利用近红外光谱技术对354个废旧涤/棉混纺织物进行研究, 通过偏最小二乘法(partial least squares, PLS)和定性鉴别系数建立了不同光谱特征的涤/棉混纺织物近红外定性分析模型。 染色涤/棉混纺织物NIR光谱主要有两大类, 一类具有正常涤/棉光谱特征, 另一类光谱由于样本中染料、 颜料和消光剂等化学助剂的影响, 使光谱谱线成斜线, 失去其光谱特征。 如以全部样本建模, 模型识别率较低。 故将样本分为两类: 斜线光谱样本和正常光谱样本, 分别建立NIR定性分析模型。 NIR定性分析模型建立后, 根据验证结果分别对建模的谱区、 预处理方法和主因子数进行优化, 以提高模型的稳健性和可靠性。 结果表明, 样本分别建模后, 模型的识别率大大提高, 用验证集样本进行内部验证, 正常光谱和斜线光谱所建模型的识别率均达99%, 其校正集相关系数RC均为0.991, 验证集相关系数RP分别为0.983和0.984、 校正标准差SEC分别为0.887和0.453、 预测标准差SEP分别为1.131和0.573。 用150个界外样本分别对正常光谱样本模型和斜线光谱样本模型进行外部预测检验, 模型识别率分别达91.33%和88.00%, 表明所建NIR定性分析模型能够较好地在回收现场进行涤/棉混纺织物的鉴别。

A series of 354 polyester-cotton blend fabrics were studied by the near-infrared spectra (NIRS) technology, and a NIR qualitative analysis model for different spectral characteristics was established by partial least squares (PLS) method combined with qualitative identification coefficient. There were two types of spectrum for dying polyester-cotton blend fabrics: normal spectrum and slash spectrum. The slash spectrum loses its spectral characteristics, which are effected by the samples’ dyes, pigments, matting agents and other chemical additives. It was in low recognition rate when the model was established by the total sample set, so the samples were divided into two types of sets: normal spectrum sample set and slash spectrum sample set, and two NIR qualitative analysis models were established respectively. After the of models were established the model’s spectral region, pretreatment methods and factors were optimized based on the validation results, and the robustness and reliability of the model can be improved lately. The results showed that the model recognition rate was improved greatly when they were established respectively, the recognition rate reached up to 99% when the two models were verified by the internal validation. RC (relation coefficient of calibration) values of the normal spectrum model and slash spectrum model were 0.991 and 0.991 respectively, RP (relation coefficient of prediction) values of them were 0.983 and 0.984 respectively, SEC (standard error of calibration) values of them were 0.887 and 0.453 respectively, SEP (standard error of prediction) values of them were 1.131 and 0.573 respectively. A series of 150 bounds samples reached used to verify the normal spectrum model and slash spectrum model and the recognition rate reached up to 91.33% and 88.00% respectively. It showed that the NIR qualitative analysis model can be used for identification in the recycle site for the polyester-cotton blend fabrics.