Abstract: Coral reefs are of great significance to the study of marine ecological environment. The ecological environment of coral reefs can be evaluated by analyzing the distribution and health status of benthic materials. In this paper, an object-based image classification method is used to determine the optimal segmentation scale of different landforms through experiments. Among them, the optimal segmentation scale of terrestrial and deep-sea is 150, and the optimal segmentation scale of various benthic substances is 30. Furthermore, Sentinel-2A satellite remote sensing images are used to extract benthic materials from coral reefs in Sanya Coral Reef Nature Reserve, Hainan province, and confusion matrix is used to evaluate the accuracy of extraction results. The results show that the overall classification accuracy of benthic material extraction is 87.91%, and the Kappa coefficient is 0.83. The object-oriented classification method can effectively combine the texture characteristics and spectral characteristics of benthic material of coral reef and make full use of the combination characteristics of different bands of remote sensing image, and the extraction results can provide methodological support for reef protection and management in Sanya.

Key words: coral reef, object orientation, classification system, remote sensing images, accuracy evaluation