A new publication from Opto-Electronic Advances; DOI 10.29026/oea.2024.240119 , discusses remodeling the "brain" of distributed sensor systems.

Sensors are like the "perceptive organs" of modern technology, designed to pick up physical signals such as temperature and vibrations. Think of a Distributed Sensing System (DSS) as a network of "skin" spread across large areas, capable of gathering detailed data from its surroundings. Just like how the skin on your fingertips is more sensitive than the skin on your back, the sensors in DSS don’t work the same everywhere. Differences in installation or environmental conditions can cause its performance to vary, leading to inconsistencies of signal feature for the same events in different regions. Traditionally, AI models are trained using labeled data from one region (called the source region) and then directly applied to other regions (called the target regions). However, these models often struggle to handle significant differences between regions, lacking the “brain-like” ability to adapt to unique regional characteristics. While retraining models with labeled data specific to each target region could help, it’s too expensive and unrealistic to do this for every possible scenario.

An alternative solution comes from unlabeled data, which is automatically collected by DSS after they’re set up. This data is low-cost and contains valuable information about regional characteristics and event responses. It has the potential to improve the adaptability of AI models. Traditional methods use model’s predicted results on unlabeled data as pseudo labels, and directly apply them to train model iteratively. However, When the differences between source and target regions are large, pseudo-label errors will pile up, making the model’s training process unreliable.

Recently, the research team of INTELLISENSE LAB ON X-fiber at Huazhong University of Science and Technology introduced an Adaptive Decentralized Artificial Intelligence (ADAI) framework designed for signal recognition tasks in DSS. Using intrusion signal recognition in distributed optical fiber sensor systems (DOFS) as a case study, the team validated ADAI's effectiveness across regions with significantly different geological conditions, such as utility tunnels, tunnels, and perimeter security zones (Figure 1). To address regional variations, the ADAI framework remodels the DSS "brain" By leveraging unlabeled data from each target region, the ADAI framework fine-tunes the model from source region (AI-S), to generate adaptive cross-region model for each region (e.g., AI-T1, AI-T2).

In the process of "brain" remodeling, the basic idea is to ensure both the retention of valuable information and real-time adaptation to environmental differences. To this end, the team use labeled data from source region for model training, to maintain classification ability of cross-region model. Furthermore, they establish data relationships between regions by measuring the feature distance between the source and target regions, and adapt the cross-region model to the target region by reducing the feature distance (Figure 2). Experimental results show that compared to baseline models, the ADAI-enhanced "brain" achieved average accuracy increases of 33.2% and 73% in two target regions, respectively. Additionally, the models exhibited exceptionally low false alarm rate (FAR) and missed alarm rate (MAR), at less than 4.3% and 2.7%, respectively, showcasing the robustness of ADAI framework.

This breakthrough reduces DSS dependence on large volumes of labeled data, illustrating a new paradigm for integrating AI with sensing technologies. Since it addresses the general challenge of cross-region model adaptation, the ADAI framework has broad potential for technological transfer, particularly in resource-limited or label-scarce scenarios. Looking ahead, the research team plans to incorporate physical simulation and digital twin technologies to establish a precise and efficient dynamic monitoring system, further advancing the intelligent application of DSS.

Keywords: artificial intelligence (AI) / signal recognition / distributed sensor systems (DSS) / distributed optical fiber sensors (DOFS)
# # # # # #
The research team of Intelligence Lab on X-fiber (iF-Lab) at Huazhong University of Science and Technology is supported by the National Engineering Research Center for Next-Generation Internet Access Systems. This team is dedicated to research on micro-nano structured optical fiber sensing technologies and their applications, including micro-nano structured specialty optical fibers and devices, distributed optical fiber sensing technology, all-fiber ultrasonic transducer and imaging technology, ultrafast laser precision measurement technology, and their applications in large-scale engineering and health monitoring (Research Team Website: http://mnofs.oei.hust.edu.cn/index.htm). Professor Sun Qizhen, the team leader, has won the National Science Fund for Distinguished Young Scholars and Excellent Young Scientists Fund, and was granted by the EU Marie Curie Fellow, and Hubei Province Innovation Group. The team has undertaken over 20 national-level projects, including NSFC (Distinguished Young Scholars, Excellent Young Scientists fund, Key Project, Joint Fund for Regional Innovation and Development, General Project, and Young Scientists fund), key research and development projects and instruments projects from the Ministry of Science and Technology, etc. The team has been authored/co-authored over 120 academic papers in the leading scholarly journals in optical field, including Opto-Electronics Advances, Light: Science and Applications, PhotoniX, Optica, Photonics Research, Advanced Science, etc., and has been authorized 47 invention patents，of which the applied patent technologies on fiber-optic acoustic sensing has successfully transferred. The corresponding research achievements have been applied in many fields and recognized by several awards, including the first prize for technical invention of China Society of Communications, the first prize for innovative products of the Chinese Society of Optical Engineering and the gold medal of the Geneva International invention Exhibition, etc.
# # # # # #
Opto-Electronic Advances (OEA) is a rapidly growing high-impact, open access, peer reviewed monthly SCI journal with an impact factor of 15.3 (Journal Citation Reports for IF2023). Since its launch in March 2018, OEA has been indexed in SCI, EI, DOAJ, Scopus, CA and ICI databases over the time, and expanded its Editorial Board to 34 members from 17 countries.
# # # # # #


More information: http://www.oejournal.org/oea
Editorial Board: http://www.oejournal.org/oea/editorialboard/list
All issues available in the online archive (http://www.oejournal.org/oea/archive).
Submissions to OEA may be made using ScholarOne (https://mc03.manuscriptcentral.com/oea).
ISSN: 2096-4579
CN: 51-1781/TN
Contact Us: oea@ioe.ac.cn
Twitter: @OptoElectronAdv (https://twitter.com/OptoElectronAdv?lang=en)
WeChat: OE_Journal
# # # # # #
Zhang SX, Li H, Fan CZ et al. Adaptive decentralized AI scheme for signal recognition of distributed sensor systems. Opto-Electron Adv 7, 240119 (2024). doi: 10.29026/oea.2024.240119