论文部分内容阅读
本文利用高光谱成像技术(Hyperspectral imaging)对常温下贮存的450个未剥皮香蕉样本光谱数据进行采集,首先检测样本果肉可溶性固形物含量(TSS)、坚实度(FIM),采用SPSS单因素方差分析,然后运用线性优化岭回归分析-偏最小二乘法(RR-i PLS)建立了香蕉成熟度理化指标的光谱和图像特征分类模型,结果表明通过实验平台获取光谱数据预测香蕉可溶性固形物含量以及坚实度的相关系数R2值分别为0.92和0.94。再通过连续投影法(successive projections algorithm,SPA)法以及主成分分析法(principal component analysis,PCA)分别选取特征波长,建立基于特征波长的极限学习机(extreme learning machine,ELM)对光谱数据进行建模交叉验证。通过比较RR-i PLS,SPA-ELM与PCA-ELM三种分类预测模型,表明基于特征波长的PCA-ELM分类模型具有较好的预测性能。交叉验证准确率达到99%。为能快速无损识别香蕉果实品质提供一种有效的预测研究,基本满足对香蕉成熟度分类检测且显示出有效建模分析,且能达到有效的经济效益。
In this paper, 450 un-banded banana samples stored at room temperature were collected by Hyperspectral imaging. First, the content of soluble solids (TSS) and firmness (FIM) of the samples were measured and analyzed by SPSS ANOVA , Then the spectral and image feature classification models of the physical and chemical indicators of the ripeness of banana were established by linear regression Ridge Regression Analysis-Partial Least Squares (RR-i PLS). The results showed that the spectral data of the experimental banana can be used to predict the soluble solid content of banana The correlation coefficients of degrees R2 were 0.92 and 0.94, respectively. Secondly, the characteristic wavelengths were selected by successive projections algorithm (SPA) and principal component analysis (PCA) respectively, and an extreme learning machine (ELM) based on characteristic wavelength was established to construct spectral data Cross-validation. By comparing RR-i PLS, SPA-ELM and PCA-ELM three classification prediction models, it is shown that the PCA-ELM classification model based on characteristic wavelength has better prediction performance. Cross-validation accuracy of 99%. It can provide an effective prediction research for rapid and nondestructive identification of banana fruit quality, basically meet banana ripeness classification test and show effective modeling analysis, and can achieve effective economic benefits.