Point cloud organ segmentation and phenotypic information extraction of cotton plant at seedling stage based on machine learning

The physiological state and health status of each organ of cotton plant reflect the growth and development state of cotton plant, and the phenotypic analysis of each organ can effectively represent the health degree of cotton. Aiming at the problem that three-dimensional point cloud is difficult to achieve new leaf organ segmentation at the top of cotton seedling stage, this study proposed a new leaf segmentation method combining Laplacian skeleton extraction algorithm and Quickshift++ based on distance field. The method mainly includes five steps: skeleton extraction, skeleton decomposition, new leaf extraction, new leaf organ segmentation and phenotype parameter extraction. In the process of skeleton extraction, the skeleton extraction algorithm based on Laplace was used to extract the skeleton of cotton plant. In the process of skeleton decomposition, according to the morphological characteristics of cotton plant, the plant skeleton is decomposed into stem subskeleton and partial leaf subskeleton. The plant space coordinate system was established in this study. The Z-axis of the coordinate system overlaps with the stem axis, which can better represent the characteristic morphology of cotton plant. The point cloud of cotton plant is transformed from the original coordinate system to the spatial coordinate system of cotton plant, and the subsequent skeleton decomposition steps are carried out in the spatial coordinate system of cotton plant. However, for cotton plants with new leaves, especially the distance between new leaves is very small, the Laplacian skeleton extraction algorithm has limited effect on the extraction of new leaf skeleton, and it is easy to over-segment new leaves. To solve this problem, this study proposed a distance field organ segmentation method based on Quickshift++, which can accurately segment new leaf organs, so as to realize the whole organ segmentation of cotton plant at seedling stage. This method can encode the global spatial structure and local feature relationships of the new leaf, and realize the rapid localization of the new leaf and the precise segmentation of the organ. In the process of segmentation of newborn leaves, Quickshift++ algorithm with distance field as input value is used to extract the tip point clouds of different organs of newborn leaves. Secondly, the neophylla apex cloud was identified according to the local geometric features of the neophylla apex cloud. Then combined with the median normalized vector growth segmentation algorithm, the stem point cloud of new leaves is segmented. Finally, Quickshift++ algorithm was used to segment the point cloud of the newborn leaf organs. The results showed that the average accuracy of organ segmentation was 0.977, the average recall rate was 0.978, the average F1 score was 0.976, and the average overall accuracy was 0.983. The determination coefficients of stem height, leaf width, leaf length and leaf area were 0.982, 0.962, 0.971 and 0.949, respectively. The root mean square errors were 0.119 cm, 0.182 cm, 0.163 cm and 1.458 cm², respectively. The method studied in this study can accurately segment cotton plant organs at seedling stage, and provides an effective technical means for high-throughput phenotypic detection of cotton.