Three-dimensional branch segmentation and phenotype extraction of maize tassel based on deep learning

Area of Science:

• Plant Science
• Computational Biology
• Agricultural Technology

Background:

• Maize tassel morphology is crucial for plant growth, reproduction, and yield.
• Accurate phenotyping is essential for Distinctness, Uniformity, and Stability (DUS) testing.
• Point cloud deep learning offers automated maize tassel trait acquisition but requires large datasets and struggles with adherent organs.

Purpose of the Study:

• To develop an efficient method for maize tassel organ segmentation and phenotype analysis.
• To create an automated system for high-throughput phenotyping of maize tassels.
• To address limitations in existing point cloud deep learning methods for plant organ segmentation.

Main Methods:

• An innovative incomplete annotation method for point cloud data was proposed.
• The MaizeTasselSeg system was developed, utilizing PointNet++ for tip feature learning and segmentation.
• A shortest path algorithm was employed for complete branch segmentation.

Main Results:

• The MaizeTasselSeg system achieved high segmentation accuracy (IoU: 96.29%, Precision: 96.36%, Recall: 93.01%).
• Six morphological traits (branch count, length, angle, curvature, volume, dispersion) were automatically extracted.
• High correlation coefficients (R² values up to 0.9897) and low RMSE were observed for key traits.

Conclusions:

• The proposed method offers an efficient scheme for high-throughput maize tassel organ segmentation.
• The system enables automatic extraction of maize tassel phenotypic traits.
• The incomplete annotation approach presents a novel strategy for morphology-based plant segmentation.