Recognition and Localization of Maize Leaf and Stalk Trajectories in RGB Images Based on Point-Line Net

Area of Science:

• Agricultural Science
• Computer Vision
• Plant Biology

Background:

• Plant phenotype detection is vital for agriculture and crop breeding, offering insights into growth and environmental responses.
• Traditional methods for quantifying plant traits like leaf number and growth trajectory are labor-intensive and costly.
• Deep learning in complex field environments faces challenges like occlusions and intricate backgrounds for accurate plant analysis.

Purpose of the Study:

• To explore the application of deep learning for automated plant phenotype acquisition in maize fields.
• To develop a method for determining the number and growth trajectory of leaves and stalks using field images.
• To enhance the efficiency of plant breeding through AI-driven phenotyping.

Main Methods:

• Developed Point-Line Net, a deep learning model based on the Mask R-CNN framework.
• Utilized RGB images from maize fields for automated recognition of plant structures.
• Introduced a lightweight keypoint detection branch for precise localization and growth tracking.

Main Results:

• Point-Line Net achieved an object detection accuracy (mAP50) of 81.5%.
• The keypoint detection branch demonstrated effectiveness with a custom distance verification index of 33.5.
• The method successfully determined the number and growth trajectory of maize leaves and stalks.

Conclusions:

• The Point-Line Net offers a promising AI-driven solution for automated plant phenotype detection in challenging field conditions.
• This approach significantly improves efficiency and reduces labor in acquiring crucial plant traits for crop breeding.
• Findings provide valuable insights for future research in field plant phenotyping, especially with dot and line annotations.