基于SwinPoinTr的视角受限下杏鲍菇表型参数测量方法

Phenotypic Parameter Measurement Method for Pleurotus eryngii Based on SwinPoinTr under Limited View Conditions

摘要:

针对菇房内杏鲍菇表型参数测量任务中，由于扫描设备视角受限，扫描的杏鲍菇点云出现残缺问题，基于AdaPoinTr(Adaptive geometry aware point transformers)提出了改进的SwinPoinTr模型，实现了对残缺杏鲍菇点云的准确补全和杏鲍菇表型参数的测量。该方法在使用提出的特征重塑模块的基础上，构建具有几何感知能力的层次化Transformer编码模块，提高了模型对输入点云的利用率和模型捕捉点云细节特征的能力。然后基于泊松重建方法完成了补全点云表面重建，并测量到杏鲍菇表型参数。实验结果表明，本文所提算法在残缺杏鲍菇点云补全任务中，模型倒角距离为1.316×10-4，地球移动距离为21.328 2，F1分数为87.87%。在表型参数估测任务中，模型对杏鲍菇菌高、体积、表面积估测结果的决定系数分别为0.958 2、0.959 6、0.960 5，均方根误差分别为4.421 3 mm、10.818 5 cm3、7.577 8 cm2。结果证实了该研究方法可以有效地补全残缺的杏鲍菇点云，可以为菇房内杏鲍菇表型参数测量提供基础。

Abstract:

In the task of measuring the phenotypic parameters of Pleurotus eryngii in the mushroom house, due to the limited viewing angle of the scanning equipment, the point cloud of the scanned Pleurotus eryngii is incomplete. Aiming to address this problem, an improved SwinPoinTr was proposed based on adaptive geometry-aware point transformers (AdaPoinTr), achieving accurate completion of the incomplete Pleurotus eryngii point cloud and measurement of its phenotypic parameters. Based on the proposed feature reshaping module, a hierarchical Transformer encoding module was constructed with geometric perception ability, improving the model’s utilization rate of the input point cloud and its ability to capture the detailed features of the point cloud. Then based on the Poisson reconstruction method, the surface reconstruction of the completed point cloud was completed, and the phenotypic parameters of Pleurotus eryngii were measured. The experimental results showed that in the task of completing the incomplete Pleurotus eryngii point cloud, the chamfer distance of the model was 1.316×10-4, the earth mover’s distance was 21.328 2, and the F1-score was 87.87%. In the task of phenotypic parameter estimation, the determination coefficients of the model’s estimation results for the height, volume, and surface area of Pleurotus eryngii were 0.958 2, 0.959 6, and 0.960 5, respectively, and the root mean square errors were 4.421 3 mm, 10.818 5 cm3, and 7.577 8 cm2, respectively. The results confirmed that the method can effectively complete the incomplete Pleurotus eryngii point cloud and provide a basis for the measurement of the phenotypic parameters of Pleurotus eryngii in the mushroom house.