寧夏赤霞珠葡萄水分含量的高光譜無損檢測研究
發(fā)布時間：2018-10-10 17:49
【摘要】：利用可見近紅外高光譜成像技術(shù)對寧夏赤霞珠葡萄含水量的無損檢測進行了初步探討。通過高光譜成像系統(tǒng)(400~1000 nm)采集了136幅赤霞珠葡萄圖像,對原始光譜、平均平滑、高斯濾波、中值濾波、卷積平滑、歸一化、多元散射校正、標(biāo)準(zhǔn)正態(tài)化、基線校準(zhǔn)、去趨勢化等預(yù)處理的偏最小二乘回歸(PLSR)模型進行對比分析;采用主成分分析(PCA)、偏最小二乘回歸(PLSR)、連續(xù)投影算法(SPA)、競爭性自適應(yīng)重加權(quán)(CARS)方法選擇特征波長,建立4種特征波長下的PLSR的葡萄含水量預(yù)測模型,優(yōu)選CARS提取特征波長的方法。在此基礎(chǔ)上,對比分析了全波段與特征波長下的MLR、PCR、PLSR的葡萄含水量預(yù)測模型。結(jié)果表明:采用多元散射校正(MSC)光譜建立的PLSR模型優(yōu)于原始光譜和其他預(yù)處理光譜的PLSR模型;CARS提取特征波長建立的PLSR模型優(yōu)于多元線性回歸(MLR)、主成分回歸(PCR)模型,預(yù)測集的相關(guān)系數(shù)(R)和預(yù)測均方根誤差(RMSEP)分別為0.806、0.144。因此,利用可見近紅外高光譜成像技術(shù)提取特征波長進行寧夏赤霞珠葡萄含水量的檢測是可行的。
[Abstract]:The nondestructive detection of water content of Cabernet Sauvignon grape in Ningxia was studied by using near infrared hyperspectral imaging technique. 136 Cabernet Sauvignon images were collected by a hyperspectral imaging system (400 nm). The original spectra, average smoothing, Gao Si filtering, median filtering, convolution smoothing, normalization, multivariate scattering correction, normalization, baseline calibration, The partial least squares regression (PLSR) model with detrend and other preprocessing is compared and analyzed, and the (SPA), competitive adaptive reweighted (CARS) method is used to select the characteristic wavelengths by using the (PCA), partial least squares regression (PLSR), continuous projection algorithm, the principal component analysis, and the (PLSR), continuous projection algorithm. The prediction model of grape water content under four characteristic wavelengths of PLSR was established, and the method of extracting characteristic wavelength of CARS was selected. On this basis, the water content prediction model of MLR,PCR,PLSR at full wavelength and characteristic wavelength was compared and analyzed. The results show that the PLSR model based on multivariate scattering correction (MSC) spectrum is superior to the PLSR model of the original spectrum and other pretreatment spectra, and the PLSR model based on CARS extraction characteristic wavelength is superior to the multivariate linear regression (MLR), principal component regression (PCR) model. The correlation coefficient (R) and the root mean square error (RMSEP) of the prediction set are 0.806 and 0.144, respectively. Therefore, it is feasible to detect water content of Cabernet Sauvignon grape in Ningxia by extracting characteristic wavelength by using near infrared hyperspectral imaging technology.
【作者單位】： 寧夏大學(xué)農(nóng)學(xué)院;寧夏大學(xué)土木與水利工程學(xué)院;
【基金】：2014年度國家級大學(xué)生創(chuàng)新創(chuàng)業(yè)訓(xùn)練計劃項目(141074917)
【分類號】：TS255.7;O657.3
[Abstract]:The nondestructive detection of water content of Cabernet Sauvignon grape in Ningxia was studied by using near infrared hyperspectral imaging technique. 136 Cabernet Sauvignon images were collected by a hyperspectral imaging system (400 nm). The original spectra, average smoothing, Gao Si filtering, median filtering, convolution smoothing, normalization, multivariate scattering correction, normalization, baseline calibration, The partial least squares regression (PLSR) model with detrend and other preprocessing is compared and analyzed, and the (SPA), competitive adaptive reweighted (CARS) method is used to select the characteristic wavelengths by using the (PCA), partial least squares regression (PLSR), continuous projection algorithm, the principal component analysis, and the (PLSR), continuous projection algorithm. The prediction model of grape water content under four characteristic wavelengths of PLSR was established, and the method of extracting characteristic wavelength of CARS was selected. On this basis, the water content prediction model of MLR,PCR,PLSR at full wavelength and characteristic wavelength was compared and analyzed. The results show that the PLSR model based on multivariate scattering correction (MSC) spectrum is superior to the PLSR model of the original spectrum and other pretreatment spectra, and the PLSR model based on CARS extraction characteristic wavelength is superior to the multivariate linear regression (MLR), principal component regression (PCR) model. The correlation coefficient (R) and the root mean square error (RMSEP) of the prediction set are 0.806 and 0.144, respectively. Therefore, it is feasible to detect water content of Cabernet Sauvignon grape in Ningxia by extracting characteristic wavelength by using near infrared hyperspectral imaging technology.
【作者單位】： 寧夏大學(xué)農(nóng)學(xué)院;寧夏大學(xué)土木與水利工程學(xué)院;
【基金】：2014年度國家級大學(xué)生創(chuàng)新創(chuàng)業(yè)訓(xùn)練計劃項目(141074917)
【分類號】：TS255.7;O657.3
【相似文獻】
相關(guān)期刊論文 前10條
1 李利梅;王秀芹;楊培培;黃衛(wèi)東;戰(zhàn)吉[，
本文編號：2262721
本文鏈接：http://sikaile.net/kejilunwen/huaxue/2262721.html
最近更新
教材專著
