宽带光谱成像系统最优训练样本选择方法研究

Optimal Training Sample Selection for Broadband Spectral Imaging System

为解决现有标准色卡或颜色样本集因数量大、存在严重颜色冗余而导致的光谱成像工作繁重的问题,提出一种基于宽带光谱成像系统光谱重建误差最小化的最优训练样本选择方法.通过现有颜色样本集中最有效样本的选择,实现宽带光谱成像系统训练样本的优化.研究通过伪逆方法进行光谱重建,以光谱均方根误差作为评价依据,从颜色样本集中逐步挑选训练样本,实现每次迭代所确定训练样本对样本集重建光谱误差的最小化.实验结果表明,在选择相同数量训练样本条件下,本研究方法所构建训练样本的光谱和色度精度明显优于现有方法.

The existing standard colorcharts or databases always have large sample size and suffer from color redundancy,which inevitability leads to a time-consuming process for practical spectral imaging.In order to resolve this problem,an optimal training sample selection method was proposed whose main idea was choosing the most effective samples from existing database based on error analysis of spectral reconstruction.A typical spectral imaging workflow was set up where the pseudoinverse（PSE）method was employed for spectral reconstruction and spectral root-mean-square error（RMS）was used as evaluation metric.Through minimizing the RMS error for each iteration,the method selected the optimal samples one by one from existing databases.The experimental results show that the proposed method has higher effectiveness both in spectral and colorimetric accuracy than the current existing methods when choosing the same number of training samples.