Corn Registration And Ear Recognition Based On Laser Point Cloud In Farmland

As an important food crop and cash crop in the world,corn is widely used and has high research value in various fields.Crop yield is the most important issue in crop research,and it is also the most concerned issue of researchers.The estimation of corn yield is mainly based on the actual number of plants and ears,but the traditional statistics of corn plants and ears is still in the manual stage,with high time cost and labor cost,and has certain damage to crops,Gradually unable to meet the current development.With the advent of the information age,digital agriculture,intelligent agriculture and other concepts have gradually become the mainstream research direction.Laser point cloud technology has become a key research direction in the field of agriculture due to its high-precision 3D scene reproduction ability,non-destructive and real-time.Based on the non rigid characteristics of corn itself,the accuracy of 3D point cloud data obtained by laser scanning in farmland environment will be affected by the wind environment factors,resulting in different degrees of stratification phenomenon,so it is necessary to analyze the accuracy of data collected in different wind environment.How to correct the precision of the scanned point cloud data of maize population and reduce the stratification is the premise of plant number identification and ear recognition.Based on the laser measurement technology,this paper studies the point cloud data collection,data accuracy analysis,non rigid registration,plant number extraction and ear recognition of Corn Population in farmland environment,The main research results are as follows:(1)Corn population point cloud data acquisition and registration:Aiming at the growth status and characteristics of corn population,a four station scanning method based on corn height is designed to achieve the acquisition of corn population point cloud data from multiple perspectives,and a rigid registration method based on automatic extraction of target ball is proposed,and the complete data of corn field point cloud is obtained.(2)Research on stratification phenomenon of corn point cloud and non rigid registration: Aiming at the influence of wind speed on the yield of corn point cloud In this paper,based on the stratification characteristics of corn leaf and corn stalk,the stratification phenomenon of point cloud was studied.The results showed that the stratification degree of maize leaves was positively correlated with wind speed,and the stratification area gradually diffused downward with the increase of wind speed.Stratification began to appear when the wind speed was 1.2m/s,and the overall stratification occurred when the wind speed was 3.1m/s.When the wind speed is below1.2m/s,the stem does not shift obviously,and when the wind speed is 3.1m/s,the stem shakes violently,so the data has no practical value.In view of the above layered phenomenon,this paper proposes a non rigid registration method of Laplace grid deformation,which realizes the deformation registration of multi site scanning point cloud data.The results show that the average hierarchical distance and distance variance are significantly reduced after non rigid registration under different wind speeds.The algorithm solves the problem of data registration in windy environment and provides data basis for crop segmentation and recognition.(3)Plant number identification and ear identification of Maize Population: This paper proposes a method to identify the number of corn plants in farmland.The number of corn plants is counted by extracting corn stalks.The results showed that the recognition rate of corn planting number reached 86.1%-92.1%,which provided data basis for crop yield estimation.In view of the serious mutual occlusion of corn in group state and irregular geometric and color characteristics of corn ear,it is impossible to recognize the ear directly.This paper proposes a method of corn ear recognition based on SVM.This method uses the distance segmentation algorithm based on curvature to segment the ear of corn population data,and randomly selects the training set and experimental set with the ratio of 1:1 for SVM training and classification to complete the ear recognition.The results showed that the recognition rate of ear was 88.3%-98.9%,which provided a theoretical basis for ear counting and ear refinement research.