本文研究了自然场景下不同颜色葡萄果实的识别方法，针对晴天顺光、晴天逆光及夜间三种光照情况下采集的葡萄图像进行处理，在图像预处理的基础上提出了基于不同颜色空间的图像分割方法，利用最大类间方差法和直方图双峰法分别获取最佳阈值，得到葡萄串的轮廓图像，实现目标果实和复杂背景区域的分割。根据目标果实轮廓图像绘制其最小外接矩形，并利用Harris角点检测法提取果实重心及采摘点坐标等特征信息。实验结果表明：葡萄识别算法可对图像中果实部分快速准确地识别出来，并在很大程度上降低了光照强度等因素对果实识别效果的影响，其中绿色葡萄在晴天顺光、晴天逆光、夜间的识别率分别为93.3%、86.7%、96.7%，紫色葡萄在三种光照情况下识别率分别为90%、83.3%、96.7%。

This paper is mainly about how to recognize different color grapes in nature, The grape images that taken under a fine day of front lighting, back lighting and night conditions, A novel grape image segmentation method based on different color space was proposed. The maximum classes square error method and the histogram method were used to obtain the best threshold, to get the grape bunches contours images and to cut the complex background region apart. The minimum enclosing rectangle was drawn on account of the image contour of target grapes, the feature information of grape barycenter and the plucking positions were extracted by the Harris corner test. The results showed that the identification algorithm of grapes could accurately identify the grapes in the picture and reduce impacts of illumination intensity in a large extent, among which the recognition ratio of green grapes in front lighting, back lighting and night conditions were 93.3%、86.7% and 96.7%, while the recognition ratio of purple grapes were 90%、83.3% and 96.7%.

国家科技支撑计划项目(2015BAD19B03)；大学生创新创业训练计划项目(201410759037)