基于低空遥感的果树冠层信息提取方法研究

Research on the extraction method of fruit tree canopy information based on low-altitude remote sensing

【目的】果树冠层信息是反映果树生长状况的重要参数,准确提取果树的冠层信息对于果园的精细管理十分必要。【方法】文章以苹果树和桃树为研究对象,利用无人机遥感获取果园影像数据,首先通过Mask R-CNN实例分割算法对果园果树冠层进行逐一分割,同时提取果树冠幅和冠层面积信息。然后利用果园正射影像结合QGIS软件,对果树冠层位置信息进行提取和可视化并通过目视解译对果树冠层参数信息提取结果进行评估。【结果】当交并比为0.5时,模型检测分割结果最优,测试集语义分割精确度为66.3%,目标检测精确度达到63.9%。总体冠层面积实测值与模型预测面积之间的平均相对误差为12.44%,均方根误差为0.5 m2。冠幅实测值与模型预测的面积之间的平均相对误差为16.39%,均方根误差为0.39 m,在一定范围内验证了模型对冠层面积和冠幅信息提取的可靠性。【结论】结合无人机遥感数据和Mask R-CNN实例分割算法可有效地将果树冠层分割出来,并且能够较为准确地提取果树冠层的相关参数信息,可为果园的精细管理提供一定的技术支撑。

[Purpose]Canopy information of fruit trees is an important parameter to reflect the growth condition of fruit trees,and accurate extraction of canopy information of fruit trees is necessary for fine management of orchards.[Method ]In this study,apple trees and peach trees were used as the research objects,and orchard image data were obtained by UAV remote sensing.Firstly,the canopy of fruit trees in orchards were segmented one by one by Mask RCNN instance segmentation algorithm,and the canopy width and canopy area information were extracted at the same time.Then the orthophoto of orchard combined with QGIS software was used to extract and visualize the fruit tree canopy location information and evaluate the results of fruit tree canopy parameter information extraction by visual interpretation.[Result]The results show that when the intersection ratio is 0.5,the model detects the best segmentation results,the test set semantic segmentation accuracy is 66.3%,and the target detection accuracy reaches63.9%.The average relative error between the overall canopy area measured and the area predicted by the model was 12.44%,and the root mean square error was 0.5.The average relative error between the canopy width measured and the area predicted by the model was 16.39%,and the root mean square error was 0.39,which verified the reliability of the model for canopy area and canopy width information extraction within a certain range.[Conclusion]The combination of UAV remote sensing data and Mask R-CNN instance segmentation algorithm can effectively segment the canopy of fruit trees,and can extract the information of relevant parameters of the canopy of fruit trees more accurately,which can provide some technical support for the fine management of orchards.