抗干扰的农作物种植模式自动提取方法

It is very significant for the agricultural information updating and food security to extract agricultural cropping pattern from remote sensing images automatically. For research on remote sensing monitoring of agricultural cropping patterns, there are two important issues needed be addressed: one issue is the data quality, which was generally associated with noise; another issue is the intra-class variability of vegetation indices (VI) temporal profile. This project aims to develop an anti-disturbance automatic method for deriving agricultural cropping patterns. First, the spatiotemporal process and its driving mechanism of vegetation dynamic are investigated, and original VI time series datasets are reconstructed for the purpose of eliminating noise under the multi-level adaptive spatiotemporal modeling framework. During this process, the characteristic of intra-class variability of vegetation indices (VI) temporal profile are evaluated. Second, with the reference of pattern identification methods, characteristics of robustness will be derived through the technical thinking from original profile, frequency-temporal profile to characteristic profile and correlation profile, with consideration of intra- and inter-annual variability of vegetation and climate indices. Finally, an anti-disturbance automatic method for deriving agricultural cropping patterns will be developed. The datasets of agricultural cropping pattern will be derived automatically through its application in main cropping areas of China, and then the spatiotemporal dynamic of cropping index in main cropping areas of China over the years will also be investigated. This method can easily be adapted to other vegetation dynamic monitoring based on remote sensing time series datasets. It is hoped that the outcomes of this project will greatly improve the technical level of time-series classification, and also provide new thinking for solving the problem of the same object with different spectrums and different spectra with the same spectrum.

农作物种植模式的遥感自动提取对于农业信息更新并确保粮食安全具有重要意义。消除植被指数时序数据噪声与类内变化的干扰，是农作物种植模式遥感监测所面临的关键技术问题。本项目拟提出一种能抗干扰的农作物种植模式自动提取方法：首先基于多层次自适应时空建模框架，探索植被演变的时空过程与驱动机制，开展植被指数时序信号分析与重构，以合理消除数据噪声，并科学评估与量化类内变化的干扰特征；进而借鉴模式识别领域的技术方法，综合植被与气候因子的年内、年际变化等多维度信息，通过提取其时频图谱-特征图谱-相关性图谱的鲁棒性特征，以有效避免类内变化干扰，最终创建农作物种植模式自动提取方法。在算法研究的基础上，实现对中国主要农业区农作物种植模式的自动精确提取，并揭示历年耕地复种指数时空演变规律。该方法可拓展用于其他植被覆盖变化遥感监测，促进时序分类技术的发展，并有望为解决遥感分类"异物同谱"与"同物异谱"问题开拓新思路。

农作物种植模式的遥感自动提取，对于及时获取农业信息并确保粮食安全意义重大。消除植被指数时序数据噪声与类内异质性的干扰，是农作物种植模式遥感监测所面临的关键科学技术问题。本项目针对这一国际学术前沿问题开展研究，获得了如下研究成果：(1)针对不同区域、年份与气候条件等多种因素导致的植被复杂类内异质性问题，提出了基于自适应时频域的植被遥感监测方法；(2)提出基于图谱的耕地复种指数自动监测方法，能有效地消除植被物候期变化、植被长势与生长期长度差异等各种类内异质性问题；(3)创建了基于生长期植被指数增量的冬小麦大范围快速识别方法，获得了相当准确的分类精度；(4)建立了基于植被与水体变化比值指数的水稻自动遥感监测方法，能有效地避免降水以及数据来源差异与数据噪声干扰，提高了水稻识别精度；(5)针对时序遥感影像数据可获得性差异带来的挑战，创新性地提出了水稻自适应制图方法；(6)创建了基于生长盛期NMDI增减比值指数的玉米自动制图方法，能适用于全国尺度连续多年玉米种植分布信息快速自动提取；(7)建立全国大范围长时间序列耕地复种指数和水稻、冬小麦和玉米等大宗农作物分布时空连续数据库，深入揭示了其时空演变规律和驱动机制，并有力校验了所创建技术方法的大范围长时序应用推广能力。本项目将深化农作物种植模式自动提取以及时序遥感影像类内异质性的认识，开拓与带动大范围长时序农作物自动制图的研究与应用，为推动克服植被类内异质性挑战建立高精度快速自动植被遥感监测技术开辟新的技术方法与研究思路。