Deep Transfer Learning-Based Multi-Object Detection for Plant Stomata Phenotypic Traits Intelligent Recognition.

Area of Science:

• Plant Science
• Computational Biology
• Agricultural Technology

Background:

• Accurate analysis of plant stomata is vital for understanding crop stress tolerance.
• Manual methods for stomatal trait analysis are insufficient for high-throughput screening.
• Developing automated systems for stomata recognition is a key research challenge.

Purpose of the Study:

• To develop an automated, end-to-end framework for intelligent detection and recognition of plant stomata.
• To address the limitations of manual stomata analysis in high-throughput plant phenotyping.
• To improve the accuracy and efficiency of studying stomatal physiological characteristics.

Main Methods:

• Utilized a multi-object detection approach, framing stomata recognition as such.
• Developed an end-to-end framework integrating feature weights transfer learning with the YOLOv4 network.
• Applied the method to analyze high-throughput plant epidermal cell images.

Main Results:

• The proposed method precisely locates and identifies multiple stomata in complex microscope images.
• It automatically provides phenotypic traits of stomata, including size and count.
• Demonstrated superior accuracy, lower training cost, and strong generalization ability compared to existing methods on maize data.

Conclusions:

• The developed framework offers an efficient and accurate solution for plant stomata detection and recognition.
• It significantly facilitates high-throughput analysis of stomatal phenotypic traits for crop improvement.
• The system is user-adjustable, enhancing accuracy and scalability for diverse applications.