基于高光谱成像技术的羊肉挥发性盐基氮检测方法研究

Detection Method of TVB-N Content in Mutton Based on Hyperspectral Imaging Technology

挥发性盐基氮(TVB-N)含量是评价羊肉新鲜度的重要指标,传统的感官评价方法主观性较强,尝试采用可以对农畜产品无损检测的高光谱图像技术。以内蒙古锡林郭勒羊肉为研究对象,使用标准正态变量方法(Standard normal variate,SNV)校正其原始光谱图像。采用二维主成分分析法(2D-PCA)对原始高光谱图像数据降维,优选出1 257.49 nm、1 396.19 nm、1 736.15 nm波长下的特征图像。对特征图像提取纹理和颜色特征参数共计54个,筛选出与TVB-N含量相关性高的特征参数,作为人工神经网络(BP-ANN)和偏最小二乘回归(PLSR)模型输入量,构建羊肉TVB-N含量的预测模型。结果发现,BPANN模型对预测集的决定系数R2为0.86,预测均方根误差为3.33,PLSR模型对预测集的决定系数R2为0.81,预测均方根误差为3.96,BP-ANN的预测效果明显优于PLSR模型。结果表明,利用高光谱成像技术进行羊肉TVB-N含量快速无损检测具有较高的可行性。

The TVB-N content is an important indicator for evaluating the mutton freshness. The traditional sensory evaluation method is strongly subjective. This paper aimed to develop a new detection method based on the hyperspectral image technology,which can be non-destructively detect agricultural and livestock products. The xilingol mutton from Inner Mongolia was used as the research object,and its original spectral images were corrected by using Standard normal variate(SNV). The Two-dimensional principal component analysis(2D-PCA)was used to reduce the dimensionality of the original hyperspectral image data,and the characteristic images at the wavelength of 1 257. 49 nm,1 396. 19 nm and 1 736. 15 nm were optimized. A total of 54 texture and color feature parameters were extracted from feature images. The characteristic parameters with high correlation with TVB-N content were screened out as the input of artificial neural network(BP-ANN)and partial least squares regression(PLSR)model. The prediction model of TVB-N content was then constructed. The results showed that the correlation coefficient of BP-ANN model was 0. 86,and the predicted root mean square error was 3. 33. The correlation coefficient of the PLSR model was 0. 81,and the predicted root mean square error was 3. 96. The prediction effect of BP-ANN was significantly better than that of the partial least squares regression(PLSR). The results indicated that using the hyperspectral imaging technology for the rapid nondestructive detection of mutton TVBN content had a high feasibility.