| The acquisition of rape phenotypic traits is extremely important in the cultivation and breeding research of rape.Currently,the monitoring of rape phenotype is mainly done manually,and manual measurement often causes certain damage to crops,which is timeconsuming,laborious and subjective.In recent years,computer technology has gradually been applied to agriculture,and the rapid development of three-dimensional reconstruction technology of plants has made non-destructive monitoring of plants and the acquisition of phenotypic parameters a research hotspot.In this paper,linear laser binocular stereo vision technology was used to obtain threedimensional point clouds from rape at different growth stages,and then preprocessed the point cloud such as coloring,denoising and target point cloud extraction,and then extracted different phenotypic traits according to different calculation methods,and accurately analyzed with the artificial measured values.The specific research contents are as follows:The acquired point clouds were colored according to the elevation value,making the point cloud map more intuitive.According to the source and classification of point cloud noise,the type of noise in the acquired rape point cloud was judged,and corresponding denoising measures were carried out.For land point cloud,this experiment combined plane fitting and pass-through filtering,which had a good effect on removing land point cloud.In this paper,the outlier noise points generated during the scanning process were denoised by statistical filtering,and a relatively smooth point cloud of rape was obtained.In this paper,the threshold of the condition filtering can be set by the spatial coordinates of a single rape plant,which can better separate the point cloud of a single rape.When rape was at the seedling stage,this paper used the segmentation based on regional growth to find appropriate parameter threshold to completely segment the leaves of rape.But at the bud stage,the change of plant type,the increase of leaf area and the overlap of leaves made this method not applicable to the segmentation of rape leaves,so this paper proposed to use the LCCP clustering method to segment the leaves to divide the complete leaves.In addition,in order to obtain the accuracy of subsequent point cloud phenotypic parameter extraction,this paper only retained the leaf edge by the point cloud edge extraction method,which made the determination of key points more accurate.In this paper,the relevant phenotypic parameters were extracted from the threedimensional point cloud of rape,including plant height,leaf length,leaf width and leaf area of rapeseed at seedling stage,and leaf chord length and leaf inclination angle of rape at the bud stage.In this paper,accuracy analysis was carried out from the extracted values in the three-dimensional point cloud and the artificially measured values,and the following conclusions were reached:compared with the manually measured values,the extracted plant height of each variety of rape was equal to the measured plant height with R2 above 0.95,the RMSE was 0.3506cm;At the seedling stage,the leaf length R2of Qinyou 7,ZheYou za 108 and Huyou 039 were all above 0.95 and the RMSE were 0.1335cm,0.1311cm and 0.1388cm respectively;The leaf width R2 were all above 0.92,and the RMSE were 0.1512cm,0.1888cm and 0.1500cm respectively;R2 of leaf area were all above 0.98,and each RMSE was 0.2960cm2,0.2309cm2 and 0.2589cm2.At the bud stage,the leaf chord length of rape extracted by three-dimensional point cloud had a high correlation with the true value,the R2of the leaf chord length of the three rape varieties were all above 0.89;R2 of leaf inclination angles were all above 0.89,included 0.920,0.894 and 0.942,and each RMSE was 7.2972°,2.0246°and 2.2834°respectively. |
PDF Full Text Request