基于毛孔尺度特征分析的带皮猪肉追溯研究

发布时间：2019-02-16 09:54

【摘要】：猪肉是我国居民的主要肉食品种,虽然政府付出了大量的人力物力来对猪肉生产的各个环节进行监管,但是近年来还是出现了一系列严重的猪肉食品安全事故。目前我国猪肉的追溯主要依靠猪皮上的合格印章和各个环节工作人员进行的信息记录,并没有依靠猪肉自身的特征来实现猪肉的追溯。毛孔作为带皮猪肉自身所具备的特征在猪皮上是普遍存在的,并且在流通过程中能够基本保持性状的不变。本研究致力于探索和提取猪皮图像中毛孔所携带的具有唯一性的特征信息,并基于此来进行局部猪皮图像与总体猪皮图像的匹配从而实现带皮猪肉的追溯。本文的主要研究内容和结论如下。(1)对高清猪皮图像的毛孔特征提取的研究。针对本研究的目标应用场景到市场上购买猪皮样本,并进行高清猪皮图像的采集。将采集的高清猪皮图像作为研究对象,对猪皮图像中的猪皮毛孔进行建模。根据猪皮图像的毛孔特性和目标应用场景,本研究在毛孔特征提取算法PSIFT(Pore Scale Invariant Feature Transform)的基础上提出了多方向的毛孔特征提取算法MPSIFT(Multi-orientation Pore Scale Invariant Feature Transform),使得最终所提取的毛孔特征具有一定的旋转不变性。利用MPSIFT算法对猪皮图像的毛孔进行检测和特征提取,最终为每一个毛孔特征点至少生成一个特征点描述向量。(2)局部猪皮图像与总体猪皮图像的匹配研究。基于MPSIFT算法提取的高清猪皮图像的毛孔尺度特征,本文进行了局部猪皮图像与总体猪皮图像的匹配研究。本研究利用特征点描述向量的夹角作为猪皮图像毛孔特征点的相似度度量,利用不同特征点描述向量与被匹配特征点描述向量夹角大小的比值来衡量特征点之间的匹配度。在两张猪皮图像的毛孔特征匹配完成之后,本文利用离群点检测的方法对匹配成功的特征点对进行筛选,剔除错误匹配的特征点对,从而保证匹配成功的毛孔特征点对的质量。(3)基于局部猪皮图像与总体猪皮图像的匹配研究本文提出了基于毛孔尺度特征分析的带皮猪肉追溯技术,并给出了带皮猪肉的追溯原理与该技术的应用场景。(4)猪皮图像毛孔特征提取与匹配的实验与分析。本研究通过实验探明了不同数据特性对局部猪皮图像与总体猪皮图像匹配效果的影响,并最终在所采集的全部猪皮图像中实现了94.14%的局部猪皮图像与总体猪皮图像的匹配正确率,在保证追溯成功结果的可靠性情况下实现了局部猪皮图像86.73%的追溯成功率。
[Abstract]:Pork is the main meat food in our country. Although the government has paid a lot of manpower and material resources to supervise the pork production, there have been a series of serious pork food safety accidents in recent years. At present, the traceability of pork in our country mainly depends on the qualified seal on the skin of the pig and the information recorded by the personnel in every link, and it does not depend on the characteristics of the pork itself to realize the tracing of the pork. Pores, as the characteristics of skinned pork, are common in pig skin, and can basically maintain the same characters in the circulation process. The purpose of this study is to explore and extract the unique feature information carried by pores in porcine skin images and to match the local porcine skin images with the overall pig skin images so as to achieve the traceability of skinned pork. The main contents and conclusions of this paper are as follows: (1) the study of pore feature extraction from high-definition porcine skin image. According to the target of this study, the pig skin samples were purchased from the market, and the high-definition pig skin images were collected. The porcine pores in the high-definition pig skin images were modeled. Based on the pore characteristics of porcine skin image and the target application scene, a multi-directional pore feature extraction algorithm, MPSIFT (Multi-orientation Pore Scale Invariant Feature Transform), is proposed based on the pore feature extraction algorithm (PSIFT (Pore Scale Invariant Feature Transform). The resulting pore features have a certain rotation invariance. MPSIFT algorithm is used to detect and extract pores of porcine skin image, and at least one feature point description vector is generated for each pore feature point. (2) matching between local pig skin image and total pig skin image. Based on the pore size feature of high-definition pig skin image extracted by MPSIFT algorithm, the matching between local pig skin image and total pig skin image is studied in this paper. In this study, the angle of feature point description vector is used as the similarity measure of porcine skin image pore feature point, and the ratio of different feature point description vector and matching feature point description vector angle is used to measure the matching degree between feature points. After the pore feature matching of two porcine skin images is completed, the outlier detection method is used to screen the matching feature pairs, and the false matching feature pairs are eliminated. So as to ensure the quality of the matching pore feature points. (3) based on the matching between local porcine skin image and total porcine skin image, a new technique of pork traceability based on pore size feature analysis is proposed in this paper. The traceability principle of skinned pork and the application scene of this technique are also given. (4) the experiment and analysis of porcine skin image pore feature extraction and matching. In this study, the effects of different data characteristics on the matching effect between local pig skin image and total pig skin image were investigated. Finally, 94.14% of the local pig skin image and the total pig skin image are matched correctly in all the pig skin images collected. The success rate of local pig skin image was 86.73%.
【学位授予单位】：西北农林科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TS251.51;TP391.41

【参考文献】