Research on cotton plant type identification method based on multidimensional vision.

Area of Science:

• Agricultural Science
• Computer Vision
• Plant Phenomics

Background:

• Current plant type judgment relies on subjective experience, hindering efficient crop breeding and precision cultivation.
• Lack of automatic analysis and identification methods restricts progress in plant genomics and cultivation knowledge modeling.

Purpose of the Study:

• To establish a rapid analysis and identification method for cotton plant types.
• To construct a digital structure model of cotton plants using multi-dimensional vision.

Main Methods:

• Constructed 3D point cloud models of 50 cotton plants using Structure From Motion and Multi View Stereo (SFM-MVS) algorithms.
• Preprocessed point cloud data and projected the 3D model into 2D for analysis.
• Utilized the fast convex hull algorithm to analyze 2D projections for plant type identification.

Main Results:

• Achieved R2 values greater than 0.90 for plant height and width extraction.
• Demonstrated rapid processing times for point cloud reading (0.402s), multi-view projection (2.275s), and convex hull construction (0.018s).
• Established classification intervals for cotton plant types: 0-0.2 for cylinder and 0.4-1.5 for tower.

Conclusions:

• The proposed cotton plant type identification method is fast and efficient.
• Provides a solid theoretical basis and technical support for cotton plant type identification.
• Facilitates advancements in crop breeding and precision cultivation.