Abstract Organs classification and fruit counting on pomegranate trees are of great significance for horticulture works and robotic picking. However, there are still some challenges: (1) illumination is uncontrollable in the natural environment; (2) traditional 2D image-based methods for classification and recognition are limited by occlusion on pomegranate trees. In this paper, a method for organs classification and fruit counting on pomegranate trees based on multi-features fusion and Support Vector Machine (SVM) was proposed. It was constructed by the following steps: (1) Three-dimensional point clouds of pomegranate trees were obtained by an RGB-D camera; (2) Three-dimensional point clouds were preprocessed; (3) Color and shape features were extracted to train the SVM classifier; (4) The obtained classifier model was used for organs classification on pomegranate trees; (5) A K-nearest neighbor (KNN) smoothing based on weighted Euclidean distance was used to improve the accuracy of classification; (6) An agglomerative-divisive hierarchical clustering was used to count pomegranate fruit. The experiment results showed that the SVM classifier based on color and shape feature had an accuracy of 0.75 for fruit and 0.99 for non-fruit. The fruit counting based on agglomerative-divisive hierarchical clustering had a recall of 87.74 % and a precision of 78.15 %. Compared with density-based spatial clustering of applications with noise (DBSCAN), the recall has improved significantly. This method was aimed at the whole fruit tree, so it has advantages in the completeness of information. The results indicated that the proposed method was effective and feasible for organs classification and yield estimation on pomegranate trees in the natural environment.

中文翻译：

---

一种基于多特征融合和3D点云支持向量机的石榴树器官分类及果实计数方法

摘要 石榴树的器官分类和果实计数对于园艺工作和机器人采摘具有重要意义。但是，仍然存在一些挑战：（1）光照在自然环境中是不可控的；(2) 传统的基于二维图像的分类识别方法受到石榴树遮挡的限制。本文提出了一种基于多特征融合和支持向量机(SVM)的石榴树器官分类和果实计数方法。其构建步骤如下：（1）通过RGB-D相机获取石榴树的三维点云；(2) 对三维点云进行了预处理；(3)提取颜色和形状特征训练SVM分类器；(4)将得到的分类器模型用于石榴树的器官分类；(5)采用基于加权欧氏距离的K-近邻(KNN)平滑，提高分类精度；(6)采用凝聚-分裂层次聚类法对石榴果实进行计数。实验结果表明，基于颜色和形状特征的SVM分类器对水果的准确率为0.75，对非水果的准确率为0.99。基于凝聚-分裂层次聚类的水果计数具有 87.74% 的召回率和 78.15% 的准确率。与基于密度的带有噪声的应用程序空间聚类（DBSCAN）相比，召回率有了显着提高。该方法针对的是整棵果树，在信息完备性上具有优势。