引用本文
  •    [点击复制]
  •    [点击复制]
【打印本页】 【下载PDF全文】 查看/发表评论下载PDF阅读器关闭

←前一篇|后一篇→

过刊浏览    高级检索

本文已被:浏览 979次   下载 4115 本文二维码信息
码上扫一扫!
基于高光谱成像技术和力学参数对贡梨冲击损伤的定量研究
李斌1,2,邹吉平1,2,张烽1,2,苏成涛1,2,刘燕德1,2,肖毅华1*
0
(1.华东交通大学 机电与车辆工程学院, 南昌 330013;2.华东交通大学 水果智能光电检测技术与设备国家与地方联合工程研究中心, 南昌 330013)
摘要:
为实现准确评估和预测贡梨的冲击损伤,采用波长为397.5~1 014.0 nm的高光谱成像技术与力学参数相结合对贡梨的冲击损伤进行定量研究。利用基于单摆原理的碰撞装置以及智能数据采集系统获得峰值力、平均接触力、损伤面积和平均压强等力学参数,并对力学参数进行统计分析。利用高光谱成像系统获得损伤贡梨的光谱数据。使用Gap-segment求导、SG求导和基线校准(Baseline)3种光谱预处理方法对原始光谱进行预处理,将光谱数据与力学参数相结合分别建立偏最小二乘回归(Partial least squares regression,PLSR)和主成分回归(Principal component regression,PCR)模型。基于基线校准(Baseline)预处理方法,采用竞争性自适应重加权(Competitive adaptive reweighting sampling,CARS)和无信息变量消除(Uninformative variable elimination,UVE)2种算法进行特征波长的选取,将选取的特征波长作为输入变量并结合力学参数建立PLSR模型。力学参数统计分析和建模的结果表明:1)力学参数在一定程度上可以表征贡梨冲击损伤程度。力学参数的平均值随损伤程度的增加而增大,峰值力平均值从138.40 N增大至335.86 N;平均接触力平均值从77.13 N增大至188.20 N;损伤面积平均值从208.07 mm2增大至544.42 mm2;平均压强平均值从0.34 MPa增大至0.42 MPa。2)Baseline-CARS-PLSR模型对力学参数的预测效果最优,其峰值力、平均接触力、损伤面积和平均压强的预测集相关系数(RP)和预测集均方根误差(RMSEP)分别为0.892和31.527 N、0.883和18.861 N、0.895和54.411 mm2、0.661和0.045 MPa。通过高光谱成像技术与力学参数相结合对贡梨冲击损伤进行定量预测具有一定的可行性,可为贡梨的品质分选及包装方面提供理论支持。
关键词:  贡梨  高光谱成像  力学参数  冲击损伤  偏最小二乘回归  主成分回归
DOI:10.11841/j.issn.1007-4333.2023.02.16
投稿时间:2022-05-25
基金项目:青年科学基金项目(12103019)
Quantitative study on impact damage of Gongli based on hyperspectral imaging technology and mechanical parameters
LI Bing1,2,ZOU Jiping1,2,ZHANG Feng1,2,SU Chengtao1,2,LIU Yande1,2,XIAO Yihua1*
(1.School of Mechatronics & Vehicle Engineering, East China Jiaotong University, Nanchang 330013, China;2.National and local joint engineering research center of fruit intelligent photoelectric detection technology and equipment, East China Jiaotong University, Nanchang 330013, China)
Abstract:
In order to accurately evaluate and predict the impact damage of Gongli, a combined hyperspectral imaging technology and mechanical parameters in the wavelength range of 397. 5-1 014. 0 nm method was proposed in this study to quantitatively detect the impact damage of Gongli. Firstly, the mechanical parameters such as peak force, average contact force, damage area and average pressure were obtained by using the collision device based on single pendulum principle and intelligent data acquisition system, and the mechanical parameters were statistically analyzed. Then, spectral data on damaged Gongli was obtained by a hyperspectral imaging system. The original spectrum was preprocessed by Gap-segment derivative, SG derivative and baseline calibration(Baseline). The partial least squares regression(PLSR)and principal component regression(PCR)models based on spectral data combined with mechanical parameters were established. Finally, the competitive adaptive reweighting sampling(CARS)and uninformative variable elimination(UVE)algorithms were used to select the characteristic wavelength. The selected characteristic wavelength was used as the input variable and the PLSR model based on characteristic wavelength combined with mechanical parameters was established. The results of the statistical analysis and modelling of the mechanical parameters show that: 1)The average value of mechanical parameters increases with the increase of release angle. The average value of peak force increases from 138. 40 N to 335. 86 N, the average value of contact force increases from 77. 13 N to 188. 20 N, the average value of damage area increases from 208. 07 mm2 to 544. 42 mm2, and the average value of pressure increases from 0. 34 MPa to 0. 42 MPa. The change of mechanical parameters can be used to characterize the impact damage degree of Gongli to a certain extent. 2)The baseline-CARS-PLSR model displays the best prediction performance on mechanical parameters. The values of correlation coefficient(RP)and root mean square error(RMSEP)of peak force, average contact force, damage area and average pressure are 0. 892 and 31. 527 N, 0. 883 and 18. 861 N, 0. 895 and 54. 411 mm2, 0. 661 and 0. 045 MPa, respectively. The overall results show the feasibility of quantitative prediction of impact damage of Gongli by hyperspectral imaging combined with mechanical parameters, and provide a theoretical support for quality sorting and packaging of Gongli.
Key words:  Gongli  hyperspectral imaging  mechanical parameter  impact failure  partial least squares regression  principal component regression