Parallel Processing System for Spectroscopic Understanding of Color Images by Neural network and its Application to Vegetation Monitoring

光谱图像包含海量数据。多变量分析通常被用来提取图像中的重要信息。相反，在所提出的系统中，我们要研究的图像中的重要光谱信息是在学习阶段预先获取的，并作为优化滤光片安装在系统中。讨论了基于自组织映射的滤波器优化问题。构建了实现优化滤光功能的三种宽带滤光系统原型，包括液晶器件和光学元件。利用该系统对四幅滤光片图像进行了光谱重建实验，并与常规方法的测量结果进行了比较。人们期待一种对物体光谱特性敏感的新型智能眼睛。

A spectral image include huge volume of data. Multi-variable analysis is usually used to extract important information in the image. On the contrary in the proposed system, important spectroscopic information in the images which we want to investigate is taken in advance in a learning phase and installed in a system as optimized filters. Filter optimization by self-organized map is discussed. Three proto-types of broad-band filtering systems to implement optimize filter functions which include liquid-crystal devises and optical components were constructed. Experimental results of spectral reconstruction from four filterer images by proposed system were compared with measured result by conventional method. A new intelligent eye which is sensitive to spectral characteristics of an object will be expected.