Abstract: In traditional watershed algorithm, there are all kinds of problems with image segmentation due to improper threshold, especially the over-segmentation problem. Based on the algorithm, by using the image of Lingwu Long Jujubes as the research object, the genetic algorithm is used to optimize the threshold that is selected randomly. For the 20 images of Lingwu Long Jujubes in the natural light environment, the improved watershed algorithm is used to segment the images. Firstly, based on the traditional watershed algorithm, in this paper, the experiment searches for the optimal threshold by using the genetic algorithm, and obtains the optimal segmentation threshold of images, and then the images are reprocessed with the OTSU method and mathematical morphology. Finally, the experiment obtains the corresponding images that are segmented. Images with new segmentation algorithm are compared with those obtained by artificial segmentation, the segmentation accuracy can reach 89.99% and the segmentation effect is far superior to the traditional watershed algorithm. The experiments show the method can obtain the optimal segmentation threshold and can meet machine recognition’s the requirement to image segmentation.

Key words: Lingwu Long Jujubes, image segmentation, selection of threshold, watershed algorithm, genetic algorithm

摘要： 传统分水岭算法常常会因阈值选择不当而导致图像分割出现各种各样的问题，尤其是过分割问题。在传统分水岭算法的基础上，以灵武长枣图像为研究对象，运用遗传算法对随机选取的阈值进行优化选择；对自然光照环境下的20幅灵武长枣图像，采用改进后的分水岭算法对其进行分割。首先在传统分水岭算法的基础上，利用遗传算法对阈值进行寻优，得到最优的图像分割阈值，再利用最大类间方差法和数学形态学等方法对图像进行后处理，最终得到分割图像，将分割图像与人工分割得到的图像进行比较，分割的正确率能达到89.99%，且分割效果远远优于传统分水岭算法。实验表明，该方法能够得到最优分割阈值并且能够满足机器识别对图像分割的要求。

关键词: 灵武长枣, 图像分割, 阈值选择, 分水岭法, 遗传算法