基于多时相GF-6遥感影像的水稻种植面积提取

为获取高精度水稻种植面积提取方法和分析红边信息在作物识别能力上的优越性，该研究选取辽宁省盘锦市为研究区域，利用2020年水稻关键物候期的多时相高分6号宽幅相机（GF-6 WFV）遥感影像，构建归一化植被指数（Normalized Difference Vegetation Index，NDVI）、归一化水体指数（Normalized Difference Water Index，NDWI）、比值植被指数（Ratio Vegetation Index，RVI）和归一化差异红边1指数（Normalized Difference Red-Edge 1 Index，NDRE1），根据各地物类型进行时序分析，在获得水稻面积粗提取结果的基础上对其他地类进行掩膜，准确提取水稻种植面积。对2020年盘锦市水稻提取结果进行精度分析，结果表明，基于实测数据进行精度验证的总体精度为94.44%，基于目视解译数据进行精度验证的总体精度和Kappa系数分别为95.60%和0.91。根据目视解译数据对有无红边波段参与的水稻提取结果进行对比分析可知，红边波段的引入使总体分类精度、水稻制图精度和Kappa系数分别提高了3.20个百分点、6.00个百分点和0.06。该研究证明红边波段可以有效降低作物的错分、漏分情况，对水稻精准估产和丰富农作物遥感监测方法具有重要作用，显示出国产红边卫星数据在作物分类、面积提取方面具有巨大应用潜力。

Extraction of paddy rice planting areas based on multi-temporal GF-6 remote sensing images

Efficient extraction from high-precision remote sensing images has widely been one of the most important ways to determine the superiority of red-edge information in crop classification. This study aims to quickly and accurately map the paddy rice planting area using GF-6 WFV time-series images in Panjin City, Liaoning Province of China. Six feature types of paddy rice were divided into the construction land, water body, natural vegetation, natural wetland, and dry land, according to the principle of spectral consistency. The Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), Ratio Vegetation Index (RVI), and Normalized Difference Red-Edge 1 Index (NDRE1) were established by the GF-6 WFV images taken in the periods of May 11th, May 25th, June 1st, June 6th, July 20th and August 22nd in 2020. Five stages of images were also divided into the trefoil, transplanting, returning green, booting, and heading stage of paddy rice, according to the phenological rhythm in the study area. Among them, the returning green stage image was covered by June 1st and June 6th. As such, a remote sensing extraction of paddy rice was established, according to the dynamic change of NDVI, NDWI, RVI, and NDRE1 of various feature types over time. Firstly, the NDRE1 at the transplanting and heading stages of paddy rice were selected to preliminarily extract paddy rice. Secondly, some masks were established to remove the impacts of other feature types. The water body and construction land were eliminated by NDWI and maximum RVI, respectively, from the trefoil to heading stage. The natural vegetation was eliminated by NDVI of paddy rice at the trefoil stage. The natural wetland was eliminated by NDVI of paddy rice at the transplanting stage, while, the dry land was eliminated by NDWI in transplanting or returning green stage of paddy rice. Finally, the remaining pixels were taken as the paddy rice. Results showed that the extraction area of paddy rice was 111 058.71hm2 in the study area in 2020, mainly distributed in Dawa District and Panshan County, accounting for 54.47% and 37.95% of the total extraction area, respectively. The overall accuracy was 94.44% under 36 field verification points. Specifically, the overall accuracy was 95.60% with the Kappa coefficient of 0.91, while the mapping accuracy of paddy rice was 95.55% with the user accuracy of 97.28%, after the accuracy verification by 250 visual interpretation points using Google Earth high-resolution images. As such, the distribution map of paddy rice without red-edge bands was obtained using the same remote sensing images and masks, substituting NDVI for NDRE1 in the preliminary paddy rice extraction. More importantly, the extraction with red-edge bands showed the increases of 3.20 percentage points, 6.00 percentage points, and 0.06 in the overall accuracy, mapping accuracy of paddy rice, and Kappa coefficient, respectively. By contrast, the extraction with or without red-edge bands was superimposed on the remote sensing image, indicating that the paddy rice distributions were similar, but the extraction without red-edge bands presented an obvious omission. This finding proved that the red-edge bands effectively reduced the classification error and omission of crops. Consequently, the domestic red-edge satellite data can provide a great application potential to the crop classification and area extraction.