Research On High-throughput Phenotyping System For Open Field Maize Crops

With the rapid rising of global population,the demand for increasing the worldwide crop production has become more and more urgent.Improving the current breeding technology will help us to solve the issue in crop production.Crop phenotypic information can reflect the features of the good traits.Therefore,researchers believe that using modern sensing technologies to obtain high-throughput crop phenotypic information can greatly enhance the development of new breeding methods.At present,high-throughput phenotyping technology has been initially applied to scenarios with controllable environment,such as greenhouse.However,the good traits selected from controllable environment can hardly survive in open field under complex whether conditions.As a result,the paper has carried out research on high-throughput phenotyping system for open field maize crops based on 3D LiDAR.The contents of the paper consist of:First,the integration of high-throughput crop phenotyping robotic system was completed.We mounted a 3D LiDAR on a small agricultural robot mobile platform,and developed data collection and analysis codes based on ROS.Second,field data collection for maize crop was conducted and point cloud from multiple sampling spots were registered and fused.Data collection was conducted with a “Go-Stop and Scan-Go” manner;point cloud registration was carried out with the help of landmarks.We extracted landmarks with improved DBSCAN algorithm.Then,we employed SAC-IA and ICP to implement initial registration and precise registration,respectively.Third,row spacing and plant height of maize crop were extracted from the fused point cloud.We used Hough transform to perform row detection,based on peak points in the point density histogram.The histogram was obtained by counting the number of scan points in the horizontal segmentation region of depth zones.Thus,we could get the row spacing information by calculating the distance between adjacent crop rows.For plant height calculation,single-row point cloud was projected onto a horizontal plane.Single crop identification could be carried out by meshing the plane and counting the point density for each grid.The plant height was calculated by using the minimum enclosing box of single crop point cloud.Fourth,the phenotypic data of manual collection was used as the basis for comparison in the paper.Our solution for high-throughput phenotyping was evaluated with three parameters: extraction accuracy for landmarks,registration performance and error analysis for row spacing and plant height.The experimental results show that the high-throughput phenotypic information acquisition and analysis system can extract two typical phenotypic data of field maize: row spacing and plant height.At the same time,it is found that the accuracy of the experimental data is proportional to the density of the point cloud to a certain extent,which can provide a scientific reference for the future extraction of phenotypic information.