猪肉水分含量线性、非线性定量分析模型比较研究

打开文本图片集

Abstract： Inordertocompare thediference inaccuracybetweenquantitativepredictive models forthemoisturecontent in pork developedbylinearandnon-linearalgorithms，the linear variable flter near-infrared （LVFNIR）spectraof whole pieces of pork and minced pork were collectedbya portable LVFNIR spectrometer，and partial least square regresson （PLSR）as a linear algorithm and support vector regression （SVR）as anon-linear algorithm were used todevelopquantitative predictive models forthe moisturecontents in the two meat samples.Optimizationof wavebands anddata preprocessng methods were conductedforeachofthefourmodels （wholepiecesofporkPLSRmodel，mincedporkPLSRmodel，whole piecesofpork SVR model，minced porkSVRmodel）.Theresultsshowedthatthe SVR model forminced porkbasedonfullbandspectral datawith data centralization + smoothing + standard normal variate transformation preprocessing performed best among all themodels tsted，withpusnteent，erelramete，eteiatioeientofalbaioootmar error ofcalibration，determination coeficientofrossvalidation，root meansquareerorofvalidationandratioperformance deviationof14400，0.020，0.7787，1.04，0.8714，0.89and2.69，respectively.The predictiveresultsfortheeteal validationset （externalblindsamples）demonstrated that theSVRmodelfor mincedpork hadthebestperformance，with root mean squareeroofpredictionand corelationcoefficientof predictionof1.24and0.821O，respectively.This study indicatedthatthe SVR modelshad higher predictive accuracy than did the PLSR models forboth pork samples.However， thestabilityoftheSVRmodelsstillneedsfurtherimprovement.

Keywords： linear variable filter near-infrared spectroscopy; quantitative model; moisture content; pork

DOI：10.7506/rlyj1001-8123-20250520-160

中图分类号：O657.33；TS251.7 文献标志码：A 文章编号：1001-8123（2026）04-0061-08

随着人民生活水平的提升，消费者对猪肉品质的要求日益提高。（剩余15591字）