基于特征融合的猪肉新鲜度高光谱图像检测

Feature Fusion for Detection of Pork Freshness Based on Hyperspectral Imaging Technology

DOI：10.3969/j.issn.1673-1689.2015.03.004

中文关键词: 猪肉 高光谱图像 挥发性盐基氮 特征融合 特征降维 可视化检测

英文关键词: pork,hyperspectral image,TVB-N,feature fusion,feature dimension reduction,visual detection

基金项目:

摘要点击次数: 324

全文下载次数: 871

中文摘要:

利用高光谱反射图像技术研究了猪肉新鲜度的无损检测。采集了180个猪肉样本在400~1 000 nm范围内的高光谱反射图像,提取了高光谱图像的光谱均值和熵两类特征;分别利用连续投影算法、主成分分析,以及连续投影算法结合主成分分析3种特征降维方法,提取了反映肉类新鲜度信息的重要特征变量;并建立了这些特征变量与挥发性盐基氮(TVB-N)的最小二乘支持向量机(LSSVM)预测模型;在此基础上提出了猪肉TVB-N含量的可视化检测方法。研究结果表明:相比于单一特征模型,利用光谱均值和熵融合特征的LSSVM模型可显著提高模型的准确度;连续投影算法结合主成分分析的特征降维方法,可显著降低模型的复杂度,提高模型准确度。利用光谱均值和熵两类特征,通过连续投影算法和主成分分析相结合的特征降维方法所建立的LSSVM预测模型,可取得最佳的预测准确度,其预测集的均方根误差RMSEP为1.96,相关系数(RP)为0.948,剩余预测偏差(RPD)为3.12,可满足实际检测需要。建立在此基础上的可视化方法,可直观显示肉类的腐败区域和程度。

英文摘要:

It is of great importance for quick and nondestructive detection of meat freshness to ensure the quality of meat products and reduce the risks of food safety accidents. In this manuscript,the hyperspectral imaging technology has been used to study the nondestructive detection of pork freshness. The hyperspectral reflectance images between 400 and 1 000 nm of 180 pork samples were acquired. The mean and entropy feature were calculated from the hyperspectral reflectance image. Successive Projection Algorithm(SPA),Principal Component Analysis(PCA),and SPA combined with PCA were respectively used to extract the important feature variables which reflect well the characteristics of the meat freshness. And the Volatile Base Nitrogen(TVB-N) prediction models using Least Squares Support Vector Machine (LSSVM) were developed. On this basis,a visual detection method for the TVB-N content of pork was raised. The results demonstrated that:it can significantly improve the precision of model using the fusion of the mean and entropy features;and the feature dimension reduction method of SPA combined with PCA can significantly reduce the complexity of the model,while improving the model accuracy. The LSSVM prediction model using the fusion feature of mean and entropy based on dimension reduction method of SPA combined with PCA can obtain the best prediction accuracy. The Root Mean Square Error of Prediction (RMSEP) is 1.96,and the Relevance of Prediction (RP) is 0.948. The Residual Prediction Deviation(RPD) is 3.12,which meet the needs of the actual detection for pork freshness. The visualization detection method for pork freshness can display directly and clearly the area and degree of meat corruption.

查看全文 查看/发表评论 下载PDF阅读器