Enhancing Marine Species Detection: A Comparative Study of Backbone Networks in fastRCNN with Limited Dataset Constraints

This study aims to harness a camera system for collecting marine ecological data and potentially uncovering novel species. Although numerous research teams have successfully employed machine learning models to analyze visual data, these models frequently encounter challenges in adapting to the dynamic marine environment. In response to this issue, our research is centered on refining the application of the Fast RCNN machine learning model through the acquisition of extensive visual data from marine habitats. A comparative analysis of various backbone feature extraction networks, including VGG16-BN, ResNet50-FPN, and MobileNetV3-FPN, revealed that VGG16 exhibits the highest effectiveness in accurately identifying, quantifying, and classifying marine species.