光谱学与光谱分析, 2017, 37 (7): 2032, 网络出版: 2017-08-30
基于邻近集计算的光谱相似性测度方法研究
Spectral Similarity Measure Method Based on Neighborhood Counting
近红外光谱 邻域保持投影 邻近集计算 相似性测度 Near infrared spectra Neighborhood preserving projections Neighborhood counting Similarity measure
摘要
在近红外光谱数据相似性测度时, 由于光谱数据高维、 非线性、 重叠等特点, 会出现测度距离失效、 数据信息处理困难等难题。 针对传统相似性测度方法在高维空间出现的不适应性, 提出了基于邻近集计算的光谱相似性测度方法。 首先, 采用邻域保持投影neighborhood preserving projections(NPP)算法对原始光谱数据进行降维处理, 该降维方法可以很好的保留原始光谱数据非线性结构信息和数据点的邻域信息。 然后, 在光谱数据降维后的低维空间中, 采用改进的邻近集计算方法, 实现对近红外光谱数据的相似性测度。 实验结果表明, 基于邻近集计算的光谱相似性测度方法, 有效的实现了光谱数据的相似性测度, 在烟叶风格判定和品质分析方面有较好的应用前景, 同时也为高维光谱数据相似性测度提供了一个良好的解决方法。
Abstract
In terms of the near infrared spectrum data similarity measurement,due to the fact that spectral data is high dimensional, non-linear and overlapping, data processing is full of difficulties failures to measure distance. The traditional method of similarity measurement in high dimensional space presented unadapted,so this paper proposed the spectral similarity measure method based on neighborhood counting. Firstly, using the (NPP) algorithm to handle the original spectral data, as a dimension reduction method, can preserve the original nonlinear spectral data structure and neighborhood information.Then, in the low dimensional space, improved neighborhood counting method is used to realize the similarity measure of the near infrared spectrum data. Experimental results show that the spectral similarity measuring method based on neighborhood counting is effective in the spectral data similarity measurement, which has a good prospect in tobacco style determination and quality analysis. Besides, it provides a good similarity measurement solution in high dimensional spectral data.
宋春静, 丁香乾, 徐鹏民, 吕光杰. 基于邻近集计算的光谱相似性测度方法研究[J]. 光谱学与光谱分析, 2017, 37(7): 2032. SONG Chun-jing, DING Xiang-qian, XU Peng-min, L Guang-jie. Spectral Similarity Measure Method Based on Neighborhood Counting[J]. Spectroscopy and Spectral Analysis, 2017, 37(7): 2032.