Tracking And Counting Of Regenerated Buds And Extraction Of Spike Phenotype Research Based On Micro-CT And 3D Structured Light In Ratooning Rice

In recent years,the total rice production of China has been increasing year by year,but with the gradual social and economic development,China’s rice production is facing problems such as shortage of rural labor resources and low economic efficiency.Ratooning rice has the advantages of one sowing and two harvests,improving replanting index,enhancing yield and economic efficiency.The yield of ratooning rice during regeneration season plays an important role in increasing total rice production and improving grain productivity.The yield during regeneration season is closely related to the development of early regenerated buds and rice spikes,so it is important for the research on the development pattern and genetic breeding of ratooning rice through accurate counting of early regeneration buds and monitoring the dynamic growth of rice spikes.In contrast,manual regenerated bud measurement and rice spike growth monitoring have the disadvantages of serious damage,poor reproducibility and low efficiency.Therefore,this paper proposes a new method based on Micro-CT and 3D structured light imaging to realize the tracking and counting of rice regenerated buds and the phenotype calculation of dynamic developmental growth process of rice spikes in regeneration season.(1)Based on Micro-CT for tracking and counting of regenerated buds,8 ratooning rice spikes with different treatments were selected as experimental materials,and the optimal scanning parameters of ratooning rice were selected through pre-experiments,5days after the first season harvest was chosen as the buding time.The voxel information of the reconstructed scanned ratooning rice was preprocessed and the dataset was produced,YOLOv5s was used as a detector for multi-target tracking,and the improved Deep SORT algorithm was used as an accurate tracking and counting method for regenerated buds.The Deep SORT algorithm was improved as follows:optimized the ID error during regenerated bud tracking;increased the number of matches for regenerated bud target tracking to improve the ID Switch problem;increase the height information in regenerating bud tracking to realize the discrimination of valid buds in regenerating buds.The tracking results show that the MOTA of the improved Deep SORT algorithm is77.61%on average,which is 1.51%better than the original algorithm;the IDSW is reduced by 94%;the improved HOTA(0)is 83.60%,which is 11.69%better than the original algorithm;the FPS of the improved Deep SORT changes from 25 frames/s to 24frames/s,shows no significant tracking speed difference.In addition,the comparative analysis of the number of regenerated buds from tracking counts and manual measurements in this paper showed that the R~2was 0.9826,the RMSE was 3.46,and the MAPE was 5.6%.Therefore,it was concluded that the Micro-CT-based and improved Deep SORT algorithm proposed in this study had high agreement with the manual statistical values for the regenerated buds tracking count results.This paper proposed an evaluation index of early regeneration power of ratooning rice,which was defined as the ratio of the effective number of buds to the number of stalks in the first season of rice 5 days after the first harvest.Based on YOLOv5s network,the 300th frame height tomogram of CT voxels was selected for the first season stalk number detection statistics.Then,by tracking the number of effective buds counted and the number of target-detected stalks,the correlation analysis between the early regeneration power and the actual yield of regeneration season of 2 varieties of ratooning rice was performed,and their R~2was 0.7947 and 0.7635,indicating that the early regeneration force and the yield of regeneration season of ratooning rice proposed in this paper had significant correlation,which provided a theoretical basis for realizing rice yield prediction.(3)Phenotype extraction of regeneration season rice spikes based on 3D structured light imaging.Three varieties of ratooning rice were selected as experimental materials,and 3D point clouds of ratooning rice were obtained by scanning and reconstructing with3D structured light imaging system during the complete growth period of regeneration season.Firstly,point cloud pre-processing such as voxel down-sampling,PCA coordinate transformation and straight-pass filter cropping were performed to reduce the redundancy of point clouds.Then the color-based regional growth segmentation was carried out.By analyzing the differences in RGB component classification of rice spikes and leaves,a suitable color threshold was selected for point cloud threshold segmentation,a single rice spike was segmented by Euclidean clustering,and the number of rice spikes per ratooning rice plant was calculated;the segmented rice spikes are converted to a grid by greedy projection triangulation,the volume and surface area of rice spikes were calculated,and finally the leaf vein extraction algorithm was used to calculate The spike length was calculated,and the four phenotypes of the three varieties were analyzed according to the above methods,and it can be concluded that the yield result of variety Sanchuanyou 6203is superior.Based on the above two 3D imaging methods,the early regenerated bud and effective bud were counted effectively,3D information rice spike phenotypic parameters was extracted and the early regeneration power of ratooning rice was analyzed for the complete growth period of regeneration season of ratooning rice,providing new methods and theoretical basis for the in-depth understanding of the growth mechanism of ratooning rice,breeding and generation of ratooning rice.