Wide-field imaging and recognition through cascaded complex scattering media.

Area of Science:

• Optics and Photonics
• Medical Imaging
• Artificial Intelligence

Background:

• Wide-field imaging through complex scattering media is crucial for minimally invasive healthcare but remains challenging.
• Extracting features from chaotic speckle patterns formed by scattering media requires advanced techniques.
• Current methods often rely on computationally intensive self-attention mechanisms.

Purpose of the Study:

• To develop an efficient deep learning approach for wide-field imaging and pathological screening through cascaded complex scattering media.
• To overcome limitations of existing methods in handling chaotic speckle patterns.
• To enable deployment of deep learning models on edge computing devices for clinical applications.

Main Methods:

• A novel deep learning architecture, SMixerNet, was established.
• SMixerNet utilizes parameter-free matrix transposition and multi-layer perceptrons (MLP) for efficient feature extraction.
• The model was trained and validated on extensive datasets for imaging and pathological screening.

Main Results:

• SMixerNet achieved effective wide-field imaging and pathological screening through multimode fibers and turbid media.
• The approach demonstrated superior performance with significantly fewer learning parameters compared to self-attention methods.
• Successful imaging and recognition were achieved even with cascaded complex scattering media.

Conclusions:

• Deep learning, specifically SMixerNet, effectively facilitates imaging and recognition through complex scattering media.
• The developed method offers a computationally efficient solution for clinical healthcare and industrial monitoring in challenging environments.
• This research expands the potential of medical and industrial imaging in minimally invasive and non-destructive applications.