应用时间序列EVI的MERSI多光谱混合像元分解

应用时间序列EVI的MERSI多光谱混合像元分解 Decomposition of MERSI multispectral mixed pixels by EVI time series 李耀辉 LI Yaohui 1 first-author 王金鑫 WANG Jinxin 1 李颖 LI Ying 2 3 郑州大学水利与环境学院 郑州大学水利与环境学院 School of Water Conservancy and Environment, ZhengZhou University School of Water Conservancy and Environment, ZhengZhou University 中国气象局河南省农业气象保障与应用技术重点实验室 中国气象局河南省农业气象保障与应用技术重点实验室 Henan Key Laboratory of Agrometeorological Support and Applied Technique,CMA Henan Key Laboratory of Agrometeorological Support and Applied Technique,CMA 河南省气象科学研究所 河南省气象科学研究所 Henan Institute of Meteorological Sciences Henan Institute of Meteorological Sciences 针对风云3数据的特点,本文将EVI生长曲线引入多光谱混合像元的分解。首先,利用Landsat8 OLI影像,采用支持向量机的分类方法,提取研究区域的耕地信息,利用该信息对风云MERSI数据进行掩膜处理,获得研究区域的耕地影像。接着,利用MERSI时序影像,计算像元EVI值,通过SG滤波,构建农作物（端元）和混合像元的EVI生长曲线。通过实地调查,获取研究区的农作物端元,尤其对主要的农作物玉米,在空间上均匀选取了14个端元。然后,采用传统的方法,将14种玉米端元生长曲线分别与其它端元组合,进行混合像元分解。发现分解的效果差异很大,提取的玉米种植面积从191.90 km2到574.83 km2不等。为提高分解精度,借用光谱匹配（光谱夹角最小）的方法（用生长曲线代替光谱曲线）自适应选择与混合像元EVI曲线最相似的玉米端元作为组合端元,进行混合像元分解。结果得到玉米的种植面积为589.95 km2,比传统方法的最好（相对）精度提高了2%。 Remote-sensing technology features and the environmental elements of surface complexity together determine mixed pixels in remote-sensing images. Many mature methods of hyper spectral mixed-pixel decomposition are available, but research on the multispectral decomposition of mixed pixels are rare. The purpose of this study is to decompose mixed pixels based on their multispectral imaging characteristics.Hyperspectral images with high spectral resolution may benefit from the spectral unmixing of end-members.By contrast,FY3 multispectral（MERSI）image shavea lower spectral resolution but a higher temporal resolution. Thus,MERSI-EVI time series is introduced in this paper to decompose mixed pixels. The basic parameters of the experiment areas are as follows: study area: Hebi City, Henan Province, China; data: 79 MERSI images acquired from May 1, 2013 to October 15, 2013（89 days had no data） and a Landsat 8 OLI image of the year; purpose: extraction of 2013 corn acreage from the data images. First, the remote-sensing images were processed, and the support-vector-machine classification method was used to extract information on farmlands with the use of a Landsat 8 OLI image. Then, SG-filtered MERSI time-series images were used to calculate EVI; the EVI growth curves of the mixed pixels and the crop end-numbers were then generated. The end-members were determined by field investigation. Corn is the main crop in the area. A total of 14 corn end-members were evenly selected in the space.Then, using the traditional method, the 14 corn end-members were combined with other end-members for unmixing. Finally, the spectral angle matching（SAM） method was used to improve the accuracy of the decomposition and adaptively select the most similar corn end-member with mixed pixels. In this case, a growth curve was used instead of a spectral curve. The results of the traditional decomposition methods vary widely; the extracted corn acreage ranges from 191.90 km2 to 574.83km2,whereas the generated corn acreage of the new decomposition method is 589.95 km2.The 2013 summer corn acreage in Hebi City is780.39 km2. Thus, compared with the best result generated by the traditional methods, the relative accuracy of the new method is improved by 2%. This study shows that using vegetation growth curves to decompose mixed pixels is effective for multispectral images.Of course, this study focused on plains, where crop planting structure is relatively simple. For areas with complex geographical environments and/or planting structures, the performance of the proposed method has yet to be confirmed. 混合像元分解 增强型植被指数 时间序列 风云MERSI数据 作物生长曲线 S513;O211.61 中国气象局·河南省农业气象保障与应用技术重点实验室开放基金（编号:AMF201507,AMF201407）;河南省基础研究计划项目（编号:142300410048） 2021-06-09 3 20