基于高光谱图像的猪肉品质检测和等级分类理论与技术

项目名称： 基于高光谱图像的猪肉品质检测和等级分类理论与技术

项目编号： No.61303116

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 自动化技术、计算机技术

项目作者： 曾山

作者单位： 武汉轻工大学

项目金额： 23万元

中文摘要： 传统的猪肉品质检测采用人工感官评定法和标准的理化方法，感官评定法劳动强度大、主观性强，理化方法需要破坏样品，耗时长，不适合大规模的在线生产检测。本课题开展利用高光谱图像技术检测猪肉品质的研究，通过高光谱图像技术采集反应猪肉外部感官品质(颜色、形状、纹理等)的图像信息和反应内部理化品质(PH值、TVB-N、滴水损失、脂肪含量等)的光谱信息，采用三维主成分分析法去除高光谱图像数据中的冗余信息，结合猪肉各项品质（嫩度、新鲜度等）的光谱学机理特征，探讨采用流形学习等方法寻找和提取最能表征猪肉各项品质的特征波段图像及特征，构建猪肉品质高光谱数据库和专家知识库，设计FCM和SVM相结合的挖掘分类算法对其进行知识和规则挖掘，揭示猪肉各项品质及相应的理化指标与其高光谱图像之间的关系，建立猪肉品质理化指标的定量检测模型和品质综合评判模型。本课题为猪肉品质快速无损检测和分级提供了思路，具有重要意义。

中文关键词： 高光谱图像；猪肉质量检测；张量分解；模糊 C-均值聚类；

英文摘要： Traditionally, the detection of pork quality is implemented by means of artificial sensory evaluation method or the standard physical and chemical method. The sensory evaluation method is labor-intensive and very subjective, while physical and chemical method often involves the destruction of the sample, which takes longer time and is not applicable to large-scale online production testing. The subject we carry out is a study on detecting pork quality using hyperspectral image technique. The technique is carried out through the collection of the image information that reflect the pork external sensory quality (such as pork's color, shape, texture, etc.) and the spectral information that reflect the internal physical and chemical quality (such as pork's PH value, TVB-N, drip loss, fat content, etc.). With the combination of the pork quality's spectroscopy mechanism characteristics, We can make use of the three-dimensional principal component analysis method to remove the redundant information in hyperspectral image data. Then, we explore and extract characteristic band image that most indicate the quality of the pork (such as tenderness, freshness) using supervised local manifold learning method. We establish multiple regression models based on hyperspectral image technique. We fuse the unsupervised FCM algorithm

英文关键词： Hyperspectral image；pork quality inspection；tensor decomposition；fuzzy C-means clustering；