多源时序数据特征优选的南方丘陵山区农作物分类研究  

Crop Classification in Southern Hilly Mountainous Areas Based on Feature Optimization of Multi-source Time Series Data

作　　者：江济强[1,2,3] 郑华健 刘洪顺 JIANG Jiqiang;ZHENG Huajian;LIU Hongshun(Surveying and Mapping Institute Lands and Resource Department of Guangdong Province,Guangzhou 510663,China;Key Laboratory of Natural Resources Monitoring in Tropical and Subtropical Area of South China,Ministry of Natural Resources,Guangzhou 510663,China;Guangdong Science and Technology Collaborative Innovation Center for Natural Resources,Guangzhou 510663,China;Guangdong Surveying and Mapping Technology Co.,Ltd.,Guangzhou 510663,China)

机构地区：[1]广东省国土资源测绘院,广东广州510663 [2]自然资源部华南热带亚热带自然资源监测重点实验室,广东广州510663 [3]广东省自然资源科技协同创新中心,广东广州510663 [4]广东省测绘技术有限公司,广东广州510663

出　　处：《地理空间信息》2024年第11期24-31,共8页Geospatial Information

摘　　要：以广东省揭阳市揭西县为例,基于Sentinel-2多光谱影像、Sentinel-1时序雷达数据、时序植被指数和地形数据,通过计算J-M距离和特征分析对18个特征波段进行特征组合优选,并对比分析了支持向量机(SVM)、随机森林(RF)、最大似然比(MLC) 3种分类器对南方丘陵山区农作物的分类效果。结果表明,时序植被指数特征中NDVI效果最优,时序雷达特征中VH极化方式最优,地形特征中DEM最优;从不同农作物类型来看,时序植被指数特征和时序雷达特征均能提升晚稻、秋玉米的分类精度;对于晚稻而言,时序雷达特征和地形特征对其分类精度均有提升作用。不同分类器对比结果表明,SVM的总体精度比RF和MLC分别高4.67%和7.84%;Kappa系数分别高5.88%和10.48%,可为南方丘陵山区农作物分类提供有效思路和方法参考。In this study,we selected the J-M distance and feature analysis methods to optimize the feature combination of 18 characteristic bands,using multi-source time series data including Sentinel-2 multi-spectral image,Sentinel-1 time series radar data,time series vegetation index and topography data.We compared the classification effects of three classifiers,such as SVM,RF and MLC on crops in the hilly mountainous areas of southern China.The results show that the NDVI feature of time series vegetation index has the best effect,followed by the VH polarization of radar feature and the DEM of topography feature.The classification accuracy of crops is improved for both late rice and corn.The time series radar feature and topography feature also have a significant impact on the classification accuracy of late rice.Comparing the results of different classifiers,the overall accuracy of SVM is the highest,4.67%and 7.84%higher than that of RF and MLC,respectively.The Kappa coefficient is also higher,by 5.88%and 10.48%,respectively.Therefore,the optimization of multi-source time series data features can improve the classification accuracy of crops,providing an effective method for the classification of crops in hilly mountainous areas of southern China.

关 键 词：农作物分类 特征优选 时序雷达特征 时序植被指数特征 南方丘陵山区 

分 类 号：P237[天文地球—摄影测量与遥感]