【作者】 王芳；

【导师】 文友先；

【作者基本信息】 华中农业大学 ， 农业电气化与自动化， 2013， 博士

【摘要】 皮蛋是我国独创的一种蛋类加工产品,其营养丰富、容易消化、具有清凉解热等作用,深受人们喜爱。目前在生产加工中主要采用眼看、手掂等人工方法来检测并剔除破损皮蛋,本文结合偏振光技术、图像处理技术及模式识别技术,研究出缸皮蛋蛋壳裂纹的自动识别方法。主要研究内容如下：(1)搭建了自然光图像采集系统,对图像的噪声滤除、图像增强、边缘检测、形态学运算等进行了深入分析和比较,研究出适合提取皮蛋蛋壳裂纹的图像处理算法。通过大量试验验证,提出了形态学边缘提取算法,试验结果表明：形态学边缘提取算法能有效区分皮蛋蛋壳信息和裂纹信息。(2)在形态学边缘提取算法处理蛋壳图像的基础上,分别提取了图像中裂纹信息的形状特征参数和纹理特征参数,具体为：裂纹长度L、裂纹深度K、投影变换后水平方向和垂直方向的灰度最大值Tmax、角点数等四个特征参数。采用支持向量机分类器进行自动识别,分别采用线性、多项式、径向基、Sigmoid等四种SVM核函数分别对样本进行分类处理,根据不同的训练样本和测试样本数,综合得出径向基核函数SVM分类器的识别率最高,检测率最高可达94%。(3)针对皮蛋蛋壳表面大块黑斑和裂纹的偏振度的不同,研制了偏振图像采集系统,根据本课题研究的需要选用了Stokes矢量法描述偏振光,根据Stokes矢量描述法中所需测量的偏振角为0°、45°、90°、-45°的偏振图像,搭建了偏振图像采集系统,详细介绍了偏振图像采集系统的硬件设计和软件设计,其中包括软件界面的设计、界面可直接控制CCD的各个参数设计、采集图像、控制电机自动旋转所需角度、自动获取所需要的各角度偏振图像等多种操作,整个图像采集系统能完成控制电机旋转一周,能自动采集相应的偏振图像。(4)利用偏振图像采集系统采集样本偏振角为0°、45°、90°、-45°的偏振图像,拟合各样本的偏振图像,通过试验数据表明,皮蛋蛋壳表面大块黑斑的偏振度相对稳定,大多在0.1左右,而裂纹的偏振度不太稳定,但大多数情况下一般大于表壳黑斑的偏振度,本文先采用阈值方法滤除样本表壳黑斑信息,然后以偏振图像中像素最高且连通区域最大部分作为中心,截取100*100的像素图像,提取该图像的裂纹长度、均方比、偏度和峰度等四个特征参数,采用Kmeans聚类分析算法识别蛋壳裂纹。通过实验证明,该方法效果良好,其中好壳皮蛋的识别准确率为100%、裂纹蛋的识别准确率为88.3%,综合准确率为93%。(5)对同一批100枚皮蛋样本分别用自然光图像处理方法和偏振光技术进行样本裂纹识别,其检测结果均为91%,但自然光图像处理方法中好壳蛋误判率较高,而偏振光技术裂纹蛋的误判率较高,最后将所有样本先经过自然光检测,再经过偏振光技术识别,其检测率高达96%,说明多特征融合算法提高了检测精度。本文结果表明该方法能有效地识别皮蛋蛋壳的裂纹,具有较高的应用价值,可作为工业自动检测装备研制的重要理论依据。更多还原

【Abstract】 Pidan or century egg also known as preserved egg is one of the most traditional and popular preserved egg products in China. Pidan is widely adopted into the local cuisine as an appetizer, its nutritional value and a pleasant fragrant taste that is preferred by most people in Southeast Asian countries. The traditional detecting crack of pickled egg was based on manual work in many factories. On the basis of analyzing relevant researches, This dissertation presented a method of detecting preserved eggshell cracks automatically based on image processing techniques and polarized optical technology in the experiment. The main contents are as follows:(1) Set-up image acquisition system on ordinary source, a series of image processing foundational algorithms, such as image denoising, image inhancenment, edge detection operators and morphology calculus, were analyzed in this work. It is configured in these ways by selecting different image processing algorithms in each stage. According to experiments and desired results obtained from this experiment, it was found that morphology calculus algorithms was the most effective way to detect cracks on preserved eggshell.(2) Some shape feature parameters and texture parameters were extracted after the image processing with morphology calculus algorithms. The four characteristic parameters were the crack length L, the crack degree coefficient K, the maximum between horizontal and vertical gray Tmax, the angler points. The inact and cracked preserved eggs was classified with SVM model. A comparison study of the classification capacibility of different kernel function of SVM model, training sample size and test smple size, it was found that the SVM model with RBF kernel function procided the best classification performance and the classification accuracy was up to94%.(3) For the different polarization degree of black spots and cracks on preserved eggshell, a polarization optical system was designed in this work. It accepted the Stokes Vector method to describe polarization image. Using the polarization optical system acquired different polarization angles images, including degree of zero, degree of forty five, degree of negative forty five and degree of nintey. It detailed introduced the hardware and software design of the polarization optical system. The image of different polarization angles was acquired by analyzer rotating automatically, while the analyzer was controlled by stepper motor. Every angle of analyzer rotating was obtained automatically by computer controlling the stepper motor using DLL (dynamic link library) and instructions.(4) According to different depolarization mechanism for black spots and cracks on preserved eggshell, it can distinguish black spots and cracks on polarization image. The polarization image after pretreated processing, it took the most connected area of high gray value as center to cut an image about100*100pixel area. Four characteristic parameters was extracted to distinguish the cracks on preserved eggshell, including the length of the crack as L, mean variance ratio as R, skewness as α and kurtosis as β. We put forward four characteristic parameters and use cluster analysis to detect cracks. This paper employed Info-Kmeans clustering algorithm, and the clustering of high-dimensional sparse data extract from images. The result showed that all preserved eggs were classified into intact and cracked groups with the Info-Kmeans clustering algorithm, and the accuracy was up to93%.(5) Comparing two different identification methods on the same samples, a fusion based on multi-recognition technology and multi-feature method was proposed. The results showed that the total classification accuracy rate was up to96%.It showed that image processing techniques and polarized optical technology distinguished intact and cracked preserved eggs efficiently, and these results could provide a theoretical basis for detecting cracks of preserved egg on product lines.更多还原