基于SLAM与神经辐射场的柑橘幼苗三维重建方法

打开文本图片集

中图分类号：S24；S666 文献标志码：A 文章编号：1001-411X（2025）03-0429-10

3D reconstruction of citrus seedlings based on SLAM and NeRF

GUO Jun1， YANG Dacheng1，MO Zhenjie1，LAN Yubin²， ZHANG Yali1，2 （1 College ofEngineering，South China Agricultural University，Guangzhou5064，China;2National CenterforInteatioal Collaboration Research on Precision Agricultural Aviation Pesticide Spraying Technology，Guangzhou510642，China）

Abstract： 【Objective】Aiming at the problem that it is difficult to obtain the accurate 3D point cloud of citrus seedlings and their 3D phenotypic parameters to characterize the state of seedlingswith the existing 3D reconstruction techniques，this paper proposes a method based on the simultaneous localization and mapping （SLAM）and neural radiance fields （NeRF） for 3D reconstruction of citrus seedlings. 【Method】 One-year old citrus seedlings were taken as the research object.Firstly，a depth sensor was used to capture the RGB map and depth map of the citrus seedling. Secondly， SLAM was employed to obtain the poses of the depth sensor in each frame of the image.Then， NeRF was trained for citrus seedlings，and the multi-view images with attached positional pose were fed into the multilayer erceptron （MLP）. Finally，through supervised training with volume rendering，a high-precision 3D realistic point cloud model of citrus seedlings was reconstructed. 【Result】The 3D modelofcitrus seedlings reconstructed by this method was highly realistic in terms ofcolor and texture， with clear contours and distinct layers，and had real-world level accuracy. Based on this model， the 3D phenotypic parameters of citrus seedlings could be effectively extracted with the accuracy of 9 7 . 9 4 % for plant height， 9 3 . 9 5 % for breadth length， 9 4 . 1 1 % for breadth width and 9 7 . 6 2 % for stem thickness. 【Conclusion】 This study helps to accelerate the selection and nursery process of excelent citrus seedlings and provides a technical support for the sustainable development of the citrus industry.

Key Words： Citrus seedling; Plant 3D phenotype; 3D reconstruction; Neural radiance fields （NeRF）; Simultaneous localization and mapping （SLAM）; Deep learning

广东是全国传统的柑橘优势产区之一，柑橘种植历史悠久，年产量逐年上升。（剩余13301字）