基于多核支持向量回归的光谱反射率重建方法

针对传统方法重建光谱反射率过程中未考虑多光谱训练数据维度高、冗余大的特点,导致重建精度低、重建模型学习能力和泛化能力较差的问题,提出了多核支持向量回归的光谱反射率重建方法。首先综合全局、局部核函数的特点引入柯西核函数与多项式核函数的乘积作为多核核函数,然后使用试凑法从训练样本中获取提高模型性能的参数。最后使用多核支持向量回归模型对测试样本进行反射率重建,通过光谱误差及适应度等进行评价。实验结果表明:与伪逆、单核支持向量回归法相比,本文重建方法的光谱误差值降低了0.4～0.785,决策系数提高了2.84～5.27%,平均适应度系数值提高了2%～3%。本文方法在颜色复制中重建精度高、色差较小,满足人眼视觉可容忍范围内。

Spectral reflectance reconstruction based on multi-kernel support vector regression

In the process of spectral reflectance reconstruction by traditional methods, it doesn't take into account of the characteristics of highly dimensional and redundant multi-spectral training data, which results in low reconstruction accuracy, poor learning ability and generalization ability of the reconstruction model. In order to solve such problems, the spectral reflectance reconstruction of multi-kernel support vector regression was proposed. Firstly, the product of Cauchy kernel function and polynomial kernel function was introduced as multi-kernel function combining with the characteristics of overall situation and part of kernel function, and then the parameters to improve the performance of the model were obtained from the training samples by cut-and-trial. Finally, the multi-kernel support vector regression model was used to reconstruct the reflectance of the test samples, and evaluation was made through spectral error and fitness. The experimental results show that compared with the pseudo-inverse and single-kernel support vector regression, the spectral error of the reconstruction method is reduced by 0.4-0.785, the decision-making coefficient is increased by 2.84-5.27%, and the average fitness coefficient is increased by 2%-3%. In color reproduction, the reconstruction accuracy is high and the chromatic aberration is small, so that it can meet the tolerance range of human vision.