Development of a deep-learning phenotyping tool for analyzing image-based strawberry phenotypes.

Area of Science:

• Agricultural Science
• Computer Science
• Plant Science

Background:

• Traditional strawberry phenotyping methods are labor-intensive and time-consuming.
• Accurate measurement of phenotypic traits is crucial for effective strawberry plant monitoring and management.

Purpose of the Study:

• To develop an image-based Strawberry Phenotyping Tool (SPT) using deep learning (DL) for digital strawberry plant phenotyping.
• To create a robust DL-based tool for natural scene or stored image analysis.

Main Methods:

• Developed SPT using two DL architectures (YOLOv4 and U-net).
• Created two versions (V1 and V2) differing in dataset size and annotation method (single-target vs. multitarget labeling).
• Utilized smartphone-captured images with diverse backgrounds.

Main Results:

• SPT successfully facilitated strawberry phenotype measurements.
• Detection accuracy improved from 60.24% (V1) to 82.28% (V2) with increased data and multitarget labeling.
• Developed an image-based regression model to predict fresh strawberry weight (R² = 0.92).

Conclusions:

• The developed SPT efficiently recognizes six strawberry phenotypic traits in complex environments.
• This tool supports accurate and efficient decision-making for strawberry plant management, potentially increasing productivity.
• The study highlights the effectiveness of DL in advancing digital plant phenotyping.