High-Throughput Field Plant Phenotyping: A Self-Supervised Sequential CNN Method to Segment Overlapping Plants

Area of Science:

• Agricultural Science
• Computer Vision
• Plant Biology

Background:

• High-throughput plant phenotyping relies on accurate plant segmentation.
• Manual data labeling for training segmentation models is time-consuming and labor-intensive.
• Overlapping plants in field conditions pose a significant challenge for segmentation.

Purpose of the Study:

• To develop an efficient, self-supervised pipeline for plant image processing in high-throughput phenotyping.
• To overcome the limitations of manual data labeling for plant segmentation.
• To enable accurate separation and growth analysis of overlapping plants in field environments.

Main Methods:

• A self-supervised sequential convolutional neural network (CNN) pipeline was developed.
• Greenhouse images were used to initially segment non-overlapping plants.
• Segmentation results from early-stage images served as training data for later stages.
• Functional principal components analysis (FPCA) was integrated for genotype-growth relationship analysis.

Main Results:

• The pipeline accurately segmented foreground plants even when overlapping with background plants.
• Plant height estimation was accurate using the proposed segmentation method.
• The system demonstrated efficiency and eliminated the need for human-labeled data.
• The approach successfully assessed the impact of treatments and genotypes on plant growth.

Conclusions:

• The proposed self-supervised pipeline offers an efficient solution for plant segmentation in high-throughput phenotyping.
• This method reduces the labor associated with training data preparation.
• The pipeline facilitates accurate computer vision-based assessment of plant growth dynamics and genotype-phenotype relationships in field settings.