农业工程学报
農業工程學報
농업공정학보
2014年
21期
301-307
,共7页
孙俊%金夏明%毛罕平%武小红%杨宁
孫俊%金夏明%毛罕平%武小紅%楊寧
손준%금하명%모한평%무소홍%양저
无损检测%主成分分析%图像采集%大米掺假%支持向量机%高光谱图像
無損檢測%主成分分析%圖像採集%大米摻假%支持嚮量機%高光譜圖像
무손검측%주성분분석%도상채집%대미참가%지지향량궤%고광보도상
nondestructive examination%principal component analysis%image acquisition%rice adulteration%support vector machine%hyperspectral imaging
为了有效判别出优质大米中是否掺入劣质大米，该文研究了一种针对大米掺假问题的快速、无损检测方法。从市场上购买了东北长粒香大米和江苏溧水大米，按纯东北长粒香大米、3∶1、2∶2、1∶3和纯江苏溧水大米共5个掺合水平进行大米试验样本的制备。利用可见-近红外高光谱图像采集系统（390～1050 nm）获取了200个大米样本的高光谱图像。采用ENVI软件确定高光谱图像的感兴趣区域（region of interest, ROI），并提取出所有样本在ROI内的平均高光谱数据。采用支持向量机（support vector machine，SVM）建立全光谱波段下的大米掺假判别模型，径向基（radial basis function，RBF）核函数模型交叉验证准确率为93%、预测集正确率为98%。由于高光谱信息量大、冗余性强且受噪声的影响较大，该文采用主成分分析方法（principal component analysis，PCA）分别对大米高光谱图像和高光谱数据进行处理，从特征选择和特征提取2个角度对原始高光谱数据进行处理，通过主成分权重系数图选择了531.1、702.7、714.3、724.7、888.2和930.6 nm 6个特征波长，通过留一交叉验证法（ leave-one-out cross-validation ， LOOCV ）确定并提取出 PCA 降维后的最优主成分数（ number of principal component，PCs）为9。最后分别将优选出的特征波长和提取出的最优主成分数作为模型的输入，建立SVM模型。试验结果表明，基于特征波长SVM模型的交叉验证准确率为95%、预测集正确率为96%，基于最优主成分数SVM模型的交叉验证准确率为94%、预测集正确率为98%。该研究结果表明，该文建立的基于特征波长和基于最优主成分数的SVM模型均具有较优的预测性能，且利用高光谱图像技术对大米掺假问题进行检测是可行的。
為瞭有效判彆齣優質大米中是否摻入劣質大米，該文研究瞭一種針對大米摻假問題的快速、無損檢測方法。從市場上購買瞭東北長粒香大米和江囌溧水大米，按純東北長粒香大米、3∶1、2∶2、1∶3和純江囌溧水大米共5箇摻閤水平進行大米試驗樣本的製備。利用可見-近紅外高光譜圖像採集繫統（390～1050 nm）穫取瞭200箇大米樣本的高光譜圖像。採用ENVI軟件確定高光譜圖像的感興趣區域（region of interest, ROI），併提取齣所有樣本在ROI內的平均高光譜數據。採用支持嚮量機（support vector machine，SVM）建立全光譜波段下的大米摻假判彆模型，徑嚮基（radial basis function，RBF）覈函數模型交扠驗證準確率為93%、預測集正確率為98%。由于高光譜信息量大、冗餘性彊且受譟聲的影響較大，該文採用主成分分析方法（principal component analysis，PCA）分彆對大米高光譜圖像和高光譜數據進行處理，從特徵選擇和特徵提取2箇角度對原始高光譜數據進行處理，通過主成分權重繫數圖選擇瞭531.1、702.7、714.3、724.7、888.2和930.6 nm 6箇特徵波長，通過留一交扠驗證法（ leave-one-out cross-validation ， LOOCV ）確定併提取齣 PCA 降維後的最優主成分數（ number of principal component，PCs）為9。最後分彆將優選齣的特徵波長和提取齣的最優主成分數作為模型的輸入，建立SVM模型。試驗結果錶明，基于特徵波長SVM模型的交扠驗證準確率為95%、預測集正確率為96%，基于最優主成分數SVM模型的交扠驗證準確率為94%、預測集正確率為98%。該研究結果錶明，該文建立的基于特徵波長和基于最優主成分數的SVM模型均具有較優的預測性能，且利用高光譜圖像技術對大米摻假問題進行檢測是可行的。
위료유효판별출우질대미중시부참입렬질대미，해문연구료일충침대대미참가문제적쾌속、무손검측방법。종시장상구매료동북장립향대미화강소률수대미，안순동북장립향대미、3∶1、2∶2、1∶3화순강소률수대미공5개참합수평진행대미시험양본적제비。이용가견-근홍외고광보도상채집계통（390～1050 nm）획취료200개대미양본적고광보도상。채용ENVI연건학정고광보도상적감흥취구역（region of interest, ROI），병제취출소유양본재ROI내적평균고광보수거。채용지지향량궤（support vector machine，SVM）건립전광보파단하적대미참가판별모형，경향기（radial basis function，RBF）핵함수모형교차험증준학솔위93%、예측집정학솔위98%。유우고광보신식량대、용여성강차수조성적영향교대，해문채용주성분분석방법（principal component analysis，PCA）분별대대미고광보도상화고광보수거진행처리，종특정선택화특정제취2개각도대원시고광보수거진행처리，통과주성분권중계수도선택료531.1、702.7、714.3、724.7、888.2화930.6 nm 6개특정파장，통과류일교차험증법（ leave-one-out cross-validation ， LOOCV ）학정병제취출 PCA 강유후적최우주성분수（ number of principal component，PCs）위9。최후분별장우선출적특정파장화제취출적최우주성분수작위모형적수입，건립SVM모형。시험결과표명，기우특정파장SVM모형적교차험증준학솔위95%、예측집정학솔위96%，기우최우주성분수SVM모형적교차험증준학솔위94%、예측집정학솔위98%。해연구결과표명，해문건립적기우특정파장화기우최우주성분수적SVM모형균구유교우적예측성능，차이용고광보도상기술대대미참가문제진행검측시가행적。
