Lightweight Infrared Image Denoising Method Based on Adversarial Transfer Learning.

Area of Science:

• Computer Vision
• Machine Learning
• Image Processing

Background:

• Infrared images often suffer from noise due to limited data.
• Existing denoising methods may struggle with insufficient sample variety.

Purpose of the Study:

• To propose a lightweight and efficient infrared image denoising method.
• To leverage adversarial transfer learning to overcome data limitations in infrared imaging.

Main Methods:

• Utilized a generative adversarial network (GAN) framework.
• Employed a phased transfer learning strategy: pre-training on visible light data, then fine-tuning on infrared data.
• Incorporated structural reparameterization, edge convolution, and progressive multi-scale attention block (PMAB) for enhanced feature extraction.

Main Results:

• The method effectively removes additive white Gaussian noise from infrared images.
• Demonstrated outstanding denoising performance on public and real-world datasets.
• Achieved significant reductions in model parameters and complexity for improved efficiency.

Conclusions:

• The proposed adversarial transfer learning method offers a robust solution for infrared image denoising.
• The integration of advanced network components enhances edge and texture feature recognition.
• The lightweight architecture ensures efficient denoising, suitable for practical applications.