中文摘要：

针对基于DSP的小型农产品机器视觉分选机，设计了一种基于图像识别的桂圆破损检测方法。该方法首先提取桂圆的彩色分量R和G，通过R—G特征值分割提取桂圆外壳区域，并求出桂圆观测面积；然后通过外壳边缘求出其近似最小外接圆的面积；最后根据观测面积和最小外接圆面积的比值来判断桂圆的破损情况。采用桂圆外壳破损和完好的共792个样本进行分选测试，结果表明，分选的准确率达到86．24％。该检测方法简单快速，有一定的准确性，适用于大多数小型球状果实的分选。

英文摘要：

A damage longan rapid detection method based on image recognition was designed for small agricultural products vision sorting machine based on DSP. The method first extracted the color components R and G of longan image, according to the characteristics of R-G value segmentation longan shell region, and found the longan observation area; secondly, with the help of shell edge, calculated the approximate area of the minimum circumscribed circle. Finally, according to the ratio of observation area and minimum circumscribed circle area to determine the longan＇s damage. A total of 792 longan samples contained broken and intact shell were tested, the results showed that, the sorting accuracy rate reached 86.24%. The detection method is simple, fast and has certain accuracy, and it is also suitable for the separation of most small spherical fruits.