| Chinese agriculture science research pays more and more attention to the digitization,intelligence and simulation of agriculture so that digital agriculture and smart agriculture are promising.It is basic for the realization of digital agriculture and intelligent agriculture to obtain the related parameters of crop growth,which include crop row information,plant height,plant density and crop canopy amplitude.However,the acquisition of traditional crops'growth parameters is inefficient and low-precision,which cannot meet the need of modern agricultural scientific research.3D laser point cloud technology overcomes the limitation of traditional measurement of crop parameters and it is able to make non-destructive,efficient and high-precision measurement of crop growth parameters.Therefore,this technique becomes a research hotspot in the measurement of growth parameters of crop population.The point cloud data of crop population at the farmland scale from 3D laser point cloud technology contains not only the phenotypic structure and spatial information of the crop,but also other non-target noise.Therefore,it is necessary to deal with point cloud data of crop population.The critical point of crop population growth parameter measurement is to keep complete extraction and segmentation of crop 3D point cloud data under the premise of enough accuracy of point cloud data.Based on 3D laser point cloud technology,this paper studied the registration,target extraction,segmentation,crop density calculation,row detection and height measurement of crop population's(rapeseed,maize)3D point cloud data at the farmland scale.The main procedures and results are as follows:(1)The collection of point cloud data of crop population:this paper constructed a point cloud data collection system of farm crops which based on a 3D laser scanner,designed a point cloud data collection method for farm crops and realizes the point cloud data collection of rapeseed and maize crops.(2)The registration of point cloud data of crop population:as the basic procedure of point cloud data registration,a point cloud registration method based on the fitting of the sphere target was proposed.This method calculated the spherical center coordinates by fitting the sphere,took the spherical center coordinates of different stations as the key points and achieves the registration of point cloud data of crop population.According to the experimental analysis,the accuracy of the target sphere is between 96.8%and 97.47%,which greatly improves the work efficiency compared with the traditional ICP algorithm.(3)The simplification of point cloud data of crop population:the method of hierarchical simplification of crop point cloud data can not only advanced the latter experimental efficiency,but also approved lower performance of computer software and hardware.Combined with the algorithm of random sampling and curvature sampling,this paper simplified the cluster point cloud data of rapeseed and maize through the method of layered simplification.The results show that the comprehensive simplification rates of point cloud data are respectively 94.13%(rapeseed)and 82.83%(maize),which can greatly simplify the point cloud data as well as maintain the morphological characteristics of crop population.(4)The extraction of target point cloud data:this paper proposed a method for extracting crop population canopy based on HSI color space model.The method sets appropriate thresholds(rapeseed of 0.1785 and maize of 0.1604)and extracted the canopy point cloud of crop population,which greatly improved the working efficiency under the premise of enough the experimental precision,compared with the traditional extraction method.On this basis,the method also extracted the soil surface layer via conditional filtering method,which provided data support for the point cloud data segmentation of crop population.(5)The segmentation of point cloud data of crop population:a cylinder space model clustering segmentation algorithm based on Euclidean distance clustering segmentation was proposed and it broke the limitations of the Euclidean distance clustering segmentation algorithm in crop plant segmentation.The results show that the plant segmentation accuracy of rapeseed crop population reached 90.12%,and that of maize population reached 96.63%.The algorithm not only solves the problem of fast clustering and segmentation of scattered point clouds,but also establishes a theoretical basis for high-precision measurement of crop plant growth parameters.(6)The measurement of plant growth parameters of crop population:(1)The plant density calculation of the crop population based on the point cloud data was realized by extracting the number of plants in the crop population and calculating the soil area.The results show that the relative accuracy of rapeseed planting density calculation reaches91.23%,and that of maize planting density reaches 93.63%.(2)The paper took center point of crop population as the feature point of crop line detection and proposed an axial clustering line detection algorithm.It set the axial detection threshold(rapeseed of 0.12and maize of 0.35)to realize all the correct detection of two crop population lines,and the detection result equations were output.(3)Based on the segmentation results of point cloud data of crop population,this paper extracted the canopy apex data of crop plants,and uses the fitted soil plane as the root point of the plant roots to achieve the crop group plant height measurement,which is analyzed by linear regression with the data of artificial height measurement.The fitting coefficients(R~2)of the two crops are both above 0.95,and the RMSE of rapeseed and maize are respectively 0.0375m and0.0271m.The experiment verifies the accuracy of the crop population height measurement method that bases on point cloud data. |
PDF Full Text Request