Object Detection Model for Marine Organisms Based on Faster R-CNN

Abstract

With the development of marine resources, image-based biological target detection technology has gradually become the core method of marine ecological monitoring. This paper adopts Faster R-CNN technology, combined with two deep learning models, VGG and ResNet50, to improve the efficiency of target detection and recognition of underwater organisms. By combining large-scale annotated seabed image datasets for training, accurate localization and recognition of biological targets in images can be achieved. Compared to ResNet50, VGG performs better in complex seabed environments, with its mAP 1.75% higher than ResNet50, indicating higher detection accuracy and robustness. Besides, this study provides a practical and feasible solution for underwater ecological monitoring, verifying the excellent performance of ResNet50 in marine biological target detection, and providing an important and reliable support tool for deep-sea scientific research and ecological protection.