摘要
从数学的角度分析比较了主成分分析(PCA)与独立分量分析(ICA)的原理和特点,给出光谱矩阵在两种不同分析方法下的不同分解;同时结合线性回归和神经网络回归,提出"两步法"来确定不同成分含量测定的最优模型.进而采用PCA与ICA对实际测得的玉米近红外光谱进行了处理,比较分析了两种不同分解所得矩阵的化学含义,以及PCA与ICA两种不同分解对玉米光谱分析结果的影响.仿真结果表明,ICA从独立性角度对光谱数据矩阵进行分解,所得结果更接近实际光谱.最后,利用"两步法"对玉米三种主要成分水、淀粉、蛋白质分别建立了各自最优含量测定模型.结果表明,所建模型符合快速测定要求,具有一定的实用价值.
The principles and characteristics of PCA and ICA were analyzed from the point of view of mathematics, and two different decompositions of spectral matrix were given by using those two methods. Combining with linear regression or neural network, a ＂two-step method＂, that was used to build the different components＇ optimal model was proposed. ICA and PCA was used to process the measured near-infrared spectra of corn respectively; and the chemical meaning of both matrixes derived from the two decompositions and the influence of the two methods on results were analyzed. The experimental results show that ICA does the matrix decomposition from the perspective of independence, and the results are closer to the actual spectrum. The ＂two-step method＂ was applied to build the optimal model of the three major components （moisture, starch, protein） of corn samples. The results show that the models with rapid prediction requirement have practical value.
出处
《中国计量学院学报》
2008年第2期137-141,145,共6页
Journal of China Jiliang University
关键词
独立分量分析
主成分分析
神经网络
近红外光谱
玉米样品
independent component analysis principal component analysis neural network
near infrared spectroscopy corn sample
