PCPE-YOLO with a lightweight dynamically reconfigurable backbone for small object detection.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Small object detection is a critical challenge in computer vision.
• Existing methods often struggle with accuracy, complexity, and lightweight deployment.

Purpose of the Study:

• To propose PCPE-YOLO, a novel object detection algorithm for improved small object detection.
• To address limitations in accuracy, model complexity, and deployment requirements.

Main Methods:

• Introduced a dynamically reconfigurable C2f_PIG module for parameter reduction.
• Incorporated Context Anchor Attention to enhance focus on small object contexts.
• Added a small object detection layer and an Efficient Up-Convolution Block for improved localization and feature sharpening.

Main Results:

• PCPE-YOLO outperformed baseline and state-of-the-art methods on VisDrone2019, KITTI, and NWPU VHR-10 datasets.
• Achieved significant improvements in precision (3.8%), recall (5.6%), mAP50 (6.2%), and F1 score (5%) on VisDrone2019.
• Demonstrated superior precision among all compared approaches.

Conclusions:

• PCPE-YOLO effectively combines lightweight design with high small object detection performance.
• Offers a more efficient and reliable solution for small object detection in real-world scenarios.
• The proposed modules contribute to enhanced accuracy and reduced computational overhead.