Mangrove Information Extraction Based on Multi-source Remote Sensing Images

Mangrove is an important wetland ecosystem in coastal areas. It plays an important role in maintaining the coastal ecological environment, and also provides a guarantee for the survival of various organisms. In the past half century, mangrove has been seriously damaged, so it is very important to find an efficient and accurate method to monitor mangrove changes. The remote sensing technology is an important approach to monitor the dynamic changes of mangrove resources and health status. This paper takes the mangrove in Dandou Ocean of Guangxi province as the research area, uses a variety of optical satellite data sources, analyzes the key features of mangrove remote sensing information recognition, and studies the effective methods of mangrove extraction from different data sources. The experimental results show that the accuracy of mangrove extraction is mainly related to the amount of information in the spectrum feature space and the spatial resolution of the data source, and the difficulty of classification is mainly the distinction between terrestrial plants and mangroves. The accuracy of Sentinel-2A satellite using maximum likelihood classification(MLC) and support vector machine classification(SVM) is very close. Landsat8 satellite with medium and low resolution and large number of spectrum feature space and Gaofen-1 satellite with small number of high spatial resolution spectra are more suitable to be extracted by support vector machine classification. The accuracy of landsat8 satellite image extraction by support vector machine classification is the highest, the overall accuracy is 91.42%, and the kappa coefficient is 0.8819.