高光谱成像技术鉴别鱼新鲜度

摘要

采用高光谱成像技术对鱼新鲜度进行检测研究.首先,提取鱼样本感兴趣区域(region of interest, ROI)光谱,分别采用竞争性自适应重加权算法(CARS),连续投影算法(SPA)和遗传算法(GA)提取特征波长,三种算法分别得到57,31和66个特征变量,采用最小二乘支持向量机和SIMCA作为分类模型,将57, 31和66个特征变量作为LS-SVM和SIMCA模型的输入变量建立分类模型,基于SPA-LS-SVM 和CARS-LS-SVM模型预测集识别率分别达到了98% 和96%,而采用SIMCA建立的模型取得了较差的预测结果, GA-SIMCA,SPA-SIMCA和CARS-SIMCA模型预测集识别率都只是达到了52%.结果表明,LS-SVM 作为分类模型优于SIMCA模型,SPA和CARS选择的特征波长,不但可以简化模型,还可以提高模型的预测精度,采用高光谱成像技术可以有效检测鱼的新鲜度,并能准确检测出鱼不同冻融次数和冷冻时间.%This study investigated the feasibility of using near infrared hyperspectral imaging system(NIR-HIS)technique for non-destructiveidentification of fresh and frozen-thawed fish fillets.Hyperspectral images of freshness,storage time,and frozen-thawed times offillets for turbot flesh were obtained in the spectral region of 380~1 023 nm.Reflectance values were extracted from each region of interest(ROI)of each sample.Competitive adaptive reweighted sampling(CARS)algorithm,successive projections algorithm(SPA)and genetic algorithm(GA)were carried out to identify the most significant wavelengths.Based on the fifty-seven,thirty-one and sixty-six wavelengths suggested by CARS,SPA and GA,respectively,two classified models (least squares-support vector machine,LS-SVM and SIMCA)were established.Among the established models,SPA-LS-SVM model performed well withthe highest classification rate(100%)in calibration and 98% in prediction sets.SPA-LS-SVM and CARS-LS-SVM models obtainedbetter results 98% and 96% of classification rate in prediction set with thirty-one and fifty-seven effective wavelengths respectively.The CARS-SIMCA,GA-SIMCA and SPA-SIMCA models obtained poor results with 52% of classification rate in prediction set.The results showedthat NIR-HIS technique could be used to identify the varieties of fresh and frozen-thawed fish fillets rapidly and non-destructively,and SPA and CARS were effective wavelengths selection methods.