摘要
银杏果富含维生素、银杏萜内酯和银杏黄酮等成分,具有抗氧化、抗肿瘤、预防心血管疾病等功能,可药食两用。由于银杏果品种不同,其主要成分含量和品质各异。另外,银杏果某些成分含量对其贮藏和加工工艺影响较大。为实现银杏果品种高效无损鉴别,提出一种基于高光谱成像技术的支持向量机(SVM)分类模型,并利用遗传算法(GA)和粒子群算法(PSO)优化模型参数提高种类鉴别正确率。以3个品种630个银杏果为研究对象,按2∶1划分为训练集和测试集,分别为420个和210个。利用高光谱图像采集系统获取900~1700 nm范围内的银杏果图像,黑白校正后选取质心位置25×25 pixel感兴趣区域(ROI),提取该区域内平均光谱作为原始光谱数据。因原始光谱两端噪声较大,信噪比低且有效信息较少,截取945.98~1698.75 nm范围内的光谱波段作为有效波段,并对去噪后光谱波段信息做标准正态变量变换(SNV)预处理,预处理后采用连续投影算法(SPA)和竞争性自适应重加权算法(CARS)提取特征波长,将其波长反射率作为输入矩阵X,预设样本类别1、2、3作为输出矩阵Y,分别建立SNV-SPA/CARS-(GA/PSO)-SVM六种银杏果品种鉴别模型。试验结果表明:SNV-CARS-PSO-SVM模型鉴别效果最佳,分类准确率96.67%,说明CARS提取特征波长变量能代表所有波长信息,且PSO-SVM模型具有较好种类鉴别效果,可实现银杏果鉴别,为银杏果种类高效无损鉴别提供新思路。
Ginkgo fruit with antioxidant,anti-tumour and cardiovascular disease prevention functions is rich in vitamins,ginkgo lactones and ginkgo flavonoids,and can be used for both medicine and food.Due to the different varieties of Ginkgo fruit,the content of the main ingredientsis different and there are differences in quality.In addition,the content of certain components in ginkgo fruit has a greater impact on their storage and processing.In order to achieve efficient and non-destructive identification of ginkgo fruit varieties,the Support Vector Machine(SVM)classification model based on hyperspectral imaging technology was proposed,and Genetic Algorithm(GA)and Particle Swarm Optimization(PSO)was used to optimizethe parameters of the model to improve the accuracy of species identification.In this study,630 ginkgo fruits of three species were regarded as the research objects and divided into training and test sets according to 2∶1,with 420 and 210 samples respectively.The hyperspectral acquisition system acquired Ginkgo fruit images in the range of 900~1700 nm.Then region of interest(ROI)of 25×25 pixel in the center of mass position was selected after black and white correction,and the average spectrum in the region was extracted as the original spectral data.Because of the large noise at both ends of the original spectra,the signal noise ratio was lower and the effective information was less.The spectral band in the range of 945.98~1698.75 nm was intercepted as the effective band,which was pre-processed by Standard Normal Variate transformation(SNV).Successive Projection Algorithms(SPA)and Competitive Adaptive Reweighted Sampling(CARS)were used to extract the characteristic wavelengths.The wavelength reflectivity was used as the input matrix X,and the sample varieties 1,2,3 were used as the output matrix Y.Six identification models were established for the SNV-SPA/CARS-(GA/PSO)-SVM.The experimental results showed that the SNV-CARS-PSO-SVM model had the best identification performance,and the classification accuracy was 96.67%,indicating that the characteristic wavelength variables extracted by CARS could represent all wavelength information,and the PSO-SVM model had a better species identification effect,which could realize the identification of ginkgo fruit.This study provides a new idea for the efficient and non-destructive identification of ginkgo fruit species.
作者
张伏
张方圆
崔夏华
王新月
曹炜桦
张亚坤
付三玲
ZHANG Fu;ZHANG Fang-yuan;CUI Xia-hua;WANG Xin-yue;CAO Wei-hua;ZHANG Ya-kun;FU San-ling(College of Agricultural Equipment Engineering,Henan University of Science and Technology,Luoyang 471003,China;Collaborative Innovation Center of Advanced Manufacturing of Machinery and Equipment of Henan Province,Luoyang 471003,China;College of Physical Engineering,Henan University of Science and Technology,Luoyang 471023,China)
出处
《光谱学与光谱分析》
SCIE
EI
CAS
CSCD
北大核心
2024年第3期859-864,共6页
Spectroscopy and Spectral Analysis
基金
国家重点研发计划项目(2017YFD0301106)
河南省科技攻关计划项目(222102110196)
河南省高等学校青年骨干教师培养计划项目(2017GGJS062)
龙门实验室前沿探索课题(LMQYTSKT032)资助。
