AIoT-Based Rail Fastener Assessment in Low-Light Environment Using Focal Modulated Unsupervised Image Enhancement and Generative Segmentation

Ensuring the integrity of railway fasteners is essential for maintaining track stability and operational safety, particularly under high-speed and dynamic loading conditions. However, reliable inspection becomes challenging in low-light environments, where conventional sensor-based systems often fail due to high costs, limited scalability, and poor visibility. To address this problem, this study introduces a novel artificial Internet of things (AIoT)-driven framework that enables real-time and low-cost fastener inspection using mobile imaging devices. The proposed method integrates zero-reference deep curve estimation enhanced with focal modulation and squeeze-excitation attention mechanisms to effectively enhance image visibility in uncertain illumination. The enhanced images are then utilized to train a generative adversarial network (GAN) for precise segmentation and defect detection of fasteners. Unlike traditional enhancement and segmentation pipelines, this work uniquely fuses illumination correction, attention-guided modulation, and generative learning into a unified end-to-end AIoT framework. This integration not only addresses the persistent problem of low-light degradation but also introduces a scalable and adaptive solution suitable for on-site railway monitoring. Experimental evaluation demonstrates that the proposed enhancement network achieves superior visual restoration with a minimized total loss of 1.1823 and optimal curve configuration, while the GAN attains high segmentation precision and robustness when trained on augmented datasets, confirming its effectiveness under real-world lighting variations. The validated framework demonstrates strong potential for scalable, cost-efficient, and accurate fastener monitoring, contributing to intelligent and predictive maintenance in modern railway infrastructure.