基于GA-SVR的苹果可溶性固形物质量分数的高光谱检测

Hyperspectral Detection of Soluble Solids Content on Apple Based on GA-SVR

DOI：10.3969/j.issn.1673-1689.2019.09.018

中文关键词: 苹果 高光谱 BPNN 支持向量机 连续投影算法

英文关键词: apple,hyperspectral,BP neural network,support vector machine,successive projections algorithm

基金项目:

摘要点击次数: 9

全文下载次数: 23

中文摘要:

以建立一种高精度的无损苹果可溶性固形物含量的检测模型为目标，通过提取高光谱图像中圆形150像素感兴趣区域（ROI）内的平均光谱反射率，分别使用Savitzky-Golay平滑处理（S-G）、标准正态变量变换（SNV）和小波变换（Wavelet-Transform）对原始光谱数据进行预处理，然后利用连续投影算法（successive projection algorithm，SPA）提取特征波长，基于特征波长建立BP神经网络（BPNN）和遗传支持向量机（GA-SVR）预测模型。在GA-SVR建模过程中，采用遗传算法获取支持向量机的最优惩罚参数和核函数参数。研究结果表明，S-G预处理后的GA-SVR模型预测效果最佳，模型的预测相关系数为0.850 5，预测均方根误差为0.303 1，所以基于该ROI内数据建立的GA-SVR模型在提高模型性能方面是可行的。

英文摘要:

In order to established a high precision method for the determination of soluble solids content in apple. In this study，the average spectral reflectance in the region of interest（ROI） of 150 pixels in the hyperspectral image is extracted. To reduce the noise in spectral，the extracted reflectance spectra were preprocessed by Savtitzky-Golay smoothing（S-G），Standard Normal Variable Transform（SNV），and Wavelet Transform（WT） methods. The preprocessed spectra were then used to select sensitive wavelengths by Successive Projections Algorithm（SPA） method. Back-propagation neural network（BPNN） and genetic support vector machine（GA-SVR） were applied to build discriminant models with the selected wavelength variables. In the process of establishing GA-SVR model，GA method was used to select the optimal parameters of SVR automatically. The results indicated that the GA-SVR model preprocessed by S-G method was the best model. The prediction correlation coefficient of the model was 0.850 5，and the root mean square error of prediction was 0.303 1. The results show that the GA-SVR model based on the data extracted from this interest region was feasible to improve the performance of the model.

查看全文 查看/发表评论 下载PDF阅读器