文献类型： 外文期刊

作者： Chen, Pengfei ¹ ; Wang, Fangyong ⁴ ;

作者机构： 1.Chinese Acad Sci, State Key Lab Resources & Environm Informat Syst, Inst Geog Sci & Nat Resources Res, Beijing 100101, Peoples R China

2.Jiangsu Ctr Collaborat Innovat Geog Informat Reso, Nanjing 210023, Peoples R China

3.Natl Sci & Technol Infrastruct China, Natl Earth Syst Sci Data Ctr, Beijing 100101, Peoples R China

4.Xinjiang Acad Agr & Reclamat Sci, Cotton Inst, Shihezi 832000, Peoples R China

关键词： Purifying spectra; High spatial resolution image; Unmanned aerial vehicle; Corn; Wheat; Cotton

期刊名称：FIELD CROPS RESEARCH （ 影响因子：5.8； 五年影响因子：6.9 ）

ISSN： 0378-4290

年卷期： 2022 年 288 卷

页码：

收录情况： SCI

摘要： Estimating plant nitrogen concentrations (PNCs) with remote sensing technology is critical for ensuring precise field nitrogen (N) management. Compared with other remote sensing platforms, low-altitude unmanned aerial vehicles (UAVs) produce images with high spatial resolutions that can be used to clearly identify soil and vegetation. Previously, many spectral indices were designed to remove soil effects to obtain optimal PNC predictions. Herein, we attempt to enhance the PNC prediction accuracy only by removing soil pixels in high-resolution images. Thus, we aimed to collect a dataset containing different crops and image types to investigate whether removing soil pixels to purify crop spectra can improve PNC estimations. For this purpose, N fertilizer experiments were conducted on cotton (Gossypium hirsutum L.), wheat (Triticum aestivum L.) and maize (Zea mays L.), and multispectral and hyperspectral UAV images and PNCs were collected at different growth stages. The multispectral images had actual high spatial resolutions, while the hyperspectral images had virtual high spatial resolutions constructed by fusing high resolution panchromatic images and coarse resolution hyperspectral images. These images represent two typical UAV image types. First, for each crop, the relative changes and driving forces associated with the purified and nonpurified spectra were analyzed under different growth stage, N treatment. Then, three commonly used methods, the spectral index (SI), partial least squares regression (PLSR) and artificial neural network (ANN) methods were used to design PNC prediction model using purified and nonpurified spectra respectively. The results showed the differences between purified and nonpurified spectra were affected by the proportion of crop pixel, sunlit soil pixel and sunshade soil pixel in image. This influence had various trends and magnitudes among different N treatment, growth stages and crop types. It is better to remove soil pixels in imagery, when designing PNC prediction model for plants across growth stages, crop types or even in a single growth stage. The results from actual high spatial resolution images demonstrated this point, with the best PNC prediction model from purified spectra. When considering virtual high spatial resolution image, as the spectrum obtained for each vegetation pixel still represented a mixed vegetation and soil spectrum, removing soil pixels showed no improved performance for PNC estimation. These results provide a reference for others to reasonably choose an optimal data-processing method for constructing PNC prediction models.