.png){kind=icon}
.png){kind=icon}
Return
Explainable optical information flow neural network
.png){kind=icon}
PRE
AI
DOI：10.1016/j.engappai.2026.115078.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Remote identification of material types is of critical importance in industrial production, non-contact inspection, and quality control. However, conventional methods relying on hardware polarizers often suffer from light intensity loss and high costs, while existing deep learning approaches lacking physical constraints face challenges regarding computational complexity and limited interpretability. To address these issues, this paper proposes an Explainable Optical Information Flow Neural Network for remote material identification. From an artificial intelligence perspective, the proposed network embeds optical physical formulas directly into the neural architecture, replacing standard opaque approximations with transparent physical laws to reduce parameter redundancy and enhance model interpretability. Regarding engineering applications, two specific models — a polarization optical network and a refractive index-guided model — are designed to retrieve complex optical properties from simple intensity images. Experimental results demonstrate that the proposed method maintains or exceeds the performance of state-of-the-art baselines, including standard convolutional and generative adversarial networks, while reducing training time by approximately 50%. This work not only provides an efficient solution for material identification but also establishes a novel paradigm for integrating optical physics with artificial intelligence in engineering sensing systems.
Keywords:
Explainable Neural Network
Optical Information Flow
Material Identification
Physical Constraints
Deep Learning
Journal
IF:
8
Papers:
5.3K
Citations:
3.5W