Dense object detection methods in RAW UAV imagery based on YOLOv8.

Area of Science:

• Agricultural technology
• Computer vision
• Machine learning

Background:

• Precision agriculture demands accurate, fast, and lightweight methods for dense target detection.
• Drones are increasingly used for crop monitoring, necessitating efficient object detection algorithms.

Purpose of the Study:

• To develop an improved dense target detection method for identifying apricot flowers using drones.
• To enhance the accuracy and efficiency of object detection in dense agricultural settings.

Main Methods:

• An improved dense target detection method based on YOLOv8, named D-YOLOv8, was proposed.
• Incorporated Dense Feature Pyramid Networks (D-FPN) for enhanced dense feature extraction.
• Introduced a Dense Attention Layer (DAL) to focus on dense target areas and suppress irrelevant features.
• Utilized RAW data augmentation to enrich feature input for dense objects.

Main Results:

• The D-YOLOv8m model achieved 98.37% Average Precision (AP) on the CARPK challenge dataset and a constructed dataset.
• The model maintained a lightweight design with only 13.2 million parameters.
• Demonstrated significant improvements in dense target detection accuracy.

Conclusions:

• The proposed D-YOLOv8 method effectively enhances dense feature extraction and data augmentation for improved detection accuracy.
• The lightweight and accurate D-YOLOv8 network can support various dense target detection tasks in precision agriculture.