计算机工程与应用2019，Vol.55Issue(15)：204-212,9.DOI:10.3778/j.issn.1002-8331.1905-0302
基于几何特征的灵武长枣图像分割算法
Lingwu Long Jujubes Image Segmentation Algorithm Based on Geometric Features
摘要
Abstract
The main purpose of this paper is to solve the problem of slight adhesion and shadow between objects in image. When the intelligent picking robot is used to pick the fruits of Lingwu Long Jujubes, there are problems of adhesion, occlusion, overlap and shadow in the target images of collected by the visual system in a natural scene, which will lead to misrecognition of objects in the image target recognition and be extremely unfavorable to intelligent picking. In order to solve this problem, a new algorithm for segmentation of Lingwu Long Jujubes image based on geometric features is proposed. According to the elliptical appearance of Lingwu Long Jujubes, a geometric model based on Lingwu Long Jujubes appearance is established through a large number of statistics on the shape characteristics of Lingwu Long Jujubes. The binary image is obtained by a series of image preprocessing, and then the relative centroid position of target object is obtained by continuous corrosion in morphological transformation, and these centroids are marked to determine the number of objects. The marked center is taken as the center of the model, the geometric model is established in the transformed binary image, and the boundary curve of the model is used to fit the segmentation line of the target object in Lingwu Long Jujubes image, so as to realize the segmentation of Lingwu Long Jujubes image. The experimental results show that the method can solve the problem of adhesion and shadow between objects and ensure high accuracy. The segmentation accuracy of the method can reach 92. 31% for the images with slight adhesion.关键词
灵武长枣图像/目标识别/图像分割/几何特征/图像粘连Key words
Lingwu Long Jujubes image/ target identification/ image segmentation/ geometric features/ image adhesion分类
信息技术与安全科学引用本文复制引用
赵琛,王昱潭,朱超伟..基于几何特征的灵武长枣图像分割算法[J].计算机工程与应用,2019,55(15):204-212,9.基金项目
国家自然科学基金(No.31660239). （No.31660239）
