Anti-noise diffractive neural network for constructing an intelligent imaging detector array.

Area of Science:

• Optics and Photonics
• Artificial Intelligence
• Machine Learning

Background:

• Intelligent imaging detector arrays require robust computational capabilities.
• Existing optical networks often lack resilience to noise and disturbances.
• Developing noise-resilient object classification methods is crucial for advanced imaging.

Purpose of the Study:

• To propose a diffractive neural network (DNN) with enhanced robustness for intelligent imaging detector arrays.
• To investigate the effectiveness of Weight-Noise-Injection (WNI) training for improving DNN noise resistance.
• To achieve accurate and fast object classification using a robust optical network.

Main Methods:

• Development of a diffractive neural network architecture.
• Implementation of Weight-Noise-Injection (WNI) training mode for the DNN.
• Layered diffractive transformation for image processing.
• Comparative analysis of model accuracy under various noise conditions.

Main Results:

• The proposed DNN trained with WNI demonstrates strong robustness against disturbances.
• The optical network effectively learns the mapping between input images and labels.
• Significantly improved noise resistance was observed in the WNI-trained DNN.
• The model exhibits higher accuracy compared to baseline methods under different noise levels.

Conclusions:

• The Weight-Noise-Injection training method is effective in developing robust diffractive neural networks.
• The proposed DNN offers a promising solution for noise-resilient object classification in intelligent imaging.
• This approach advances the development of reliable optical computing systems for imaging applications.