Research on the quantification and automatic classification method of Chinese cabbage plant type based on point cloud data and PointNet++

Abstract

The accurate quantification of plant types can provide a scientific basis for crop variety improvement, whereas efficient automatic classification methods greatly enhance crop management and breeding efficiency. For leafy crops such as Chinese cabbage, differences in the plant type directly affect their growth and yield. However, in current agricultural production, the classification of Chinese cabbage plant types largely depends on manual observation and lacks scientific and unified standards. Therefore, it is crucial to develop a method that can quickly and accurately quantify and classify plant types. This study has proposed a method for the rapid and accurate quantification and classification of Chinese cabbage plant types based on point-cloud data processing and the deep learning algorithm PointNet++. First, we quantified the traits related to plant type based on the growth characteristics of Chinese cabbage. K-medoids clustering analysis was then used for the unsupervised classification of the data, and specific quantification of Chinese cabbage plant types was performed based on the classification results. Finally, we combined 1024 feature vectors with 10 custom dimensionless features and used the optimized PointNet++ model for supervised learning to achieve the automatic classification of Chinese cabbage plant types. The experimental results showed that this method had an accuracy of up to 92.4% in classifying the Chinese cabbage plant types, with an average recall of 92.5% and an average F1 score of 92.3%.

Plant phenotyping relevance

点群データからハクサイの草型形質を定量化し、PointNet++で自動分類する手法の開発・評価が研究の中心であるため。

Code and data availability

The supplied blocks describe Chinese cabbage point-cloud data collection, phenotypic parameter extraction, K-medoids clustering, and PointNet++ classification, but contain no data availability statement, repository deposit, or public URL for the point-cloud dataset, images, or analysis code. No paper-specific public or

No evidence-backed public reproduction asset is currently recorded.