【摘要】 在大豆油脂过氧化值近红外光谱分析中,利用间隔偏最小二乘法(interval partial least square,iPLS)实现油脂光谱特征波段选择。分别将全谱波段以10个数据点间隔和20个数据点间隔分成若干个小波段,然后对全谱和每个小波段分别用PLS回归建模,用预测残差平方和(predicted residual sum of squares,PRESS)对模型进行评价。结果表明:经过特征波段选择后,50个波长点模型的决定系数、预测误差均方根、相对误差均值分别为0.9791、0.0513和2.12%,有效地减少建模的变量数,预测精度得到提高。更多还原

【Abstract】 During analyzing peroxide value of soybean oil by near-infrared spectroscopy,iPLS(interval partial least squares) was applied to select characteristic spectral bands of soybean oil.The whole spectrum was divided into several smaller bands using data intervals of 10 and 20 points,respectively.Then,PLS regression models were established by using whole spectrum and smaller bands.These models were evaluated by PRESS(predicted residual sum of squares).The results indicated that after selecting characteristic bands,the R2,RMSEP and RSD were 0.9791,0.0513 and 2.12%,respectively.The variables in these models were decreased effectively and the prediction accuracy was improved.更多还原