作　　者： ; ; ; ; ;

机构地区： 华南农业大学工程学院南方农业机械与装备关键技术省部共建教育部重点实验室

出　　处： 《华中农业大学学报》 2012年第4期518-523,共6页

摘　　要： 以砂糖橘为对象,建立基于可见-近红外光谱的砂糖橘总酸含量的无损检测方法。试验采集170个完整砂糖橘的500～2 500nm漫反射光谱,然后采用滴定法测定总酸含量。采用Sym8小波变换对光谱进行去噪预处理,并采用连续投影算法(successive projections algorithm,SPA)结合间隔偏最小二乘法(interval partialleast squares,iPLS)优选波长,最终建立BPNN和偏最小二乘法(partial least squares method,PLS)总酸预测模型。结果表明:砂糖橘光谱的小波去噪方法产生的信噪比均值SNR=175.291 1,去噪信号与原始信号间的均方根误差均值RMSE=0.000 13,性能优于常规去噪方法。SPA与iPLS相结合构成的反向偏最小二乘法(back-ward interval partial least squares,BiPLS)_SPA波长选择法能将光谱变量从2 001个压缩到14个,能简化模型并提高建模精度和稳定性。BPNN模型具有更好的非线性映射能力,基于这14个变量的BPNN总酸预测模型的预测相关系数Rp=0.867,预测均方根误差RMSEP=0.061 6,性能优于线性的PLS模型。 Abstract Near-infrared spectroscopy was used to measure total acidity in ‘shatangju’ （Citrus reticulatablanco）. The diffuse reflection spectra of 170 intact samples within 500-2 500 nm were collected. The total acidity in intact samples were measured by titration method. After that, the spectra were denoised using the orthogonal wavelet functions sym8 （level= 3）. And then the spectra variables were optimized by successive projections algorithm （SPA） and interval partial least squares （iPLS）. Finally, the PLS calibration models of intact samples were established and compared. As a result,wavelet de-noising can produce higher SNR and lower RMSE than that of routine method. Wavelength variables were decreased from 2 001 to 14 by biPLS_SPA,and this can help to make the models more concise and robust. The BPNN model produced Rp =0. 867 and RMSEP=0. 061 6 with 14 variables as inputs.

关 键 词： 近红外光谱 小波去噪 连续投影算法 砂糖橘 总酸含量 无损检测

领　　域： [农业科学] [农业科学]