枝上柑橘果实深度球截线识别方法

发布时间：2018-01-28 02:22

本文关键词： 柑橘采摘识别深度信息球截线特征提取　出处：《农业机械学报》2017年10期 　论文类型：期刊论文

【摘要】：针对柑橘果、叶、枝对象具有球体、片体和细柱体不同的三维几何特征,提出一种识别柑橘果实的深度球截线方法。首先提出了球形果实特征提取的深度球截线方法的基本原理和关键参数,进而分别针对枝上果、叶孤立和贴碰区域提出了孤立果实的特征提取算法和贴碰果实的特征提取算法,得到了复杂枝环境下的深度数据处理与果实识别策略,并综合根据Intel Real Sense F200型深度传感器参数、柑橘果实尺寸、近景探测范围、数据预处理与特征提取需要完成了深度球截线方法的参数确定。大量室内试验结果表明,深度球截线方法对孤立果实提取的平均成功率为97.8%,贴碰区域内果实提取的平均成功率为76%,而复杂枝环境的果实提取综合成功率为63.8%。该深度球截线的识别方法仅利用有限的深度数据点,在保证原始数据精度的同时降低了运算量和果实特征提取复杂性,能有效应对果叶遮挡问题,实现对贴碰果叶的有效区分,对柑橘果实具有良好的适应性,为采摘机器人在复杂环境下的果实识别与定位提供了新的技术思路。
[Abstract]:Citrus fruits, leaves and branches have different 3D geometric characteristics of sphere, lamina and fine cylinder. In this paper, a method to identify the depth spherical section of citrus fruit is proposed. Firstly, the basic principle and key parameters of the deep spherical truncation method for extracting the characteristics of the spherical fruit are presented, and then the fruit on the branch is separately targeted. In this paper, the feature extraction algorithm of isolated fruit and the feature extraction algorithm of fruit are proposed, and the deep data processing and fruit recognition strategy under complex branch environment are obtained. And according to the parameters of Intel Real Sense F200 depth sensor, citrus fruit size and close-range detection range. Data preprocessing and feature extraction need to complete the determination of parameters of the depth spherical truncation method. A large number of laboratory experiments show that the average success rate of the deep spherical truncation method for isolated fruit extraction is 97.8%. The average success rate of fruit extraction was 76%, while that of fruit extraction in complex branch environment was 63.8%. In order to ensure the accuracy of the original data and reduce the complexity of operation and fruit feature extraction, it can effectively deal with the fruit leaf occlusion problem, achieve the effective discrimination of the fruit leaves, and have a good adaptability to citrus fruit. It provides a new technique for fruit recognition and location of picking robot in complex environment.
【作者单位】：江苏大学现代农业装备与技术教育部重点实验室;
【基金】：国家自然科学基金项目(51475212) 江苏省自然科学基金项目(BK20151339) 江苏省高校自然科学研究重大项目(16KJA210002) 江苏省高校优势学科建设工程项目(PAPD)
【分类号】：TP242;TP391.41
【正文快照】： 引言识别定位是采摘机器人的核心技术,而快速准确地提取采摘目标特征是果实识别的前提[1-3]。由于传统的基于可见光图像识别的定位方法在实时性、重叠对象分割和光照敏感性上存在着局限性,研究者陆续进行了其他途径的探索,其中利用深度信息进行果实的识别定位成为研究热点。近

【相似文献】