Rice is an important food ration of Chinese people, which contains a great number of starch, protein, fat and some nutrient elements. However, rice adulteration is becoming one of the most urgent problems and it needs to be solved as soon as possible in Chinese rice market. Therefore, the purpose of this study was to develop a rapid, precise and nondestructive method to detect the rice adulteration. In this paper, some expensive rice with high quality (Chang-Li-Xiang) and some cheap rice with relatively low quality (Li-Shui) were purchased from the local Wal-Mart in Zhenjiang province, China. Then they were mixed together in five different proportions (0:4, 1:3, 2:2, 3:1 and 4:0) by using electronic scale and the sample of rice adulteration were obtained. The visible and near infrared (VIS-NIR) hyperpectral imaging system with the spectral range of 390-1050 nm was used to capture the hyperspectral images of 200 rice samples. ENVI software was adopted to determine the region of interest (ROI) in the hyperspectral image and extract the hyperspectral data by averaging the reflectance from all the pixels in the ROI images. Then the discriminative model for rice adulteration was established by using support vector machine (SVM) and the extracted hyperspectral data in the full spectral range. The performance of the SVM model was evaluated by using the indexes of cross validation accuracy and prediction accuracy. Finally, the cross validation accuracy was 93%and the prediction accuracy was 98%in the full-spectral-SVM. As there were a large number of noise and redundant information in the raw hyperspectral images and hyperspectral data, some data processing methods should be used to remove the noise, accelerate the processing efficiency and improve the performance of the models. In this paper, the traditional principal component analysis (PCA) method was respectively used to process the hyperspectral images and hyperspectral data from the two aspects of feature selection and feature extraction. For the aspect of feature selection, a total of six characteristic wavelengths (531.1, 702.7, 714.3, 724.7, 888.2 and 930.6 nm) were picked up according to the weight coefficient distribution curve of the first four principal component images under the full wavelengths. For the aspect of feature extraction, the optimal number of principal component (PCs) was determined as 9 by using the leave-one-out cross-validation (LOOCV). Finally, the two kinds of simplified SVM models were respectively developed by using the input data at the six characteristic wavelengths and at the optimal PCs. The experiment results showed that the cross validation and prediction accuracy in the model based on characteristic wavelengths were 95%and 96%, the cross validation and prediction accuracy in the model based on optimal PCs were 94%and 98%. It indicated that the two kinds of simplified models all achieved the promising results and they all had the comparable discriminant power for rice adulteration when compared with the full-spectral-SVM. The results demonstrated that it is feasible to use hyperspectral imaging technology for the detection of the problem of rice adulteration.
