基于高光谱成像技术生长发育后期苹果糖度的无损检测  

Hyperspectral imaging based non-destructive prediction of soluble solids content in apples at late development period

作　　者：孟田源 王转卫[1] 迟茜[1] 赵凡[1] 翁小凤 

机构地区：[1]西北农林科技大学机械与电子工程学院,陕西杨凌712100

出　　处：《西北农林科技大学学报（自然科学版）》2016年第6期228-234,共7页Journal of Northwest A&F University(Natural Science Edition)

基　　金：国家科技支撑计划项目(2015BAD19B03);国家级大学生创新创业训练计划项目(201410712021)

摘　　要：【目的】研究应用高光谱成像技术无损检测生长发育后期苹果糖度的可行性。【方法】以生长发育后期的“富士”苹果为对象，基于采集到的波长900-700nm高光谱数据，建立预测苹果糖度的偏最小二乘（PLS）、支持向量机（SVM）和极限学习机（ELM）模型，并比较主成分分析（PCA）和连续投影算法（sPA）2种数据压缩或特征波提取方法对预测模型精度的影响。【结果】采用PCA方法可将全光谱压缩至9个主成分，采用SPA从全光谱的230个波长中提取出了13个特征波长，两者相比，SPA能更有效地提高模型预测能力。预测生长发育后期苹果糖度的最佳模型为基于SPA的PLS模型，其预测集相关系数为0．945，均方根误差为0．628οBrix。【结论】高光谱图像技术可以用于生长发育后期苹果糖度的无损检测，该技术的应用将有助于指导苹果的种植和适时采收。[Objective] nondestructively predict so ＇Fuji＇ apples were used as tion models, partial least s This study investigated the feasibility of using luble solids content （SSC） of apples at th samples to acquire hyperspectral images fr quares （PLS）, supp （ELM） ,were built. The effect of characteristic e late d perspectral evelopment om 900 nm to 1 700 image technique to period. [Method] nm. Three predic- ort vector machine （SVM） and extreme learning machine wavelength selection method of successive projections algo- rithm （SPA） and data compression method of principal component analysis （PCA） were compared accord- ing to model predication accuracy. [Result] Nine principal components were compressed by PCA and 13 characteristic wavelengths were selected by SPA from the full spectra （230 wavelengths）. SPA improved the prediction performance effectively. The best model for SSC prediction of apples at late development pe- riod was SPA-PLS,whose correlation coefficient and root mean square error of prediction were 0. 945 and 0. 628 Brix, respectively. [Conclusion] Hyperspectral imaging technique could be used as a noninvasive method for predicting SSC of apples at late development period. This technique is helpful to instruct apple planting and harvest timely.

关 键 词：高光谱成像技术 苹果糖度 无损检测 

分 类 号：S123[农业科学—农业基础科学] S661.1