基于改进KNN算法在近红外光谱中的模式识别研究

摘要

针对近红外光谱数据特征变量个数远大于样本数以及光谱点之间存在强相关的特点,通过主成分分析压缩光谱信息抽提独立的特征变量,在最佳主成分个数下计算各样本到不同类中心的马氏距离,进而统计整体的预测正确率.文中采用改进的KNN算法对四种牌号的卷烟近红外光谱数据进行了类别预测,在明显改进效率的同时,获得了更为准确的预测结果.%For the number of variables is much larger than that of samples and the strong correlation characteristics between spectral points, the spectral information was compressed and independent characteristic variables were extracted by principal component analysis. The Mahalanobis distance from each sample to different centers was calculated in the best number of principal components. The overall prediction correctness was counted. The improved KNN algorithmis used to forecast four brands of cigarette category of near-infrared spectral data. A more accurate prediction result was obtained, and at the same time, the efficiency was significantly improved.