基于三维点云的苹果树冠光照强度分布模型研究

Canopy light distribution model research of apple trees based on the 3D point cloud

关键词:  苹果树  三维点云  相对光照强度  有效叶面积指数  线性回归

Key words:  Apple tree  3D point cloud  relative light intensity  Effective leaf area index(ELAI)  linear regression.

摘要:

果树冠层形状和器官分布直接影响果实产量和品质。为了给果园精细管理中果树修枝整形、果实品质评价以及果实产量估算等提供科学的理论依据和技术指导，本文以果园自然开心型苹果树为研究对象，基于果树三维点云结构，进行探究果树冠层空间光照分布建模研究情况。首先，设计并进行实验，使用使用Windows Kinect相机采集果树照片，测量果树冠层样品点的光照强度。然后，获取果树多角度RGB如图像和深度图像并建立重构果树三维点云结构，并对果树冠层进行立体空间网格划分，利用点云分割技术对该果树三维点云进行分割得到对应高度的点云分层及分割交叉点 。，获得冠层各高度层的平均相对光照强度并分析获得了果树冠层光照三维空间分布。同时并，分别使用像素占比和Graham扫描算法计算各高度点云分层垂直投影面的占地面积和的有效投影面积和占地面积，并将利用二者比值作为估算各点云分层的有效叶面积指数。最后，以果树冠层不同高度层的有效叶面积指数为自变量，对不同高度层平均相对光照强度进行线性回归，获得果树冠层光照分布模型，并对模型进行了验证。试验结果表明：所建果树冠层光照分布模型的校正决定系数R_c^2达92.4%，校正均方根误差RMSEC为0.05，验证决定系数R_v^2达95.5%，验证均方根误差RMSEP为0.04，验证相对分析误差RPD为4.91，具有较高度预测精度和较强的预测能力强的预测能力，可为快速获取果树冠层光照分布提供可靠方法。

Abstract:

Canopy of an apple tree plays an important role on apples’ production and quality. In order to provide scientific theoretical foundation and technical guidance for pruning, fruit quality evaluating and fruit production predicting, this paper took the open center shaped apple trees as the research object, and canopy light distribution modelling was explored based on apple tree’s 3D point cloud structure. Firstly, the experiments were designed and carried out: (1) the apple trees’ images were taken by using Kinect camera; (2) the illuminations of the sampling points in apple trees’ canopies were measured. Then the apple trees’ 3D point clouds were reconstructed and the point cloud segmentation technology was used to get each apple tree’s point cloud at different altitude layers. And then the Pixel proportion and Graham scanning algorithms were used to calculate effective projection area of each layer and its covering area respectively, and their ratio was calculated as effective leaf area index of each layer. Finally, the apple tree canopy light distribution model was investigated between the actual illumination of each layer and its effective leaf area index. The model calibration and validation results showed that its calibration reached to 0.924, its root mean square error correction (RMSEC) was 0.05; its validation reached to 0.955, its root mean square error of prediction (RMSEP) was 0.04, and its relative percent deviation (RPD) was 4.91, which illustrated the model had high accuracy and performance, and it could be used to understand the apple tree canopy light distribution at different altitude layers.