高光谱技术在牛肉丸复合掺假类型鉴别中的应用

Application of Hyperspectral Imaging Technology in the Identification of Composite Adulteration Type in Beef Balls

肉制品加工过程的复杂性给肉制品掺假检测带来了巨大的挑战。基于高光谱技术,对不同比例的猪肉-鸡肉复合掺假牛肉丸进行了掺假类别判别分析,尤其首次提出了对熟牛肉丸复合掺假的快速检测。为建立模型,分别在牛肉糜中加入不同比例(20%、40%和80%)的猪肉/鸡肉,获得单一掺假样品。并将猪肉和鸡肉按2∶8、5∶5和8∶2的比例混合,在20%、40%和80%比例下进行复合掺假。此外,还制备了油炸掺假牛肉丸,以检验分类模型的适用性。对采集的掺假样本高光谱数据,经五种方法预处理后建立基于极限学习机分类(ELMC)和支持向量分类(SVC)的掺假鉴别模型。此外利用连续投影算法(SPA)、竞争性自适应重加权算法(CARS)分别提取特征波长并建立相应简化模型。研究表明:基于全波长建立的生/熟牛肉丸掺假类别检测模型SVC性能均优于ELMC。而基于特征波长建立的简化模型ELMC性能优于SVC。对生牛肉丸分类判别,利用SPA筛选出的44个特征波长建立的ELMC模型性能最优,其校正集和预测集的分类准确率均为97.17%。基于CARS筛选出的38个特征波长建立的ELMC模型对熟牛肉丸具有最高的分类性能,其校正集和预测集的分类准确率分别为97.17%和96.23%。因此高光谱技术可以对生肉和熟肉的复合掺假类型进行有效、快速、准确的鉴别,为便携式检测设备的研制奠定了理论基础。

The complexity of the meat processing process presents significant challenges in detecting adulteration in meat products.This study uses hyperspectral technology to identify and analyze adulteration in beef meatballs.To establish the models,different proportions(20%,40%,and 80%)of pork/chicken were added to mince beef to obtain single adulterated samples,respectively.Subsequently,pork and chicken were mixed in 2∶8,5∶5,and 8∶2 ratios to prepare samples for composite adulteration under three different gradients(20%,40%,and 80%).In addition,fried adulterated beef balls were also prepared to test the applicability of classification models.Hyperspectral data of the adulterated samples were collected and preprocessed using five different methods.Adulteration identification models were developed using the Extreme Learning Machine Classification(ELMC)and Support Vector Classification(SVC)algorithms.Feature wavelengths were extracted using the Successive Projections Algorithm(SPA)and Competitive Adaptive Reweighted Sampling(CARS),developing corresponding simplified models.The results showed that the performance of the raw/cooked beef ball classification model's SVC model based on full wavelength was better than that of ELMC.In contrast,the simplified model based on characteristic wavelength showed a contrary trend.For the classification of raw beef balls,the ELMC model(SPA-ELMC-Raw)established based on the 44 characteristic wavelengths selected by SPA yielded the best performance,with classification accuracies of 97.17%for both the calibration set and prediction set.For the classification of cooked beef balls,the ELMC model(CARS-ELMC-Cooked)established based on the 38 characteristic wavelengths selected by CARS showed the highest performance,with classification accuracies of 97.17%and 96.23%for the calibration set and prediction set,respectively.The results indicated that hyperspectral imaging technology proves to be an effective,rapid,and accurate method for discriminating between different types of adulteration in raw and cooked meat.This provides a strong theoretical basis for developing portable detection equipment.