论文部分内容阅读
针对苹果糖度近红外光谱数据的特点,分析了基于BP神经网络和偏最小二乘PLS的苹果糖度定量预测模型建立方法:,将平均影响值方法:(mean impact value)引入到近红外波长选取的过程中来,并与联合区间偏最小二乘法结合,达到波长优选的目的:。首先,利用联合区间偏最小二乘算法,筛选出与苹果的糖度相关度较大的光谱波长数据,再利用PLS-BP方法:建立预测模型。在此模型基础上,使用平均影响值方法:,对参与建模的每个波长数据进行评价,选取影响值最大的一系列波长点,重新建立模型。模型变量数为124,校正均方根误差(RMSEC)为0.1740,验证均方根误差(RMSEP)为0.4565。结果:表明,校正均方根误差,利用平均影响值与联合区间偏最小二乘方法:结合,对光谱数据进行波长的筛选,可以降低模型复杂度,同时提高模型预测精度。
According to the characteristics of near-infrared spectral data of apple brix, a method for establishing a quantitative prediction model of apple brix based on BP neural network and partial least-squares PLS is analyzed. The mean impact value is introduced into the selection of near infrared wavelength Process, and with the joint interval partial least squares method, to achieve the purpose of wavelength optimization :. First of all, by using the joint interval partial least squares algorithm, we selected the spectral wavelength data which has a high correlation with the brix of apple and then used the PLS-BP method to establish the prediction model. On the basis of this model, we use the method of average influence value to evaluate each wavelength data involved in modeling and select a series of wavelength points with the largest influence value to rebuild the model. The number of model variables was 124, the root mean square error of correction (RMSEC) was 0.1740, and the root mean square error of validation (RMSEP) was 0.4565. The results show that the root mean square error is corrected and the average influence value is used in combination with the partial least square method: The combination of spectral filtering and wavelength selection can reduce the complexity of the model and improve the prediction accuracy of the model.