基于稳健估计不同权函数的氧化铝光谱值对比

Comparison of Alumina Spectral Values Based on Robust Estimation with Different Weight Functions

氧化铝在光谱检测测中会出现粗差问题,在后续光谱数据反演时会影响拟合效果。利用稳健估计的不同权函数与传统的多元线性回归对比,在粗差较大的情况下研究2种方法的回归效果。对不同质量的氧化铝粉末进行观测,通过相关性分析筛选光谱曲线中5个敏感波段作为因变量,质量作为自变量,进行方程拟合。分析不同方法的拟合曲线、中误差、预测值,并对比稳健估计不同权函数的拟合及抗差能力。结果表明,在不含粗差的情况最小二乘和稳健估计效果接近,在含有粗差时稳健估计抗差能力更强。

Alumina will have the influence of gross errors in the spectrum observation,and it will affect the fitting effect in the subsequent inversion of the spectrum data.The different weight functions of the robust estimation were used to compare with the traditional multiple linear regression,the regression effects of the two methods were studied under the larger gross error.The alumina powder of different quality was observed,and the five sensitive bands in the spectral curve were screened as dependent variables through correlation analysis,and the quality as the independent variable to perform equation fitting.The fitting curves of different methods,medium errors and predicted values were analyzed,and robustly estimate the fitting and robustness of different weight functions were compared.The results show that the effects of least squares and robust estimation is close when there is no gross error,and the robust estimation has stronger ability to resist error when there is gross error.