融合光谱和图像特征信息的羊肉TVB-N含量无损检测

Nondestructive Detection of TVB-N Content in Mutton Based on Fused Spectra and Image Information

为实现羊肉挥发性盐基氮(TVB-N)含量的快速、无损检测,利用可见-近红外光谱与机器视觉技术提取光谱特征信息和图像特征参数,将图谱特征信息融合后建立羊肉样品中TVB-N的最小二乘支持向量机(LSSVM)预测模型。采集贮藏1~15 d的73个羊肉样品在320~1100 nm波段范围的可见-近红外光谱和图像信息,参照国家标准方法测定样品中TVBN含量。采用竞争性自适应加权算法优选特征波长作为光谱特征信息,提取样品图像的颜色及纹理特征作为图像特征信息,通过特征层融合法将光谱信息与图像信息融合成总特征参数。分别基于3种特征信息建立羊肉TVB-N含量的LSSVM预测模型。结果表明:基于图谱融合信息建立的模型预测精度优于仅利用光谱信息或图像信息的建模结果,其验证集相关系数为0.930,标准分析误差为1.873 mg/100 g,相对分析误差为2.635。该结果证实基于融合图谱特征信息无损预测羊肉贮藏过程中TVB-N含量的可行性,为实现羊肉样品中TVB-N含量的定量、快速、无损、准确预测提供了参考方法。

To realize the rapid prediction of total volatile basic nitrogen(TVB-N)content in mutton,near-infrared spectroscopy technology and computer vision technology were employed to obtain the spectral information and image characteristic parameters,and least squares support vector machine(LSSVM)prediction model for TVB-N content was established based on fused spectra and image information.Spectral information in the range of 320-1100 nm and image information of 73 mutton samples storage at 4℃from 1 to 15 days were collected,and the TVB-N contents in samples were determined referring to the national standard method.Then the spectral characteristic information was extracted using competitive adaptive reweighted sampling algorithm,and the color and texture features were extracted as the image characteristic information.Finally,spectral information and image information were fused using feature layer fusion method,and LSSVM models were established based on three kinds of characteristics information,respectively.The results showed that the model based on fused spectra and image information yielded better prediction performance than that based on spectral information or image information merely,with correlation coefficient in the prediction set of 0.930,the standard analysis error of 1.873 mg/100 g,and relative percent deviation of 2.635.The results indicated that the feasibility of the prediction for TVB-N content in mutton based on spectra and image information.It provided an effective method for quantitative,rapid,nondestructive and accurate prediction of TVB-N content in mutton samples.