Research on underwater coral identification method based on deep learning

Soft corals are an important component of marine ecosystems, which have potential medicinal and medical research value. In order to monitor information such as the coverage and abundance of underwater soft corals, A large number of images are usually acquired by manual detection methods or by using remotely operated underwater vehicles (ROVs) and autonomous underwater vehicles (AUVs). Then, These images are manually annotated by marine experts or automatically annotated using a Convolutional Neural Network (CNN) classifier. However, Manually annotating a large number of images is both tedious and time-consuming, while CNN classifiers can only classify images in a simple way, which does not adequately represent more important information in the image. To address the challenges in underwater soft coral monitoring, this paper proposes an improved Faster R-CNN detection model aimed at automatically locating and identifying different species of soft corals in an image and providing more detailed annotations. We created a dataset containing common soft coral species in the Sanya sea area through on-site photography. On this dataset, compared with the original Faster R-CNN model, the improved Faster R-CNN model improved 39.7% in mAP and 45.6%in mAP75. In addition, the model also outperforms some newer models.