Three-dimensional Reconstruction And Leaf Phenotype Monitoring Of Cotton Plants

Cotton is a particularly important cash crop,and Xinjiang is one of the major cotton growing areas in China,and cotton has become an important local cash crop and a major source of income for farmers.in2020,cotton production in Xinjiang accounted for 86.74%of the total national production,and the total planted area and total production have ranked first in the country for 27 consecutive years.For growth monitoring and yield prediction,it is important to obtain phenotypic information of cotton plants.Traditional phenotype measurement methods mainly rely on manual measurement,but there are problems such as subjectivity,lack of objectivity and easy to cause plant damage.With the development of agricultural modernization and computer vision technology,3D reconstruction technology is widely used in the field of plant phenotype research.Compared with traditional two-dimensional images,3D models can present more comprehensive information on plant growth morphology,spatial position of leaves,plant height and leaf area,which is a positive contribution to the development of cotton breeding and planting.In this paper,we propose a method for measuring cotton phenotypes based on 3D reconstruction,which is of great practical significance.This paper focuses on the following three aspects of research and application:(1)A 3D reconstruction model of cotton based on motion recovery and stereo geometric visual fusion was constructed,which is complete and clear and basically reproduces the morphology and characteristics of the plant itself and provides 3D point cloud data support for the subsequent study.By setting up the experimental environment,the data acquisition method suitable for cotton plants was determined,and 304cotton plants and more than 50,000 cotton pictures were obtained.(2)In this paper,we propose a feasible method for phenotypic measurement of cotton,based on the reconstructed 3D point cloud of cotton to manually measure 30 leaves and plant image analyzer to obtain 50leaf areas of leaves,as well as to measure the height of 10 buds and 30 cotton plants,and perform error analysis of the results.The result analysis showed that the phenotypic measurement method based on 3D reconstruction proposed in this study expressed more complete information and improved measurement accuracy significantly.(3)This paper proposes a feasible method for phenotypic measurement of cotton,based on the reconstructed 3D point cloud of cotton for leaf number,leaf area and plant height,and the error analysis of the results.The result analysis shows that the phenotypic measurement method based on 3D reconstruction proposed in this study expresses more complete information and improves the measurement accuracy significantly.Through the above research and application,the IoU value of the PointNet++network model trained to segment leaves was obtained as 0.881,and the accuracy of the predicted value was 0.917;the DBSCAN segmentation of individual cotton leaves was more effective than the Kmeans clustering method;regression analysis was performed between the algorithm values and the manual measurements,and the values of R~2for plant height at bud and boll stages were 0.96 and 0.97,respectively The values of R~2 were 0.92 and 0.98for the regression analysis of the algorithm values with the manually measured values and the leaf area obtained from the plant image analyzer,respectively.