An RT-DETR remote sensing image recognition method fusing CAFM and PKI modules
By introducing PKI and CAFM modules into the RT-DETR model, the problems of insufficient detection accuracy and multi-scale adaptability in remote sensing image recognition are solved, and high-precision and robust remote sensing image recognition is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUANGXI UNIV
- Filing Date
- 2026-06-12
- Publication Date
- 2026-07-17
AI Technical Summary
Existing remote sensing image recognition technologies lack the accuracy and multi-scale adaptability for target detection in complex backgrounds. Traditional methods struggle to fully represent deep semantic information, and convolutional neural network-based methods have limitations in long-distance dependency modeling and global context representation.
The RT-DETR remote sensing image recognition method, which introduces CAFM and PKI modules, enhances the local structure and contextual relationships of high-level semantic features by introducing the PKI module in Stage 3 of the backbone network for multi-receptive field feature extraction and introducing the CAFM module before Stage 4 for local and global information fusion.
It improves the detection accuracy and multi-scale adaptability of remote sensing image recognition, enhances robustness to complex backgrounds, and achieves high-precision, low-miss-detection, and low-false-detection multi-scale target recognition.
Smart Images
Figure CN122416296A_ABST