融合高光谱成像多类特征的名优牛肉种类鉴别

Identification of famous beef species using multiple futures of hyperspectral imaging

提出一种基于高光谱成像技术鉴别名优牛肉的新方法。首先,测量采集了来自安徽、新疆和西藏三种牛肉的120个样本的高光谱图像。然后,通过感兴趣区域提取、加和平均获得样本的400 nm~1 000 nm波段的反射率光谱,并采用局部二值模式从样本灰度图中提取纹理特征。同时,通过标准化、多元散射校正(MSC)和标准正态变量校正方法去除反射率光谱中的噪声,利用连续投影算法(SPA)、竞争性自适应重加权采样算法以及稳定性竞争自适应重加权采样算法选择重要波长。最后,基于重要波长的光谱信息和纹理特征,采用支持向量机、随机森林和K最邻近算法(KNN)构建牛肉品种分类模型。实验结果表明,融合重要光谱信息和牛肉纹理特征后,基于SPA波长选择的MSC-KNN模型具有最好的分类效果,分类结果的校正集与预测集正确率分别为100%、99.14%。所提方法为实现基于多光谱系统与成像系统的快速、简便的牛肉种类鉴别装置奠定了理论与实验基础。

Hyperspectral imaging technology is a combination of imaging technology and spectral technology,which can detect the two-dimensional geometric space and one-dimensional spectral information of the target,and obtain the continuous and narrow band image data of hyperspectral resolution. In this study,the hyperspectral imaging( HSI) was adopted to identify species of famous beef. Firstly,the HSI images of the 120 beef samples from Anhui,Xinjiang and Xizang were measured. Then,reflectance spectra of 400 nm ~ 1 000 nm were extracted,and textural features were gained by local binary patterns. Normalization,multiple scattering correction( MSC) and standard normal variable were employed to remove the noise in spectra. And successive projection algorithm( SPA),competitive adaptive reweighting sampling( CARS),and stability competitive adaptive reweighting sampling( SCARS) were to choose the important wavelength. Random forest,support vector machine and k-nearest neighbor algorithm( KNN) were used to develop the calibration models for recognition of beef species. The results showed that the MSC-KNN model based on spectral information of important wavelengths and texture gained the best classification result with the accuracy of100% and 99. 14% in calibration and prediction set. The proposed method provides the possibility for the rapid and simple identification of famous beef species based on multi-spectral and imaging system.