离散Landsat归一化植被指数的时间序列融合构建  

Time series fusion construction of discrete Landsat normalized vegetation index

作　　者：曾德裕 卢晓宁[1] 黄玥 杨柳青[1] 孟成真[1] ZENG De-yu;LU Xiao-ning;HUANG Yue;YANG Liu-qing;MENG Cheng-zhen(College of Resources and Environment,Chengdu University of Information Technology,Chengdu 610225,China)

机构地区：[1]成都信息工程大学资源环境学院

出　　处：《湖北农业科学》2019年第12期45-51,共7页Hubei Agricultural Sciences

基　　金：四川省教育厅重点(自然科学)项目(17ZA0075);成都市科技局科技惠民技术研发项目(2016-HM01-00392-SF);国家自然科学基金项目(41401103;41471305;41505012);四川省科技厅重点项目(2017SZ0169);四川省科技厅软科学项目(2016ZR0102);四川省旱涝与重点实验室2018重点项目

摘　　要：结合滑动窗的使用,提出了一种基于傅里叶逼近和线性融合降尺度的时间序列构建方法。在时间序列上逐像元对离散的Landsat数据进行傅里叶逼近,并同分辨率较低的AVHRR传感器数据进行融合,实现了试验区Landsat植被指数的时间序列构建,结合作为实际观测数据的Landsat数据对模型结果的精度进行验证。结果表明,模型结果在色调与纹理等图像特征方面与实际观测数据一致,构建了白色冰雪覆盖区域的NDVI,构建结果与周边地表具有非常好的空间连续性,且整幅影像不存在明显的块状效应;模型结果与实际观测数据具有较高的相关性(平均0.8692)和较低的均方根误差(平均0.0435),平均高达75.04%的像元误差在±0.05之间,高达97.64%的像元误差在±0.1之间,并且误差呈良好的正态分布。模型精度对输入数据量存在一定依赖,同时对混合像元的处理方面尚存在一定的改进空间。In combination with the use of sliding windows, a time series construction method based on Fourier approximation and linear fusion downscaling is proposed. The Fourier approximation of discrete Landsat data is performed on a time-by-pixel basis and merged with the lower resolution AVHRR sensor data to realize the time series construction of the Landsat vegetation index in experimental area, combined with Landsat as the actual observation data. The data verifies the accuracy of the model results. The results show that: The model results in the hue and texture, image features are consistent with the actual observation data, and the NDVI of the white snow-covered area is constructed. The construction results have very good spatial continuity with the surrounding surface, and there is no obvious blocky effect in the entire image;Model results and actual observed data has a high correlation (mean 0.869 2) and a low root mean square error (average of 0.043 5), averaging up to 75.04% of the pixels between ±0.05, up to 97.64% of the pixel error between ±0.1, and the error is in a good normal distribution. The accuracy of the model has a certain dependence on the amount of input data, and there is still room for improvement in the processing of mixed pixels.

关 键 词：LANDSAT NDVI 数据融合 傅里叶分析 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]