【作者】 金晶；

【导师】 廖桂平；

【作者基本信息】 湖南农业大学 ， 作物信息科学， 2009， 硕士

【摘要】 计算机视觉技术具有实时、客观、无损等优点,能对马铃薯表面外部品质进行快速检测。本文从图像获取装置、图像预处理、大小形状检测、绿皮检测、表面缺陷检测五个方面对单个马铃薯的外部品质静态检测进行了研究。图像获取装置和图像预处理是图像处理的基础环节,直接影响图像质量,进而影响识别的效果和检测的准确性。本文设计的马铃薯外部品质检测装置由照明设备、CCD数码相机、图像采集卡、计算机硬件和光照箱组成；采用B通道灰度化、中值滤波和Otsu分割法分别对马铃薯原始图像进行灰度化、图像平滑和阈值分割处理。利用马铃薯形状类似圆或椭圆这一特性对马铃薯大小形状进行检测。在传统系统标定方法的基础上,通过与基本矩形长和宽的比较,提出采用椭圆长短轴比作为其大小特征进行检测,其大小测量的误差率为7%。提出用提取椭圆长短轴比的方法来描述其形状特征,将马铃薯形状分成圆形、椭圆形和长筒形三类,形状检测结果为99.1%。为了检测绿皮马铃薯,论文介绍了一种基于色调域的阈值识别马铃薯绿皮的检测方法,实现了从量化角度提取马铃薯的表皮颜色信息,克服了统计的逐步判别分析方法和支持向量机SVM识别方法在构建模型时,因局限于特定样本集的特征空间的缺点,提取色调作为模式识别的特征值,并确定了区分正常和绿皮马铃薯的有效色调值区间57-64,再结合二次阈值分割方法对马铃薯的绿皮进行检测,准确率达到97.5%,且结果稳定。论文提出一套基于计算机视觉的检测马铃薯表面缺陷的新方法。使用自适应Ⅰ截留法或固定Ⅰ截留法能一次性将马铃薯表面的疑似缺陷分离出来,再结合OTSU法和形态学运算对疑似缺陷部位进行分割并分别提取面积和颜色特征,选取面积阈值和黑色比率阈值对疑似缺陷进行识别。经验证,缺陷正确分类率、缺陷正确识别率和马铃薯表面缺陷正确检测率分别为92.1%、91.4%和100%。更多还原

【Abstract】 Computer vision technology, which can detect some external characteristics of potato, has some advantages such as real-time, objectivity, and being nondestructive. The research based on computer vision includes five sections:the vision inspection device, image preprocessing, size and shape inspection, greened potato detection, external defects detection.Image acquisition and image preprocessing are fundamental steps in digital image processing which decide the quality of image and then can ensure the accuracy of recognition and inspection. The computer vision system developed to detect the defects of potato was composed of a CCD camera, lighting chamber, frame grabber and computer. B channel graying, median filtering method and Otsu segmentation are used in graying, image smoothing and threshold segmentation respectively.In consideration of the characteristics that the shape of potato is similar with circle and ellipse, on the base of traditional system calibration method with ping pong ball, by comparing the length and width of basic rectangle, the long and short axis of ellipse has been chose as characteristic values of size inspection with error rate of 7%. Similarly, the ratio of the long and short axis of ellipse has been chose as characteristic value of shape inspection to classify the shape of potato as round shape, oval shape and long cylinder shape with accuracy of 99.1%.This article recommended a method based on the hue region to detect green skin of potato, and extracted the color information from the quantitative perspective. The model based on the statistical Stepwise discriminated analysis and SVM has a certain bias which causes the results unsteadiness due to the limitations of characteristics space of a particular sample set. To overcome the foregoing shortcomings, the new method extracted hue as features, confirmed effective range of hue 57 to 64 which could distinguish between normal and greened potato. By combining the second threshold segmentation, the result showed that the accuracy of reorganization was 97.5% respectively with good stability.This paper reports a novel inspection approach to external defects of potato. Adaptive Intensity Interception (All) and Fixed Intensity Interception (FII) methods have been proposed to extract the suspect defects. Otsu segmentation combined with morphologic operation was used to remove the normal skin and background. Area threshold and black ratio threshold were used to identify defects in the suspect defects. Experiments have shown FII performed better than All in a specific circumstance. The correct classification rate of defects, the correct recognition rate of defects and the correct inspection rate of potatoes based on FII are 92.1%,91.4% and 100% respectively.更多还原