基于Roujean和Ross-Li模型算法的不同户外光照南疆冬枣BRDF特性研究

不同光照下，如何消除或减小反演数据差异，提高检测精度，是目前南疆冬枣户外检测中遇到的一大难题，因此通过用高光谱相机获得的南疆冬枣二向反射分布函数（BRDF）测量值，采用最小二乘法拟合Roujean模型和Ross-Li模型的参数，最后对比Roujean模型和Ross-Li模型反演的结果，提出何种天气何段波长用何种模型反演效果最好的建议，实验结果表明：（1）多云天气，反演南疆冬枣线偏振度（Dolp）时，Ross-Li模型的R2是0.974 8，Roujean模型的R2是0.969 9；反演南疆冬枣强度分量时，Ross-Li模型的R2是0.972 3，Roujean模型的R2是0.974 9。阴天反演南疆冬枣Dolp时，Ross-Li模型的R2是0.965 1，Roujean模型的R2是0.977 8；反演南疆冬枣强度分量时，Ross-Li模型的R2是0.942 0，Roujean模型的R2是0.968 8。晴天反演南疆冬枣Dolp时，Ross-Li模型的R2是0.965 5，Roujean模型的R2是0.926 2；反演南疆冬枣强度分量时，Ross-Li模型的R2是0.928 5，Roujean模型的R2是0.833 1。整体反演的最佳方案是多云天气下的南疆冬枣Dolp用Ross-Li模型反演，强度分量用Roujean模型反演；晴天南疆冬枣Dolp和强度分量均用Ross-Li模型反演；阴天南疆冬枣Dolp和强度分量均用Roujean模型反演。（2）多波段反演的最佳方案是：多云天气下，反演南疆冬枣强度分量时，波长为1 000~1 100 nm范围，需用Ross-Li模型，波长为1 450~1 600 nm范围，需用Roujean模型，其余波段处两种模型均可；反演南疆冬枣Dolp时，在波长为1 300 nm附近，需用Ross-Li模型，其余波段处两种模型均可。阴天反演南疆冬枣强度分量时，在1 000~1 350 nm范围，需用Roujean模型，在1 600 nm附近，需用Ross-Li模型，其余波段处两种模型均可；反演南疆冬枣Dolp时，在1 000~1 350 nm范围，需用Roujean模型，在1 600 nm附近，需用Ross-Li模型。晴天反演南疆冬枣强度分量时，波长为1 000~1 350和1 600 nm附近，需用Ross-Li模型，其余波段处则无特殊要求；反演南疆冬枣Dolp时，在1 000 nm附近，需用Roujean模型，在1 600 nm附近，需用Ross-Li模型。探索出消除或减小反演数据差异的方法，为南疆冬枣户外检测提高精度奠定基础。

A Comparative Study on Roujean and Ross Li Models of Winter Jujube in South Xinjiang Under Different Outdoor Light

How to eliminate or reduce the difference of inversion data and improve the detection accuracy under different illumination is a big problem in the outdoor detection of winter jujube in South Xinjiang. Therefore, this paper obtains the bidirectional reflection distribution function of winter jujube in South Xinjiang by using a hyperspectral camera The least square method is used to fit the parameters of the roujean model and Ross Li model. Finally, the inversion results of the roujean model and Ross Li model are compared, and the viewpoint of which weather, which wavelength and which model are the best is put forward. The experimental results show that: (1) in cloudy weather, in an inversion of Dolp of winter jujube in South Xinjiang，The R-square of Ross Li model is 0.974 8 and that of roujean model is 0.969 9; the R-square of Ross Li model is 0.972 3 and that of roujean model is 0.974 9 when the intensity component of winter jujube is retrieved. In cloudy weather, in an inversion of Dolp of winter jujube in South Xinjiang，the R-square of Ross Li model is 0.965 1, and that of roujean model is 0.977 8; in an inversion of winter jujube intensity component, the R-square of Ross Li model is 0.942 0, and that of roujean model is 0.968 8. In sunny weather, in an inversion of Dolp of winter jujube in South Xinjiang，R-square of Ross Li model is 0.965 5, R-square of roujean model is 0.926 2; in an inversion of winter jujube intensity component, R-square of Ross Li model is 0.928 5, R-square of roujean model is 0.833 1. The best scheme of the whole inversion is to use the Ross Li model for the inversion of winter jujube DOLP in cloudy weather, roujean model for the inversion of intensity component, Ross Li model for the inversion of winter jujube DOLP and intensity component in sunny weather, and roujean model for the inversion of winter jujube DOLP and intensity component in cloudy weather. (2) The best scheme of multi band inversion is: in cloudy weather, when the intensity component of winter jujube in South Xinjiang is retrieved, the wavelength is 1 000~1 100 nm, Ross Li model is needed, the wavelength is 1 450~1 600 nm, roujean model is needed, and the other two models can be used; when the DOLP of winter jujube in South Xinjiang is retrieved, the wavelength is near 1 300 nm, Ross Li model is needed, and the other two models can be used. In cloudy weather, when the intensity component of winter jujube in South Xinjiang is retrieved, the wavelength is 1 000~1 350 nm, roujean model is needed, the wavelength is near 1 600 nm, Ross Li model is needed; when the Dolp of winter jujube in South Xinjiang is retrieved, in the range of 1 000~1 350 nm, roujean model is needed, and in the range of 1 600 nm, Ross Li model is needed. On sunny days, when retrieving the intensity component of winter jujube in South Xinjiang, the wave-length is in the range of 1 000～1 350 nm and around 1 600 nm, the Ross Li model is needed, and there is no special requirement for other wave-bands; when retrieving DOLP of winter jujube in South Xinjiang, the wavelength is near 1 000 nm, the roujean model is needed, and the wavelength is near 1 600 nm the Ross Li model is needed. Thus, the method of eliminating or reducing the difference of inversion data is explored, which lays a foundation for improving the accuracy of outdoor detection of winter jujube in South Xinjiang.