Research On Density Estimation Method Of Field Mature Wheat Based On 3D Point Cloud Model Segmentation

The sustainable development of agriculture depends on the implementation of precision agriculture,and the research on field crop traits is the key to precision agriculture.Wheat is an important food crop in the world,and its quality and yield are closely related to global food security.For wheat,the ear is an important factor that constitutes the yield,and its density is one of the most important crop traits.The traditional density measurement method uses manual measurement,which is time-consuming and labor-intensive.Therefore,an efficient and convenient wheat density estimation method is urgently needed to provide timely and reliable crop yield monitoring data for the agricultural production management system in the era of big data.This paper takes mature wheat in the field as the research object.Based on computer vision technology,it uses stereo vision,lidar,and multi-source heterogeneous information fusion to research the segmentation method of the 3D point cloud model,and realizes the intelligent analysis of wheat plant density.The main research contents are as follows:(1)The density estimation method of field mature wheat based on 3D point cloud clustering and segmentation of crop rows was studied.Aiming at the problems of low efficiency and high cost of traditional manual statistics,the 3D point cloud model of wheat was reconstructed by stereo vision technology,and cluster segmentation was carried out.A row cluster segmentation method for field-dense wheat was proposed.Quick shift algorithm was used to cluster wheat ears to realize automatic segmentation of sparse wheat sample canopy.On this basis,octree splitting and voxel mesh merging algorithms were used to re-cluster the dense wheat in the field.Through linear regression analysis,the relationship model between the number of wheat ear point clouds and the number of wheat plants was established.The results showed that the determination coefficient~2 of the predicted value and the manually measured value in the field was 0.93,and the root mean square errorwas 18.53 plants,indicating that the proposed method had good accuracy.(2)The density estimation method of field mature wheat based on quantile regression of three-dimensional point cloud depth was studied.Aiming at the problems such as the influence of sunlight and the complexity of the algorithm during the measurement of wheat density by a stereo camera,a wheat density estimation method based on quantile regression of point cloud depth was proposed by collecting three-dimensional point cloud data in the wheat field by laser radar.Firstly,3D point cloud data collected were preprocessed using Cloth Simulation Filter(CSF)to filter the ground-measured point cloud.For the accurate extraction of phenotypic parameters from population point clouds,a deep quantile regression algorithm was used to extract wheat canopy position information,and a relationship model between wheat biomass density and canopy point cloud quantity was constructed to estimate wheat density.The density estimation results were highly consistent with the manual measurement results,the coefficient of determination~2 was 0.97,and the root mean square errorwas 14.26 plants,which could realize the efficient estimation of wheat density and had strong practicability.(3)The density estimation method of field mature wheat based on multi-source heterogeneous information fusion was studied.Due to the problem,the binocular camera is prone to overexposure or underexposure under the conditions of sunlight and shadow,and laser radar can not obtain object color information,a method for estimating the density of mature wheat in the field based on the heterogeneous information fusion of binocular camera and laser radar was proposed by using the two-dimensional image information and three-dimensional point cloud information collected synchronously.The two-dimensional bounding box of each target is generated by the detection of two-dimensional image information,and the corresponding three-dimensional point cloud data is obtained.On this basis,DBSCAN algorithm of point cloud clustering technology was used to segment the target and realize ear instance segmentation of the three-dimensional point cloud of wheat plants,and then the wheat density was calculated.The determination coefficient~2 of the density estimation results and the manual measurement results was 0.99,and the root mean square errorwas 8.98plants.The results showed that the method using the heterogeneous sensor information from multiple sources for comprehensive analysis had more advantages than that using the single sensor,increased the diversity of information components,and effectively improved the accuracy and accuracy of wheat density estimation.In conclusion,the three-dimensional point cloud model segmentation method proposed in this paper achieves an accurate estimation of population density parameters of mature wheat in the field,provides an important idea and direction for wheat growth and yield monitoring,and helps to promote intelligent research on wheat growth and quality cultivation,which is of great significance for food production and security.At the same time,the application of these methods can also play a role in the growth monitoring of other crops and plants,which has a wide range of application value.