【作者】 李志；

【导师】 孙永海；

【作者基本信息】 吉林大学 ， 食品科学， 2008， 硕士

【摘要】 牛胴体眼肌切面是进行牛肉等级评价的重要部位。本文采用机器视觉技术和颜色传感器技术,对牛胴体眼肌切面处牛肉分级信息自动检测技术展开研究。本文研究的主要内容如下:1.通过对当前常见的图像采集设备的比较,从课题的实际出发,筛选出适宜的图像采集设备,并采用VFW技术设计和开发实时图像捕捉程序,完成实时图像的获取工作。2.根据牛胴体眼肌切面图像特征,设计出相应的图像处理流程,研究分级信息的自动检测方法,并设计图像处理效果的评价指标,对本文设计方法的处理能力进行评价。3.采用颜色传感器对肌肉、脂肪的颜色等级标准图板颜色值进行测量,采用多元回归分析法、RBF神经网络、简易区间法三种方法建立颜色等级判别模型,并使用最优模型对眼肌切面肌肉、脂肪的颜色等级进行评定。设计开发基于颜色传感器的颜色信息自动检测程序,实现眼肌切面处肌肉、脂肪的颜色等级自动评价。4.组建基于USB接口摄像头(机器视觉)及颜色传感器的牛胴体眼肌切面分级信息自动检测系统,用计算机技术代替人工测量,实现牛肉分级信息检测自动化。更多还原

【Abstract】 Rid eye muscle of beef carcass contains a lot of characteristic informantions of beef quality, is the important part of beef grading evaluation. In the traditional beef processing industry, these grading informations are measured by manual measurement or visual observation commonly. The manual measurement methods are dependent on the individual experience and personnel mentality, and disadvantaged in considerable error, low efficiency and bad approbation, which will impact the entire link of beef grading evaluation if too serious. Based on the status of Chinese beef grading, a method was presented to replace the manual methods using computer technology, and the new method of measuring the classification information on rid eye muscle by using machine vision and color sensor technology was studied on, so as to achieve automation of beef grading information detection, and to complete initial development of beef Information online automatically detection grading system.The mean contents and conclusions are as following:1.From the fact of study, CMOS camera with an USB interface was choiced as the real-time image acquisition devices of the detection system. Based on By VFW in the video capture and data acquisition principle and by the Visual C++ 6.0 development environment, the real-time image acquisition process has been designed and developed, so as to dispense with the image acquisition card, reduce the costs, and enhance the flexibility of image acquisition devices, provide the real-time image for next processing.2.A deep study on beef carcass image segmentation algorithm has been conducted: first, a pretreatment work on rid eye muscle was done using light compensation, noise removal and image enhancement; then component— iterative—threshold segmentation was used to remove the complex background, and the image was conversed from color one into black-and-white binary one—through these steps the effective rid eye region would be extracted mainly, and be indeed precise extracted after series steps such as contour extraction, predefined seed-connectivity growth, filling empty threshold of. Based on the effective rid eye region image, the beef marbling regions were extracted out by optimum threshold, gray value subtraction mutually.3.Through the image segmentation process, three grading information are measured: statistic work for the number of black pixels from the effective rideye binary image is done in order to measuring the area of the effective rideye; combining the result image whose complex background has been removed and the rid eye regional image, the 3/4 point of the effective length was first found out, and then was made as a starting point to find the corresponding backfat fat from the brink, the backfat thickness finally was measured; regional labeling method was used to demarcate each fat region of the beef mabling binary image, a vector structure M r ( Ft,V,X,Y) was designd to record the result, which can change the marbling from image information into digital information and complete the measurement for the characteristics of beef marbling.4.Effective extraction rate, extraction accuracy rate and response time have been designed as three indicators characterization of the degree of image processing, and provides that: if effective extraction rate is more greater, extraction accuracy rate mose closer to 100%, the response time mose shorter (that is, the smaller consumption), the algorithm will be thought as more gifted. Through samples at the scene, effectiveness of the section segmentation methods referd were tested: all rid eye region effective extraction rate reaches more than 92 percent, and most of the effect rate achieve more than 91 percent, response time of image processing is short, the total time is only 991 to 1124 ms. The design of image processing methods for beef carcass images in this paper, has a strong analytical skill.5.The color values of the color grade standard charts are measured by color sensor. Study on the relationship between the color values and the corresponding grade, and establish discriminating model to actualize no manual process for muscle and fat color grading using nonlinear regression multi-analysis, RBF neural work and simple interval discriminating analysis. The accuracy rates of the models for muscle color are all 100%, and the accuracy rates of the models for fat color are 92%, 100%, 100% respectively. Through the comparison of different grading models, simple interval method is simpler, faster, more convenient, and of high efficiency advantages, of a good assessment of the habit. The simple interval model is choiced for muscle, fat color grading of beef carcass rideye.6 . Design and develop a automatically detects the color information procedures for color sensor, design mainly a collection interface and serial communication, data acquisition modules, and carry the color grading model into the process, realize the automatic evaluation for the color information on a beef carcass rid eye.7.The preliminary development of the online automatically measuring system for the grading information has been carryed out. Design the hardware and software parts of the system. The hardware mainly includes: digital camera with USB interface, color sensor and the computer hardware. In the choice of hardware, the tenet was to meet the performance requirements of simple structure, and was of low-cost convenience. The major software includes: real-time image acquisition module, image processing module, color sensor response modules and system interface. Multithreading technique is used to connect the functional modules together, and to support the System‘s multi-task work.更多还原