Abstract: The use of computer vision technology for potato postpartum quality grading has very important significance. First image analysis is carried out for the potato image obtained by computer vision system. The Otsu method is used to remove the background of the potato image, then the image processing and recognition is carried out for the potato states of damage, green skin and sprout. For the green skin potato, the Perceptron Learning Algorithm（PLA） is used to distinguish the normal potato and green peel potato. For the epidermis germinative potato, the edge detection method is used to get the image edge of potato in the area of each part. The K-Nearest Neighbor（KNN） classification algorithm is used to recognize the epidermis germinative potato, and the germination area profile is obtained by corner detection. Then for the edge detected, the median filter and the area of the solution are used to determine the potato skin injury, finally realizing the classification of potato quality. The potato quality test results：the recognition accuracy of normal potato is 96.8%, green peel potato is 89.7%, the skin injury potato is 90.4%, and the sprouting potato is 96%.

Key words: computer vision, potato classification, quality detection

摘要： 利用计算机视觉进行马铃薯产后品质检测分级有十分重要意义。首先对计算机视觉系统获取的马铃薯进行图像分析，利用Otsu法去除马铃薯图像背景，然后针对马铃薯的损伤、绿皮和发芽状态进行图像处理识别；针对马铃薯中的绿皮状态，利用感知器学习算法（PLA）区分正常马铃薯与绿皮马铃薯；然后针对表皮发芽的马铃薯，利用边缘检测法得到图像中马铃薯区域的各部位边缘，结合K-最近邻分类算法（KNN）识别表面发芽的马铃薯，同时通过角点检测确定轮廓上的发芽区域；然后对检测到的边缘利用中值滤波结合面积最大法，确定马铃薯表皮的损伤部位，最终实现马铃薯品质的分级。利用计算机视觉方法马铃薯品质检测实验结果：正常马铃薯识别正确率为96.8%，绿皮马铃薯为89.7%，表皮损伤马铃薯为90.4%，发芽马铃薯为96%。

关键词: 计算机视觉, 马铃薯分级, 品质检测