摘要
糖度是评价枇杷内部品质的重要指标.采集枇杷高光谱图像并提取不同感兴趣区域的平均光谱,分析光谱提取区域对枇杷糖度预测精度的影响规律.使用标准正态变换对原始光谱作预处理,分别利用竞争性自适应重加权算法和连续投影算法(Successive Projection Algorithm,SPA)筛选特征光谱,为枇杷糖度建立多元线性回归(Multi Linear Regression,MLR)预测模型.最后采用伪彩色技术实现枇杷糖度含量分布的可视化.结果表明:以整个样本为感兴趣区域,经连续投影算法筛选特征光谱建立的多元线性回归模型具有最好的预测性能(决定系数Rp2=0.822,均方根误差RMSEp=0.435,剩余预测偏差RPD=2.407),基于SPA-MLR模型计算枇杷每个像素点的糖度,生成可视化分布图.表明采用高光谱成像技术可实现枇杷糖度的快速无损检测及可视化研究.
Abstract
The sugar content is an important index to evaluate the internal quality of loquat.The hyperspectral images and the average spectra in different regions of interest of loquat were collected.And the effect of the spectral extraction regions on the prediction accuracy of sugar content of loquat was analyzed.The original spectra was preprocessed by standard normal variation.The competitive adaptive reweighting algorithm and the successive projection algorithm(SPA)were used to select characteristic spectra to establish multi linear regression(MLR)prediction model for predicting the sugar content of loquat.The visualization of the sugar content of loquat was realized by pseudo-color technology.The results indicated that SPA-MLR model based on the spectra in entire sample regions had the best prediction performance(Rp2=0.822,RMSEp=0.435,RPD=2.407).Visualization maps for sugar content were generated by calculating the spectral response of each pixel on loquat samples by SPA-MLR.It is indicated that hyperspectral imaging technology can realize rapid nondestructive detection and visualization of loquat sugar content.
基金项目
贵阳市科技计划项目(筑科合同[2021]43-15号)
国家级大学生创新创业训练计划项目(202210976043)
贵阳学院硕士研究生科研基金项目(GYU-YJS[2021]—45)
维普
万方数据