基于太赫兹成像检测技术与特征提取方法结合巴旦木饱满度检测方法研究  

Research on Almond Plumpness Detection Method Based on Terahertz Imaging Technology and Feature Extraction Method

作　　者：胡军[1] 吕豪豪 乔鹏 贺永[2] 刘燕德[1] HU Jun;L Hao-hao;QIAO Peng;HE Yong;LIU Yan-de(School of Intelligent Electromechanical Equipment Innovation Research Institute,East China Jiaotong University,Nanchang 330013,China;School of Mechanical Engineering,Zhejiang University,Hangzhou 310027,China)

机构地区：[1]华东交通大学智能机电装备创新研究院,江西南昌330013 [2]浙江大学机械工程学院,浙江杭州310027

出　　处：《光谱学与光谱分析》2024年第7期1896-1904,共9页Spectroscopy and Spectral Analysis

基　　金：国家重点研发计划项目(2022YFD2001805);江西省青年科学基金项目(20224BAB215042);国家青年自然科学基金项目(32302261);赣鄱俊才支持计划·青年科技人才托举项目(2023QT04)资助。

摘　　要：巴旦木是一种营养丰富的坚果,对巴旦木的品质进行检测具有重要的经济价值和实际意义。由于巴旦木具有较为坚硬的外壳,传统的检测手段较难实现内部检测,因此,采用新兴的太赫兹透射成像检测技术,开展巴旦木饱满度的检测研究。首先采集不同饱满度巴旦木的太赫兹透射图像,并且从太赫兹图像的感兴趣区域分别提取无样品区域、空壳区域和满仁区域的太赫兹光谱信息;为了提高模型的精度,减少计算量,采用竞争性自适应重加权算法(CARS)、无信息变量消除(UVE)、连续投影算法(SPA)、蒙特卡罗无信息变量消除法(MCUVE)和遗传算法(GA)对太赫兹光谱信息进行特征提取,建立对应的最小二乘支持向量机(LS-SVM)、随机森林(RF)和K-近邻(KNN)定性判别模型,对巴旦木的饱满和空壳区域进行检测和鉴别。此外,对太赫兹特征图像转为JPG格式,接着转化为RGB格式进行G通道提取和图像二值化分离出外壳和果仁图像,检测饱满度为太赫兹特征图像的壳仁像素点之比;对原始图像进行轮廓提取和图像二值化分离出外壳和果仁图像,实际饱满度为原始图像的壳仁像素点之比。通过计算检测饱满度和实际饱满度的误差,证明了太赫兹透射成像技术检测巴旦木饱满度的可行性。建立的KS-GA-RF模型的鉴别效果最优,准确率为98.21%;通过壳仁像素点之比分别计算出对应的检测饱满度和实际饱满度,误差为16%。研究验证了采用太赫兹图、谱相融合的方法,可以很好地实现对巴旦木内部种仁饱满度可视化检测,为巴旦木的准确分级提供了新的思路,也为太赫兹成像技术检测其他坚果饱满度提供了理论参考,具有重要的应用价值。As a kind of nutrient-rich nut,it is of great economic value and practical significance to test the quality of almonds.Because of the almond hard shell,it is difficult for traditional detection methods to realize internal detection.In this paper,the emerging terahertz transmission imaging detection technology is used to study almond plumpness detection.Firstly,the terahertz spectral images of almonds with different fullness are acquired.Secondly,the terahertz spectra of sample free region,empty shell region and full almond region are extracted,respectively.To improve the accuracy of the model and reduce the computational effort,Competitive Adaptive Reweighting Sampling(CARS),Uninformative Variable Elimination(UVE),Successive Projections Algorithm(SPA),Monte Carlo Uninformative Variable Elimination(MCUVE)and Genetic Algorithm(GA)for feature extraction of terahertz spectral information.The corresponding Least squares support vector machine(LS-SVM),Random forest(RF)and K-nearest neighbor(KNN)qualitative discriminant models are established to detect and identify the full and empty regions of almonds.In addition,the terahertz feature image was to jpg format and then to RGB format,the shell image and kernel image were separated by G-channel extraction and image binarization,and the ratio of shell kernel pixels in the terahertz feature image was detected.The image of shell and kernel were separated by contour extraction and image binarization.The actual plumpness was the ratio of shell kernel pixels in the original image.The terahertz transmission imaging technique's feasibility for detecting the almond's plumpness was proved by calculating the error between the detection plumpness and the actual plumpness.The established KS-GA-RF model had the best identification effect,with an accuracy of 98.21%.According to the ratio of shell and kernel pixels,the corresponding detection and actual fullness were calculated,respectively,with an error of 16%.This study verified that combining terahertz graph and spectrum could well rea

关 键 词：巴旦木饱满度 太赫兹透射成像 特征提取 RF判别模型 

分 类 号：O657.3[理学—分析化学]