基于深度图像的毗邻罗汉果分割方法  

A method for segmenting adjacent siraitia grosvenorii based on depth image

作　　者：刘姣娣 张立昌 蒋水元[3] 许洪振 张邦 张小龙[1,2] LIU Jiaodi;ZHANG Lichang;JIANG Shuiyuan;XU Hongzhen;ZHANG Bang;ZHANG Xiaolong(Key Laboratory of Advanced Manufacturing and Automation Technology,Guilin University of Technology,Guilin 541006,China;College of Mechanical and Control Engineering,Guilin University of Technology,Guilin 541006,China;Guangxi Institute of Botany,Chinese Academy of Sciences,Guilin 541006,China)

机构地区：[1]桂林理工大学广西高校先进制造与自动化技术重点实验室,广西桂林541006 [2]桂林理工大学机械与控制工程学院,广西桂林541006 [3]中国科学院广西植物研究所,广西桂林541006

出　　处：《包装与食品机械》2024年第6期73-80,共8页Packaging and Food Machinery

基　　金：广西重点研发计划项目(桂科AB21238011)。

摘　　要：针对传统RGB图像处理算法无法分割毗邻罗汉果的问题,提出一种基于深度图像的毗邻罗汉果分割方法。利用罗汉果的圆度形状特性与藤蔓垂挂特性,从果实底部获取果实的深度图像,引入流体力学中源点的理论,提出将梯度向量看作运动矢量场,计算矢量场散度的方法,实现果实的粗定位。使用K-means++聚类算法对深度图像进行分割,提出一种基于定位点的连通域分割算法,实现对毗邻罗汉果单个果实的分割。结果表明,毗邻果实分割算法具有较快的处理速度和较高的分割精度,平均单张图像用时0.139 s,分割准确度为94.43%,精确度为94.20%,召回率为95.79%;与现有算法相比,准确率提高2.64%,处理速度提升88.5%。可在复杂光照环境下对单个和毗邻的果实进行分割,规避彩色图像处理的局限性。研究为罗汉果的机械化自动采摘提供技术支撑。To solve the problem that traditional RGB image processing algorithms can not segment adjacent Siraitia grosvenorii,a segmentation method of adjacent Siraitia grosvenorii based on depth image was proposed.By using the roundness shape characteristics and vine hanging characteristics of Siraitia grosvenorii,the depth image of the fruit was obtained from the bottom of the fruit.The theory of source point in fluid mechanics was introduced,and the gradient vector was regarded as the motion vector field,and the method of calculating the divergence of the vector field was proposed to realize the rough positioning of the fruit.K-means++ clustering algorithm was used to segment the depth image,and a connected domain segmentation algorithm based on location points was proposed to segment a single fruit adjacent to Siraitia grosvenorii.The experimental results show that the algorithm has fast processing speed and high segmentation accuracy.A single image takes 0.139 second,with a segmentation accuracy of 94.43%,an accuracy of 94.20%,and a recall rate of 95.79%.Compared to the current algorithm,the accuracy is improved by 2.64%,and the processing speed is improved by 88.5%.Single and adjacent fruits can be segmented in complex lighting environments to avoid the limitations of color image processing.The research provides technical support for mechanized automatic picking of Siraitia grosvenorii.

关 键 词：图像分割 深度图像 罗汉果 果实毗邻 梯度向量 

分 类 号：TS255.36[轻工技术与工程—农产品加工及贮藏工程]