融合地理空间认知的珊瑚礁地貌单元高分遥感分类方法

近年来，在人为活动和自然因素的影响下，全球珊瑚礁面临着大规模退化问题，开展珊瑚礁监测研究对珊瑚礁生态系统评估、修复和保护工作具有重要作用。本文以西沙群岛北礁和华光礁为研究区，应用2015 年高分二号（GF-2）和WorldView-2 高空间分辨率卫星影像和现场调查数据，基于不同珊瑚礁地貌单元的空间位置特征，提出了融合地理空间认知（Geo-Spatial Cognition，GSC）的珊瑚礁地貌单元高分遥感分类方法。研究结果表明：针对因空间位置不同和底质组成高度近似导致珊瑚礁地貌单元漏分和错分的问题，本文提出的方法更能有效获取精准的珊瑚礁地貌单元信息。其中，融合地理空间认知的随机森林（Integrating Geo-Spatial Cognition-Random Forest，GSC-RF）方法展现出了最优的分类表现，在北礁和华光礁珊瑚礁地貌单元分类中总体精度分别为98.06%和91.93%，Kappa 系数分别为0.98 和0.91。相比于仅使用光谱信息的随机森林（Random Forest，RF）、多元逻辑回归（Multinomial Logistic Regression， MLR）和支持向量机（Support Vector Machine，SVM）经典分类方法，本文提出的方法对北礁和华光礁的总体分类精度分别提高了14%~25%和6%~15%。因此，本文构建的融合地理空间认知的珊瑚礁地貌单元分类方法能够大幅提高珊瑚礁地貌单元的分类精度，可为开展大范围的珊瑚礁精细监测提供技术支撑。

英文摘要:

In recent years, coral reefs around the world have faced massive degradation due to anthropogenic and natural factors, and coral reef monitoring studies play an important role in coral reef ecosystem assessment, restoration and conservation. In this paper, Bei Reef and Huaguang Reef of Xisha Islands are taken as the study areas and Gaofen-2(GF-2)and WorldView-2 high spatial resolution images and field survey data in 2015 are applied. Based on spatial location characteristics of different coral reef geomorphic units, a high - resolution remote sensing classification method of integrating Geo-Spatial Cognition(GSC) is proposed. The results of classification showed: Aiming at the problems of omission and misclassification caused by different spatial positions and highly similar sediment composition, the proposed classification method is more effective in obtaining accurate information on coral reef geomorphic units. Integrating Geo-Spatial Cognition-Random Forest(GSC-RF) method showed the best classification performance, with an overall accuracy of 98.06% and 91.93% in Bei Reef and Huaguang Reef, respectively, and the Kappa coefficient is 0.98 and 0.91, respectively. Compared with the classical classification methods of Random Forest(RF), Multinomial Logistic Regression(MLR) and Support Vector Machine(SVM), which only use spectral information. The proposed method improves the overall classification accuracy of the Bei Reef and Huaguang Reef by 14%- 25% and 6%-15%, respectively. The new method can significantly improve the classification accuracy of coral reef geomorphic units and provide technical support for fine monitoring of coral reefs on a large scale.

查看全文查看/发表评论下载PDF阅读器