摘要
我国南方丘陵山区大面积种植油茶果,而目前油茶果的采摘期主要根据节气和经验来判断,过早和过晚采摘油茶果皆会带来经济损失。旨在探索高光谱成像技术准确鉴别油茶果成熟度的可行性,应用波段范围为400~1000 nm的高光谱成像(HSI)系统采集了不同成熟度油茶果共480个样本的高光谱数据。基于SNV、SNV-detrend、SG、一阶导和二阶导5种不同预处理建立PLS-DA和PSO-SVM判别模型。选择最优预处理数据进行特征波长筛选,发现相比于SPA,CARS筛选特征波长建立的简化模型性能更优,CARS-PLS-DA和CARS-PSO-SVM模型预测集分类准确率为92.5%和89.2%,Kappa系数均超过0.86。采用颜色矩的方法提取高光谱图像中颜色特征值结合特征波长建立PLS-DA和PSO-SVM组合模型,发现仍是经CARS筛选特征波长建立的模型性能最优,其中CARS+颜色-PLS-DA和CARS+颜色-PSO-SVM模型预测集分类准确率分别为94.2%和93.3%。特征波长融合颜色特征值的组合建模比单一特征波长建模分类效果好,预测集分类准确率分别提高了1.7%和4.1%。CARS+颜色-PLS-DA模型显示出最佳预测性能,其Kappa系数为0.9231。研究表明利用高光谱成像技术结合化学计量学方法可用于油茶果成熟度检测,为实现快速、无损、准确鉴别油茶果成熟度提供了科学依据。
Camellia oleifera fruit is widely planted in hilly and mountainous areas in southern China.The harvest time of Camellia oleifera fruit is currently decided by solar terms and experience,and the prematurity or too late picking will bring economic losses.This study aimed to explore the feasibility of hyperspectral imaging(HSI)technology to identify the maturity stages of Camellia oleifera fruit accurately.The HSI system with a spectral range of 400~1000 nm was applied to collect hyperspectral images of 480 Camellia oleifera fruit samples at different maturity stages.PLS-DA and PSO-SVM models were individually developed based on spectra preprocessed with five different pretreatments including SNV,SNV-detrend,SG,first-order derivative and second-order derivative.The optimal preprocessing method was selected and further used in feature wavelength screening.Consequently,it was found that the simplified model built by feature wavelengths selected using CARS gave better performance compared to SPA.The classification accuracies of CARS-PLS-DA and CARS-PSO-SVM models in the prediction set were 92.5%and 89.2%,respectively,and the kappa coefficients were above 0.86.Furthermore,color features were extracted from the hyperspectral images by color moment approach,and PLS-DA and PSO-SVM models were built based on the combination of color features and feature wavelengths.Then,the performance of the models built by feature wavelengths screened by CARS was still found to be the best with classification accuracies of 94.2%and 93.3%for CARS+color-PLS-DA and CARS+color-PSO-SVM models in the prediction set,respectively.The models developed by combination features showed better classification results than models based on wavelengths alone,and the classification accuracies were improved by 1.7%and 4.1%in the prediction set,respectively.The optimal CARS+color-PLS-DA model gave the best predicted performance with its Kappa coefficient of 0.9231.As a result,our work indicates that the application of HSI technology combined with chemometric methods can be used to identify the maturity stages of Camellia oleifera fruit,which provides a rapid,nondestructive and accurate way in Camellia oleifera fruit maturity detection.
作者
袁伟东
鞠皓
姜洪喆
李兴鹏
周宏平
孙梦梦
YUAN Wei-dong;JU Hao;JIANG Hong-zhe;LI Xing-peng;ZHOU Hong-ping;SUN Meng-meng(Jiangsu Co-Innovation Center of Efficient Processing and Utilization of Forest Resources,Nanjing Forestry University,Nanjing 210037,China;College of Mechanical and Electronic Engineering,Nanjing Forestry University,Nanjing 210037,China)
出处
《光谱学与光谱分析》
SCIE
EI
CAS
CSCD
北大核心
2023年第11期3419-3426,共8页
Spectroscopy and Spectral Analysis
基金
国家自然科学基金青年科学基金项目(32102071)
江苏省农业科技自主创新资金项目(CX(20)3040)
江苏省高等学校自然科学研究项目(21KJB220013)资助。
关键词
高光谱成像
油茶果
成熟度
颜色特征值
Kappa系数
Hyperspectral imaging
Camellia oleifera fruit
Maturity
Color features
Kappa coefficient
