Automatic acquisition, analysis and wilting measurement of cotton 3D phenotype based on point cloud

Abstract

This study constructed a high-throughput method for the acquisition and analysis of three-dimensional phenotypes of cotton, and proposes a method for evaluating the degree of wilting of cotton varieties based on phenotype. The upgraded version of the self-developed data acquisition platform MVS-Pheno V2 was used to continuously collect point cloud data. PointSegAt deep learning network model was used to establish plant stem and leaf segmentation and leaf overlap distinction models, realising the segmentation of cotton plant stems and leaves and the distinction of leaf overlap. In addition, an algorithm called "Active Boundary Segmentation" has been developed, which achieved automatic segmentation of overlapping cotton leaves. Based on point cloud technology, the automation of plant height, leaf count, and wilted leaf area based on voxels was realised, and a set of wilt measurement methods for cotton plants was designed. The results show that the PointSegAt model proposed has good performance in stem and leaf segmentation, with a segmentation accuracy of 0.995 and mean intersection over union of 0.924. In terms of single leaf segmentation, the average accuracy reached 0.95, and the average f1-score reached 0.94. Compared with manual measurements of plant height, leaf count, leaf area, and canopy area, the correlation coefficients were 0.99, 0.96, 0.90, and 0.99, respectively, and the root mean square errors were 0.01, 0.04, 0.19, and 0.02, respectively. Finally, the proposed method was used to perform wilting quantification experiments on two different varieties of cotton plants, and quantitative analysis of drought resistance of different varieties was conducted.

Code and data availability

公開本文の所在を確認できませんでした。非公開または購読が必要な可能性があります。

No evidence-backed public reproduction asset is currently recorded.