Correlation reconstruction mechanism based on dual wavelength imaging and neural network

Area of Science:

• Optics and Photonics
• Computer Vision
• Image Reconstruction

Background:

• Natural light-field imaging often uses mixed wavelengths, posing challenges for human vision requirements.
• Existing reconstruction methods may not fully capture object details or ensure accuracy.

Purpose of the Study:

• To propose a correlation reconstruction method based on dual-wavelength imaging and a neural network.
• To enhance the quality, detail, and reliability of reconstructed object images.

Main Methods:

• Developed a compression sensing correlation fusion-reconstruction by dual-wavelength imaging and auto-encoder neural network (CSCFR-DWI-AENN).
• Utilized dual-wavelength illumination via an optical multiplexer unit (OMU).
• Employed compressed sensing ghost imaging with illumination field distribution as prior information, combined with an autoencoder neural network for noise reduction and NSML algorithm for fusion.

Main Results:

• The proposed CSCFR-DWI-AENN method demonstrated superior comprehensive object image reconstruction.
• Achieved reproduction of complete detail information, leading to more accurate and reliable object descriptions.
• Comparative analysis using non-reference image quality metrics (EN, MI, EAV, SF) validated the algorithm's effectiveness over single-wavelength and direct dual-wavelength methods.

Conclusions:

• The CSCFR-DWI-AENN algorithm provides a significant advancement in multi-wavelength imaging technology.
• Offers a theoretical basis for future developments in imaging and demonstrates strong application prospects.
• The method effectively reconstructs object images with high fidelity and detail.