基于Landsat-8 的湛江东海岛地物分类研究

“宝钢湛江项目”的实施对近十年湛江东海岛的地物分布产生剧烈影响,尤其是工业用地。本文基于2013 年、2017 年和2021 年的陆地卫星8 号（Landsat-8） 数据对湛江东海岛进行地物分类,研究该区域近十年的用地变化趋势。以2013 年数据为参照：采用归一化水体指数（Normalized Difference Water Index,NDWI） 模型和谱间关系模型实现水陆分离,比对选择分离效果较优者以提取东海岛岸线;对比最大似然法、神经网络法和支持向量机法3 种监督分类方法,选择提取地物效果最优者应用于其余数据。基于Google earth 在线地图及无人机实测数据构建验证点集,使用混淆矩阵进行精度评价。结果表明：谱间关系模型的水陆分离效果较优,提取海岛岸线的精确度有明显提升;支持向量机法的分类总体精度和Kappa 系数最高,分类结果能较好地反映研究区的真实地物分布;汇总三年数据的分类结果,发现用于发展工业的土地面积增长突出且处于持续增长趋势。谱间关系模型与支持向量机法分别实现了对东海岛岸线和地物类型的准确提取,得出近十年研究区的用地变化趋势,能为研究区的用地规划提供参考。

Classification of Land Features in the East Island of Zhanjiang Based on Landsat-8

The implementation of the "Baosteel Zhanjiang Project" has drastically changed the feature distribution of Zhanjiang East Island in the past ten years, especially the industrial land. This paper based on the Landsat-8 data of 2013, 2017 and 2021, the terrain features of Zhanjiang East Island were classified to study the land use change trend of East Island in the past decade. Taking 2013 data as reference: The NDWI model and the spectral relation model are used to realize the separation of water and land, and the better separation effect is selected by comparison to extract the coastline of East Island; Comparing three supervised classification methods: Maximum Likelihood, Neural Net and Support Vector Machine, and select the one with the best extraction effect to apply to the rest of the data. Construct a verification point set based on the Google earth online map and the measured data of the unmanned aerial vehicle, and use the Confusion Matrix for accuracy evaluation. The results show that: The water-land separation effect of the spectral relation model is better, and the accuracy of extracting island coastlines has been significantly improved; Support Vector Machine has the highest overall accuracy and Kappa coefficient, and the classification results can better reflect the real distribution of ground objects in the study area; Summarizing the classification results of the three years of data, it is found that the growth of land area used for industrial development is prominent and in a continuous growth trend. The spectral relation model and the Support Vector Machine are used to accurately extract the coastline and surface feature types of the East Island respectively, and obtained the land use change trend of the study area in the past ten years, which can provide a reference for the land use planning of the study area.