Abstract: Traditional lamb quality detection has low efficiency and will destroy samples. In order to rapidly and non-destructively test tenderness of chilled fresh lamb, taking Xilingol lamb in Inner Mongolia as research object, the spectrum was corrected by multivariate scattering correction method, and feature images at 620.23, 761.48, 819.48 nm were obtained by principal component analysis method,  and their texture features and color features were extracted to establish BP neural network and support vector machine prediction models for lamb tenderness. The results showed that the prediction effect of the BP neural network model was better than that of the support vector machine model. The determination coefficient (R2) and root mean square error (RMSEP) of the BP neural network model on the prediction set were 0.85 and their 1.86, and those of support vector machine model were 0.77 and 2.37, respectively. Above results showed that it was feasible to use hyperspectral and image information feature layer fusion method to predict the tenderness of cold fresh lamb.

Key words: hyperspectral imaging technology, lamb, BP neural network, support vector machine, tenderness

摘要： 传统羊肉品质检测方法效率低、破坏样品，为实现冷鲜羊肉嫩度快速无损检测，以内蒙古锡林郭勒羊肉为研究对象，采用多元散射校正法对光谱进行校正，利用主成分分析法获得620.23、761.48、819.48 nm波长下的特征图像，并提取其纹理特征和颜色特征，分别建立羊肉嫩度的BP神经网络和支持向量机预测模型。结果显示，BP神经网络模型预测效果优于支持向量机模型预测效果，BP神经网络模型对预测集的决定系数（R2）和预测均方根误差（RMSEP）分别为0.85和1.86；支持向量机模型分别为0.77和2.37。研究表明，利用高光谱和图像信息特征层融合方法对冷鲜羊肉嫩度进行预测具有可行性。

关键词: 高光谱成像技术, 羊肉, BP神经网络, 支持向量机, 嫩度