Spatial attention–driven hybrid Vision Transformers for automated classification of crown-of-thorns starfish in coral reef ecosystems

Abstract

Coral reefs play the main role in marine ecology, as home and shelter to many aquatic species, and quite importantly, supporting local economies. The danger to coral reefs from coral bleaching is joined by Acanthaster planci invasions that can extensively harm these sensitive colonies. The normal practice has been the manual identification of these starfish by divers and snorkelers. This study proposes an automatic deep learning-based approach to detecting and classifying such species within their natural underwater environments. In its architectural setup, the system fashions advanced types of Convolutional Neural Networks (CNN) together with Vision Transformer (ViT) models. This study evaluated several pre-trained CNN and transformer models on the marine species dataset, achieving test classification accuracy of 91.80% for MobileNetV3-Small, 94.54% for VGG19, 94.54% for ResNet50, 95.90% for DenseNet121, 97.54% for Swin-Tiny, 97.54% for CrossViT, and 97.81% for FastViT-T8. The proposed Spatial-Attention Swin + FastViT + CrossViT framework achieved an overall test classification accuracy of 99.18%. It will also make the detection process much safer and faster, besides being more dependable in coral reef protection.