and hand‑crafted‑feature‑only baselines on three‑way PN morphology classification. The manuscript also discusses limitations related to small‑sample regimes and outlines how the proposed framework can be extended to larger, more heterogeneous planetary nebula catalogues and future survey data