Leaf rolling detection in maize under complex environments using an improved deep learning method.

Area of Science:

• Agricultural Science
• Plant Biology
• Computer Vision

Background:

• Leaf rolling is a key plant adaptation to environmental stress.
• Understanding leaf rolling mechanisms can enhance crop stress tolerance, particularly in maize.
• Accurate detection of leaf rolling is crucial but challenging with manual methods.

Purpose of the Study:

• To develop a high-throughput method for detecting leaf rolling in maize.
• To improve the understanding of leaf rolling phenotypes in response to stress.
• To enhance stress tolerance in crops through advanced phenotyping.

Main Methods:

• Utilized the YOLOv8 model for leaf rolling detection in maize.
• Integrated a Convolutional Block Attention Module for enhanced feature extraction.
• Incorporated Deformable ConvNets v2 for improved adaptability to shape and scale variations.

Main Results:

• Achieved a mean average precision of 81.6% on a complex dataset.
• Demonstrated superior performance compared to existing state-of-the-art methods.
• LRD-YOLO model is computationally efficient with 8.0 G floating point operations and 3.48 M parameters.

Conclusions:

• LRD-YOLO offers an innovative and effective solution for maize leaf rolling detection.
• The model accurately detects leaf rolling in complex scenarios with occlusion and scale variations.
• The method supports real-time inference, facilitating rapid phenotyping in agricultural research.