文献类型： 外文期刊

作者： Fu, Yuanyuan ¹ ; Yang, Guijun ¹ ; Li, Zhenhai ¹ ; Li, Heli ¹ ; Li, Zhenhong ¹ ; Xu, Xingang ¹ ; Song, Xiaoyu ¹ ; Zhang, Y ¹ ;

作者机构： 1.Beijing Res Ctr Informat Technol Agr, Key Lab Quantitat Remote Sensing Agr, Minist Agr, Beijing 100097, Peoples R China

2.Natl Engn Res Ctr Informat Technol Agr, Beijing 100097, Peoples R China

3.Beijing Engn Res Ctr Agr Internet Things, Beijing 100097, Peoples R China

4.Beijing Res Ctr Intelligent Equipment Agr, Beijing 10097, Peoples R China

5.Newcastle Univ, Sch Engn, Newcastle Upon Tyne NE1 7RU, Tyne & Wear, England

关键词： Hyperspectral remote sensing; Cereal crop N status indicators; N-related hyperspectral feature analysis technique; Machine learning based regression; Radiative transfer model

期刊名称：COMPUTERS AND ELECTRONICS IN AGRICULTURE （ 影响因子：5.565； 五年影响因子：5.494 ）

ISSN： 0168-1699

年卷期： 2020 年 172 卷

页码：

收录情况： SCI

摘要： Nitrogen (N) is the most limiting nutrient for cereal crop production, which often results in over-application of N fertilization to maximize crop yield. Negative environmental impacts and long-term reductions in productivity has encouraged site-specific N fertilization approaches, but these require timely and accurate crop N monitoring. The advent of hyperspectral remote sensing potentially provides a fast and economic way to accomplish this. A framework for hyperspectral remote sensing of cereal crop N is introduced, based on a comprehensive literature survey, to help inform monitoring best practices. Existing and potential crop N status indicators are summarized, with some recommendations provided. Hyperspectral analysis techniques for extracting N-related features are also examined and categorized into spatial domain and frequency domain based methods. In-depth analyses are conducted regarding: (1) the inconsistency in selected wavebands by different band selection methods and (2) determination of optimal wavelet, scale and wavelength in continuous wavelet transformations. Characteristics and deployment of machine learning based regression methods are also presented for crop N monitoring. Further, existing strategies to alleviate the ill-posed problem in physical and hybrid methods are outlined with some examples. Finally, the strengths and weaknesses of crop N retrieval methods are summarized to improve the understanding of how these methods affect prediction quality. Existing limitations and future areas of research emphasize on the fusion of crop N-related features from different domain spaces and the improved combination of empirical and physical methods. Nitrogen (N) is the most limiting nutrient for cereal crop production, which often results in over-application of N fertilization to maximize crop yield. Negative environmental impacts and long-term reductions in productivity has encouraged site-specific N fertilization approaches, but these require timely and accurate crop N monitoring. The advent of hyperspectral remote sensing potentially provides a fast and economic way to accomplish this. A framework for hyperspectral remote sensing of cereal crop N is introduced, based on a comprehensive literature survey, to help inform monitoring best practices. Existing and potential crop N status indicators are summarized, with some recommendations provided. Hyperspectral analysis techniques for extracting N-related features are also examined and categorized into spatial domain and frequency domain based methods. In-depth analyses are conducted regarding: (1) the inconsistency in selected wavebands by different band selection methods and (2) determination of optimal wavelet, scale and wavelength in continuous wavelet transformations. Characteristics and deployment of machine learning based regression methods are also presented for crop N monitoring. Further, existing strategies to alleviate the ill-posed problem in physical and hybrid methods are outlined with some examples. Finally, the strengths and weaknesses of crop N retrieval methods are summarized to improve the understanding of how these methods affect prediction quality. Existing limitations and future areas of research emphasize on the fusion of crop N-related features from different domain spaces and the improved combination of empirical and physical methods.