The use of near infrared ( NIR) spectroscopy coupled with regression analysis prediction method to detect wood biological decay was investigated in this paper. Principal component regression ( PCR ) analysis and partial least squares regression (PLSR) analysis were compared with the results of extensive research on SIMCA and PLS-DA methods by analysis of the correlation coefficients and the model residuals. The results shown that correlation between the predicted variable of calibration and validation and the measured variable is significant with correlation coefficient (r) over 0.95 with low SEC and SEP (0.07 — 0.20) ; the discriminant accuracy for the non-decay, white-rot and brown-rot decay samples are 100% ( deviation ＜ 0. 27 ) by the PCR and PLSR models based on test set samples; the discriminant accuracy by PLSR is better than that by PCR due to the lower deviation, while both of PCR and PLSR have better discriminant accuracy than that by SIMCA pattern recognition, and has the same discriminant accuracy as PLS-DA method. It ' s suggested that NIR spectroscopy coupled with PCR and PLSR analysis prediction methods could be used to rapidly detect wood biological decay.%利用近红外光谱结合多变量回归分析中常用的主成分回归(PCR)和偏最小二乘法回归(PLSR)分析预测法来判别木材的生物腐朽,并与前期采用的SIMCA和PLS-DA 2种判别方法进行对比分析.结果表明:1)应用近红外光谱结合多变量回归分析方法对校正集样本建立的判别模型,其校正及验证结果与标准值的相关性很高,相关系数均大于0.95,SEC和SEP都很低(0.07 ～0.20),利用模型对未参与建模的样本进行检测,发现2个模型对未腐朽、白腐和褐腐3种类型样本的判别准确率均为100％(偏差都小于0.27);2)对于相同样本集的判别效果,PLSR法比PCR法的判别效果好,且二者都比采用SIMCA法的效果好,并都与PLS-DA法的判别结果相近,说明利用近红外光谱结合回归分析预测法能有效地检测木材的生物腐朽,并对生物腐朽的类型进行准确判别.
展开▼
