Study On Seed Potato Bud Eye Recognition Method Based On Laser 3D Reconstruction

Potato is the fourth largest food crop in China after wheat,corn and rice,and its planting area ranks the first in the world.In recent years,as the country attaches more importance to agriculture,agricultural mechanization and intelligent level are getting higher and higher,potato industry has realized the automation of a series of links from sowing,harvesting,classification and screening,which has greatly improved the efficiency of potato from sowing to harvesting.Potato in the process of sowing,need to seed potato cutting treatment,potato cutting can break the dormant period,promote germination,reduce potato production costs and other advantages,at present seed potato cutting mainly rely on artificial,relying on artificial seed potato cutting has high cost,low efficiency,the tool is not timely disinfection and lead to cross infection and other problems.Therefore,it is necessary to use the machine way to achieve seed potato cutting.The important premise of intelligent segmentation is to realize automatic eye recognition.At present,the bud eye recognition is realized to a certain extent by the two-dimensional image method,but the two-dimensional image is greatly affected by the illumination,and the seed potato cannot be reasonably sliced because of the two-dimensional information.In this paper,on the basis of fully studying the point cloud data processing and three-dimensional point cloud feature extraction,combined with the characteristics of the collected seed potato point cloud data,A seed potato eye recognition method based on laser three-dimensional reconstruction was proposed,and verified by experimental data.The results showed that the accuracy rate of seed potato eye recognition method based on three-dimensional point cloud was 94.9%,and the recall rate was 88.3%,which basically met the requirements of automatic seed potato cutting.The main research contents of this paper are as follows:1.Acquisition of potato point cloud based on line structure light.In this paper,a variety of current point cloud acquisition methods were studied,and several commonly used point cloud acquisition principles and methods such as grating projection method,tomography method,line structured light method were compared,combined with the current situation of seed potato point cloud acquisition,the method of line structured light was selected as the method of seed potato point cloud acquisition.The experimental platform was built and the optical system was calibrated to obtain the three-dimensional surface model of seed potato.2.The present point cloud preprocessing methods are discussed.According to the noise distribution characteristics of seed potato point cloud,this paper uses through filtering and statistical filtering to remove the noise of point cloud,and adopts voxel filtering to downsample seed potato point cloud.The distribution characteristics of seed potato eye were analyzed,and the characteristics of seed potato point cloud were analyzed from the point cloud curvature,normal vector,roughness,density,covariance,etc.It was concluded that the normal vector characteristics had the greatest influence on the feature extraction process of seed potato eye point cloud.3.A method based on three-dimensional point cloud deep learning was proposed to extract the bud eye features in seed potato point cloud,make the data set of seed potato point cloud,and conduct experiments.The experimental results showed that the three-dimensional reconstruction of seed potato and image recognition method could obtain more surface information of seed potato,which was more beneficial to the subsequent automatic cutting process of seed potato,and provided a new idea for the bud eye recognition of seed potato.It has certain guiding significance.