基于Sentinel-2A和GF-1B遥感数据的海岸带水产养殖识别方法比较

近海水产养殖为人类提供优质的动物蛋白，对海洋经济高质量发展具有重要意义。卫星遥感技术已被广泛用于海岸带水产养殖的监测，但目前相关的研究聚焦基于单一传感器卫星数据并使用单一信息提取算法而忽略与不同卫星传感器以及处理方法之间的比较。本研究以北莉岛东北部为研究区，基于多光谱Sentinel 2A卫星和高分一号B（GF 1B）卫星数据，分别使用自适应阈值法、支持向量机监督分类法以及多尺度分割面向对象分类法，开展海岸带人工水产养殖区的识别。研究表明，更高空间分辨率的GF 1B卫星对人工水产养殖水网密集区的提取正确率远优于Sentinel 2A卫星；基于高空间分辨率GF 1B卫星，多尺度分割面向对象分类法的正确率最高，为94.65%，优于支持向量机监督分类法的94.45%和自适应阈值法的84.62%；自适应阈值法更适用于中等空间分辨率卫星数据的水产养殖信息提取，其提取的水产养殖水面的面积与目视解译提取的面积差异小于4%。针对单个养殖池使用情况的业务化监测需使用高空间分辨率卫星数据，而大范围水产养殖面积的变化分析则可应用中等空间分辨率卫星数据。

A comparative study on coastal aquaculture pond identification based on Sentinel-2A and GF-1B remote sensing data

Coastal aquaculture provides human beings with high-quality animal protein, which is of great significance to the high-quality development of the marine economy. Satellite remote sensing technology has been widely applied in aquaculture monitoring, but most research focuses on a sole sensor and specified processing method with no comparisons of other sensors or processing methods. This paper selected the northeast part of Beili Island, Guangdong Province, as the study region, used adaptive thresholding, support vector machine (SVM) supervised classification method, and multi-scale segmentation object-oriented classification method to identify aquaculture ponds based on Sentinel-2A and Gaofen-1B (GF-1B) multi-spectral data. Results showed that the higher spatial resolution GF-1B satellite data was far superior to the Sentinel-2A satellite data in view of the pond identification accuracy, especially in areas of densely populated aquaculture water networks. Based on the high spatial resolution GF-1B satellite imagery, the multi-scale segmentation object oriented classification method obtained the highest detection accuracy of 94.65%, which was better than 94.45% from the SVM method, and 84.62% from the adaptive thresholding method. The adaptive thresholding method was more suitable for aquaculture pond extraction based on medium spatial resolution satellite data. The difference of aquaculture water surface areas detected by adaptive thresholding and visual interpretation was less than 4%. Thus, high spatial resolution satellite data is required for operational monitoring of every single aquaculture pond, while medium spatial resolution satellite data is suitable for analysis of large-scale change of aquaculture areas.