Enhancing signal extraction and image reconstruction through scattering media using semi-supervised learning methods

Area of Science:

• Optics and Photonics
• Machine Learning
• Image Processing

Background:

• Laser imaging systems struggle with scattering media, hindering object detection.
• Supervised learning methods require extensive manual data labeling, which is impractical.

Purpose of the Study:

• To develop a semi-supervised learning approach for signal extraction and image reconstruction in scattering environments.
• To reduce the reliance on manually labeled datasets for laser imaging systems.

Main Methods:

• A novel semi-supervised learning method utilizing the discrepancy between target signals and backscattered noise.
• Exploiting unlabeled data to improve model generalization capabilities.

Main Results:

• Superior signal extraction and image reconstruction in strong dynamic scattering environments.
• Demonstrated improvement in model generalization through the use of unlabeled data.

Conclusions:

• The proposed semi-supervised method significantly reduces the need for labeled data in laser imaging.
• This approach enhances practical applicability of laser imaging systems in challenging conditions.