Related Experiment Video
Updated: Jun 6, 2026
13:44
Detection of Architectural Distortion in Prior Mammograms via Analysis of Oriented Patterns
Published on: August 30, 2013
42.8K
Fine-YOLO: A Simplified X-ray Prohibited Object Detection Network Based on Feature Aggregation and Normalized
Yu-Tong Zhou1, Kai-Yang Cao1, De Li1
1Department of Computer Science and Technology, Yanbian University, Yanji 133002, China.
Sensors (Basel, Switzerland)
|June 19, 2024
Summary
This study introduces Fine-YOLO, a lightweight object detection model for X-ray security imaging. It achieves high accuracy with fewer parameters, improving detection of small objects in complex backgrounds.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Image Processing
Background:
- X-ray images present challenges for object detection due to complex backgrounds and small objects.
- Existing methods often require high computational costs, hindering the development of lightweight models for security applications.
Purpose of the Study:
- To propose Fine-YOLO, a novel lightweight object detection model for rapid and accurate detection in security tasks.
- To enhance the feature learning capabilities and address information loss in existing models.
Main Methods:
- Designed a low-parameter feature aggregation (LPFA) structure for the YOLOv7 backbone.
- Introduced a high-density feature aggregation (HDFA) structure to preserve local details and deep location information.
- Utilized Normalized Wasserstein Distance (NWD) to improve bounding box regression for small objects.
Main Results:
- Fine-YOLO achieved 58.3% detection accuracy with only 16.1 million parameters on the EDS dataset.
- An auxiliary validation on the NEU-DET dataset yielded a detection accuracy of 73.1%.
Conclusions:
- Fine-YOLO offers a computationally efficient and accurate solution for object detection in security domains.
- The model's effectiveness suggests potential applications in broader inspection areas beyond security.
