将分布式并行计算引入光谱学信号处理领域。 用傅里叶红外光谱仪FTIR-4100获得白砂糖、 木糖醇、 麦芽糖和葡萄糖4类糖各39个样本的光谱曲线作为测试数据。 在两台软硬件配置相同的计算机平台上运行分布式并行算法。 先运用分布式并行方法读取FTIR-4100生成的文本文件中的原始数据, 然后进行分布式并行数据预处理, 包括最大峰值标准化校正, Savitzky-Golay平滑降噪算法等, 再运用分布式并行遗传算法抽取糖特征波数共24个, 最后将提取到的24个特征波数作为用BP神经网络输入, 建立3层人工神经网络。 实验结果表明, 分布式并行计算运行结果与单机顺序计算结果比对一致, 在两台计算机并行工作模式下的计算效率比传统的单机顺序计算处理效率高33.6%, 为光谱学信号处理研究领域进行复杂科学计算和提高计算效率提供了新的方法。

The distributed and parallel computation was introduced to spectroscopy signal processing. The reflection spectra of 4 different varieties of sugar including sucrose, xylitol, maltose and dextrose were measured with FI/IR-4100 Fourier infrared spectral equipment. Each type of sugar consisted of 39 samples. The distributed and parallel algorithm was executed on 2 computers with the same hardware and software systems. First, the distributed and parallel algorithm was used to read original spectral data from the text files generated by FT/IR-4100 device. Second, the data were preprocessed by distributed and parallel algorithm. The preprocessing methods include standard normalization to the maximum peak, Savizky-Golay smoothing denoising, etc. Third, search for the key discriminative wave numbers in mass spectrometry data was performed by distributed and parallel genetic algorithm (GA). At the end, the discriminative features of 24 wave numbers extracted by GA were applied as BP neural network inputs and a 3-layer neural network was built up. The computing results generated by distributed and parallel algorithm are the same as the serial computing results generated by single personal computer. The processing efficiency using 2 personal computers is 33.6% higher than that of serial computation. So the paper presents a creative method for the complex scientific computation and enhancing the computing efficiency in spectroscopy signal processing.