| This project’s research focuses on crop row detection and navigation baseline extraction in corn fields based on Li DAR.A method for crop row detection and the extraction of the navigation baseline in the early,middle,and late stages was proposed to realize the row-torow auxiliary driving of agricultural machinery in the corn field,with the goals of addressing the issues of rural labor shortage,high labor intensity,and low efficiency.The following are the primary research findings:(1)A set of corn point cloud collection system was designed,which was mainly composed of a telescopic rod,a support rod,angle adjustment equipment,Li DAR,and a data collection platform.While the telescopic rod regulates the height of the Li DAR,the angle adjustment equipment is used to alter the angle of the Li DAR tilt,and the model of the Li DAR is the Livox Horizon.A database of corn point clouds was constructed by collecting corn point clouds in the early,middle,and late stages using the point cloud collection technique.The total collecting time for the early and middle stages of the corn point cloud database was 103 minutes,while the total collecting time for the middle and late stages of the corn point cloud database was 114 minutes.(2)An algorithm for crop row detection and navigation baseline extraction based on dynamic clustered regions for pre-crop maize is proposed.A straight pass filter is used to filter the maize point cloud for ground and weeds and to extract the detection area,and a statistical filter is used to denoise the point cloud.In order to accurately classify the clustering area,the crop row spacing is obtained based on the K-Means algorithm,after which the initial horizontal banding and clustering of feature points are determined based on the row spacing;the remaining horizontal bands are divided into clustering areas and feature points are extracted based on the information from the previous horizontal band.Crop row detection is achieved using the least squares method,and the navigation baseline is determined based on the crop row detection results.(3)A PV-RCNN++ neural network based algorithm is proposed for crop row detection and navigation baseline extraction for mid to late season maize.By dividing the collected point clouds into multiple crop row detection regions,which in turn can effectively remove more background points,a data set is formed,sent to PV-RCNN++ for training,and the optimal model is selected for prediction based on the training results,multiple crop row detection regions can be obtained.In this thesis,each crop row detection area is processed separately,firstly the point cloud in the crop row detection area is pre-processed to obtain clearer crop rows,then the feature points are clustered,then the least squares method is used to achieve crop row detection,and finally the navigation baseline is determined based on the position of the top and end of the crop row detection line.(4)The proposed algorithm was tested to verify the feasibility.The test was conducted at the Anhui Agricultural University’s Wanbei Experimental Station using a Zoomlion 4YZL-5BZH maize seed harvester as a real vehicle test platform.The test results show that the proposed algorithm can effectively detect crop rows and extract navigation datum lines,and the maize seed harvester can complete straight line driving during maize harvesting without intervention,enabling the agricultural machine to carry out operations safely and effectively... |
PDF Full Text Request