高光谱成像
规范化(社会学)
近红外光谱
像素
遥感
数学
支持向量机
偏最小二乘回归
残余物
人工智能
生物系统
模式识别(心理学)
计算机科学
光学
统计
算法
地质学
物理
社会学
生物
人类学
作者
Yiying Zhao,Chu Zhang,Susu Zhu,Yijian Li,Yong He,Fei Liu
标识
DOI:10.1016/j.postharvbio.2019.111080
摘要
Soluble solids content (SSC) is an important quality attribute in determining fruit maturity and grading after harvest. This study explored the potential of hyperspectral imaging (HSI) coupled with multivariate analysis in visible and near-infrared (Vis-NIR, 380−1030 nm) and near-infrared (NIR, 874−1734 nm) regions for measuring SSC in winter jujube fruit. The effectiveness of applying area normalization to reduce the influence of non-uniform light distribution on the spherical surface of intact fruit was explored. Then linear and non-linear regression models were developed and compared in two spectral ranges. The performance obtained by the least squares-support vector machine (LS-SVM) models based on successive projection algorithm (SPA) was satisfactory. The determination coefficient of prediction (Rp2) and residual predictive deviation (RPD) were 0.873 and 2.81 for the NIR range, and 0.894 and 3.07 for the Vis-NIR range, respectively. The SPA-LSSVM models were applied on the pixel-wise and object-wise spectra of region of interest before and after area normalization for comparison of visualization performance of corresponding prediction maps for SSC. Area normalization could effectively correct the non-uniform reflectance on a spherical object. The overall results indicated that HSI could be used to non-destructively predict and visualize SSC in winter jujubes.
科研通智能强力驱动
Strongly Powered by AbleSci AI