高光谱技术结合特征波长筛选和支持向量机的哈密瓜成熟度判别研究

发布时间：2018-06-02 00:10

本文选题：高光谱 + 哈密瓜　；参考：《光谱学与光谱分析》2017年07期

【摘要】：可溶性固形物含量(SSC)和硬度是哈密瓜划分等级的重要指标,同时也是其成熟度的表征因子。因此,为满足哈密瓜自动化分级和适宜采摘,采用高光谱技术结合特征波长筛选的方法,同时对哈密瓜的可溶性固形物含量、硬度及成熟度进行了无损检测研究。对多元散射校正(MSC)处理后的光谱分别利用连续投影算法(SPA)、竞争性自适应重加权算法(CARS)和CARS-SPA方法筛选了哈密瓜可溶性固形物和硬度的特征波长,并将原始光谱、MSC预处理后的光谱和所筛选的特征波长作为输入变量分别建立哈密瓜可溶性固形物和硬度的支持向量机(SVM)预测模型及成熟度判别模型。结果显示,MSC-CARS-SPA方法所建立的可溶性固形物和硬度SVM预测模型最优,其Rpre,RMSEP和RPD分别为0.940 4,0.402 7,2.94 1和0.825 3,35.22,1.771。同时对哈密瓜成熟度进行了判别分析,并分别建立了基于全光谱、单一的可溶性固形物或硬度特征波长和主成分分析(PCA)特征融合的哈密瓜成熟度SVM判别模型。结果显示,CARS-PCASVM模型的判别结果与全光谱SNV-SVM模型相同,其校正集和预测集判别正确率分别为95%和94%。研究表明,利用高光谱技术结合特征波长筛选方法可实现同时对哈密瓜可溶性固形物和硬度的定量预测及成熟度判别。
[Abstract]:Soluble solids content (SSCS) and hardness are important indexes for the classification of Hami melon, and they are also the characterization factors of the maturity of Hami melon. Therefore, in order to satisfy the automatic grading and suitable picking of Hami melon, the method of hyperspectral technology combined with characteristic wavelength screening was adopted, and the content of soluble solids, hardness and maturity of Hami melon were studied by nondestructive testing. The spectral characteristics of soluble solids and hardness of Hami melon were screened by continuous projection algorithm (SPAS), competitive adaptive reweighting algorithm (CARSs) and CARS-SPA method. The prediction model of soluble solids and hardness of Hami melon by support vector machine (SVM) and the maturity discriminant model were established by using the pre-treated spectrum and the selected characteristic wavelength of MSC as input variables. The results show that the SVM prediction model of soluble solids and hardness developed by MSC-CARS-SPA method is the best. Its RPD and RMS EP are 0.940 4 ~ 0.402 7 ~ 2.941 and 0.825 ~ 3 ~ 35.225.221.771respectively. At the same time, discriminant analysis of Hami melon maturity was carried out, and the SVM discriminant model of Hami melon maturity was established based on full spectrum, single soluble solids or hardness characteristic wavelength and principal component analysis (PCA). The results show that the discriminant result of CARS-PCASVM model is the same as that of full-spectrum SNV-SVM model, and the accuracy of correction set and prediction set are 95% and 94% respectively. The results showed that the quantitative prediction and maturity discrimination of soluble solids and hardness of Hami melon could be realized by using hyperspectral technique combined with characteristic wavelength screening method.
【作者单位】：石河子大学食品学院;石河子大学机械电气工程学院;中国农业大学工学院;
【基金】：国家自然科学基金项目(61263041) 国家科技支撑项目(2015BAD19B03)资助
【分类号】：O657.3;TS255.7

【相似文献】