Application of MobileNet and Xception neural networks to identify Sillago sihama populations in Vietnam's coastal waters based on otolith morphology

Classification of Indo‐Pacific whiting ( Sillago sihama ) from three coastal regions of Vietnam revealed distinct population structures using otolith morphology. All three analytical approaches – traditional morphometrics using basic dimensional parameters and shape indices (BDP‐ShI), elliptic Fourier descriptors (EFD) and deep learning models – consistently identified the presence of three distinct populations along the Vietnamese coast. EFDs and traditional shape indices achieved moderate performance, with BDPs‐ShIs and EFD reaching average accuracies of 65.92% and 84.67%, respectively. Deep learning models significantly improved classification: MobileNet and Xception achieved 90.50% and 90.33% accuracy, respectively. Linear discriminant analysis confirmed greater overlap between Son Cha and Cat Ba samples, suggesting intermediate morphological characteristics. These results demonstrate that deep learning models better capture complex otolith shape variation and offer scalable tools for fish population identification.