基于超分辨率和YOLOv8的小物体多尺度检测方法
By using Real-ESRGAN super-resolution processing and an improved YOLOv8 framework, combined with SPD-Conv multi-scale fusion convolution and PPA attention mechanism, the problem of small target detection in low-resolution images is solved, improving detection accuracy and performance.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HEBEI UNIV OF SCI & TECH
- Filing Date
- 2026-04-30
- Publication Date
- 2026-07-17
AI Technical Summary
In complex backgrounds and low-resolution images, existing technologies struggle to effectively detect small targets, especially smaller targets such as ships. This is due to difficulties in distinguishing targets from the background, the small size of the targets, and their susceptibility to noise.
Real-ESRGAN is used for super-resolution preprocessing, the YOLOv8 framework is improved, an SPD-Conv multi-scale fusion convolution module and PPA attention mechanism are introduced, and a small target-specific detection layer is added to enhance feature extraction and detection capabilities.
It significantly improves the detection accuracy of small targets in low-resolution images, better copes with background interference and scale changes, and improves detection performance.
Smart Images
Figure CN122415993A_ABST