Graph-based methods for analyzing orchard tree structure using noisy point cloud data

Fredrik Westling,Dr James Underwood,Dr Mitch Bryson

DOI: https://doi.org/10.48550/arXiv.2009.13727

2021-02-02

Abstract:Digitisation of fruit trees using LiDAR enables analysis which can be used to better growing practices to improve yield. Sophisticated analysis requires geometric and semantic understanding of the data, including the ability to discern individual trees as well as identifying leafy and structural matter. Extraction of this information should be rapid, as should data capture, so that entire orchards can be processed, but existing methods for classification and segmentation rely on high-quality data or additional data sources like cameras. We present a method for analysis of LiDAR data specifically for individual tree location, segmentation and matter classification, which can operate on low-quality data captured by handheld or mobile LiDAR. Our methods for tree location and segmentation improved on existing methods with an F1 score of 0.774 and a v-measure of 0.915 respectively, while trunk matter classification performed poorly in absolute terms with an average F1 score of 0.490 on real data, though consistently outperformed existing methods and displayed a significantly shorter runtime.

Computer Vision and Pattern Recognition,Robotics

What problem does this paper attempt to address?

The problems that this paper attempts to solve mainly focus on three key operations in using low - quality point - cloud data for orchard tree structure analysis:

1. **Trunk Location**: How to automatically identify the trunk position of each tree in low - quality point - cloud data. This is especially important in large - scale orchard automation measurement, because manually marking the trunk position is time - consuming and impractical.

2. **Individual Tree Segmentation**: How to distinguish each tree in the point - cloud data, that is, to determine which points belong to which tree. This is very useful for understanding the growth parameters of a specific tree, especially when there is an overlap between trees.

3. **Material Classification**: How to perform semantic annotation on each point in the point - cloud data, specifically to distinguish between leaves and woody parts (such as trunks and branches). This is because the woody parts do not participate in photosynthesis, and measuring the amount of photosynthetically active material in trees is of great significance for plantation management.

The method proposed in the paper aims to process low - quality point - cloud data collected by handheld or mobile LiDAR and perform the above three operations on these data, thereby providing support for orchard management and crop yield improvement. Through the graph - search algorithm, the paper demonstrates the performance of its method in trunk location, individual tree segmentation, and material classification tasks, and compares it with existing methods, showing its advantages in low - quality data processing.