基于激光点云的农田玉米种植株数数目识别

为了对玉米种植株数进行无损的自动化识别，利用FARO focus s70地面激光扫描仪、采用四站式扫描方法，采集不同视角下的玉米田块点云数据。设计了一种基于标靶球自动提取的配准算法，对各站获取的点云数据进行精确配准，取得了完整的玉米田块点云数据，并以标靶球拟合误差和标准偏差分析配准精度。对于配准好的3维点云数据，利用采样一致性算法基于圆柱体特征从完整的玉米田块点云中精确分离出茎秆点云，统计玉米种植株数。结果表明，标靶球拟合标准偏差在0.1mm~0.7mm之间，满足仪器测量精度要求; 拟合误差总体在2mm~5mm之间，能满足大场景测量配准误差5mm的要求; 对玉米种植株数的识别率达到86.1%~92.1%。这一结果对于农田环境下玉米种植株数识别的实际应用方面是有帮助的，为作物的估产提供了数据基础，为智慧农业研究提供了理论方法。

Recognition of the number of corn plants in farmland based on laser point cloud

In order to identify the number of maize plants without damage,the four station scanning method was used to collect the data of corn field point cloud from different perspectives by using FARO focus s70 laser scanner.A registration algorithm based on automatic extraction of target ball was designed.The point cloud data obtained by each station was accurately registered,and the complete corn field point cloud data was obtained.The registration accuracy was analyzed by the fitting error and standard deviation of the target ball.For the three-dimensional point cloud data,the stem point cloud was separated from the whole corn field point cloud by using the sampling consistency algorithm based on the cylinder characteristics,and the number of corn planting plants was counted.The results show that the standard deviation of the standard fitting of the target ball is between 0.1mm and 0.7mm,which meets the requirements of the instrument measurement accuracy.The fitting error is between 2mm~5mm,which can meet the requirements of 5mm in large scene measurement registration error.The recognition rate of maize plant number was 86.1%~92.1%.This result is helpful to the practical application of maize plant number identification in farmland environment,providing data base for crop yield estimation and theoretical method for intelligent agricultural research.