A lightweight endoscope image restoration method based on teacher-student network knowledge distillation
By constructing a three-stage teacher network and a lightweight student network, and combining a window Transformer and U-Net hybrid model with a supervised attention module, the problems of large number of parameters and high computational complexity in endoscopic image restoration models are solved, and efficient processing of endoscopic image restoration for various degradation types is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SUZHOU ENTROPTONG INTELLIGENT TECHNOLOGY CO LTD
- Filing Date
- 2026-04-03
- Publication Date
- 2026-07-17
AI Technical Summary
Existing endoscopic image restoration models have a large number of parameters and high computational complexity, making it difficult to meet real-time processing requirements and to handle multiple degradation types simultaneously.
We employ a knowledge distillation approach based on teacher-student networks to construct a three-stage teacher network and a lightweight student network. By combining a window Transformer and U-Net hybrid model, we design a supervised attention module and a multi-stage knowledge distillation strategy to achieve efficient image restoration through knowledge transfer.
It reduces the number of network parameters for students by approximately 66%, and the processing time for a single image is reduced from 1.267 seconds to 1.004 seconds. It can simultaneously handle various degradation types such as motion blur, Gaussian blur, low-light noise, and salt-and-pepper noise, and has strong adaptability and robustness.
Smart Images
Figure CN122415360A_ABST