中国空间科学技术

中国空间科学技术 ›› 2025, Vol. 45 ›› Issue (4): 37-47.doi: 10.16708/j.cnki.1000-758X.2025.0056

• 智能遥感应用专题I • 上一篇    下一篇

基于BRDF的多角度光谱曲线反演模型

刘振1,2，孙旭2,*，沙建军1，孙赫2，王德港2,3，高连如2   

  1. 1.哈尔滨工程大学青岛创新发展基地，青岛266000
    2.中国科学院 空天信息创新研究院，北京100094
    3.中国科学院大学, 北京100049
  • 收稿日期:2024-09-30 修回日期:2025-01-13 录用日期:2025-01-22 发布日期:2025-07-22 出版日期:2025-08-01

A multi-angle spectral curve inversion model based on BRDF

LIU Zhen1,2, SUN Xu2,*, SHA Jianjun1, SUN He2, WANG Degang2,3, GAO Lianru2   

  1. 1.Qingdao Innovation and Development Center of Harbin Engineering University, Qingdao 266000, China
    2.Aerospace Information Research Institute, Chinese Academy of Sciences，Beijing 100094, China
    3.University of Chinese Academy of Sciences, Beijing 100049, China
  • Received:2024-09-30 Revision received:2025-01-13 Accepted:2025-01-22 Online:2025-07-22 Published:2025-08-01

PDF (PC)

53

摘要： 为了解决先验光谱库与实测高光谱图像采集倾角不同导致匹配探测精度较低的问题，提出了一种基于核驱动双向反射分布函数BRDF的倾斜光谱曲线反演模型。首先将光谱曲线按照波段拆分，再依据BRDF方程拟合各波段反射率与探测倾角的变化关系，建立反射率反演模型。针对特定倾角采集的高光谱图像，采用反演算法反演出相应倾角下的先验光谱，并进行匹配探测。实验中，使用自主设计组装的野外多角度光谱测量系统，采集了荒漠伪装网及其背景干枯草地的多角度光谱数据，并对其光谱特征进行了分析，同时对比了多种反演算法的性能。实验结果表明，不同地物的反射率分别以相似的趋势随探测倾角变化，拟合光谱曲线与实测光谱曲线差异较小，两者的决定系数均超过0.83，光谱夹角均小于1°。与不同倾角采集的先验光谱相比，反演模型生成的与图像采集角度一致的先验光谱在匹配探测中的AUC值有所提高。本研究丰富了多角度遥感的相关研究，并有效地将多角度遥感应用于高光谱目标探测领域。通过建立的反演模型，提高了高光谱图像匹配探测的精度，具有理论和实际意义。

关键词: 多角度遥感, 双向反射分布函数, 光谱曲线反演, 倾斜测量, 高光谱目标探测

Abstract: To address the issue of low matching detection accuracy caused by differing inclination angles between the a priori spectral library and the measured hyperspectral image acquisition, a tilted spectral curve inversion model is proposed based on a kernel-driven bidirectional reflectance distribution function (BRDF). The spectral curves are firstly split according to the wavebands, and then the relationship between the reflectance of each waveband and the detection inclination is fitted according to the BRDF equation, to establish the reflectance inversion model. For the hyperspectral images collected at a specific inclination angle, the inversion algorithm is used to perform the a priori spectra at the corresponding inclination angle and match the detection. In the experiment, the multi-angle spectral data of the desert camouflage net and its background dry grassland are collected using the independently designed and assembled field multi-angle spectral measurement system, their spectral characteristics are analyzed, and the performance of various inversion algorithms is compared. The experimental results show that the reflectance of different features varies with the detection inclination in a similar trend, respectively, and the differences between the fitted and measured spectral curves are small, with the coefficients of determination of the two exceeding 0.83, and the spectral pinch angles of the spectra are all less than 1°. Compared with the a priori spectra acquired at different inclination angles, the a priori spectra generated by the inversion model with the same angle as the image acquisition have improved AUC values in the matched detection. The research enriches the related studies of multi-angle remote sensing and effectively applies multi-angle remote sensing to the field of hyperspectral target detection. Through the established inversion model, the accuracy of hyperspectral image matching detection is improved, which is of theoretical and practical significance. 

Key words: multi-angle remote sensing, BRDF, spectral curve inversion, inclined measurement, hyperspectral target detection