OLAM神经网络分析X荧光谱的预处理研究

在采用最优线性联想记忆(Optimal Linear Associative Memory)神经网络分析X荧光谱时，可用多种方式对谱作预处理，如去掉多余数据、取对数、横向压缩、标准化等.针对实验测得的水泥生料X荧光谱，分别用上述方法进行预处理，然后再用神经网络分析，与化学分析的偏差作对比.结果表明，去掉多余数据、横向压缩可以减少信息处理量，减少存储空间，提高运算速度，且对精度影响较小；标准化处理对含量较少的成分有优势，对含量大的成分误差太大；对数方法则不适用.

Preprocessing of X-ray spectra in analysis of optimal linear associative memory neural network

Some preprocessing methods can be used in optimal linear associative memory neural network to analyze X-ray spectra, such as deleting useless data, logarithm, horizontal compressing, and standardization. This paper uses these methods to analyze the spectra of cement, and compares with the chemistry analysis value respectively. The result show that, deleting useless data and horizontal compressing method can reduce the process information, decrease storage memory, and make the program calculate faster,while the resolution is almost the same; the standardization method is better for less concentration constituents, and the logarithm method is not suitable for OLAM network.