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 seedlings with 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 model of citrus seedlings reconstructed by this method was highly realistic in terms of color 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 97.94% for plant height, 93.95% for breadth length, 94.11% for breadth width and 97.62% for stem thickness.
Conclusion This study helps to accelerate the selection and nursery process of excellent citrus seedlings and provides a technical support for the sustainable development of the citrus industry.