A Survey of Methods and Architectures for Dynamic Path-Controllable Deep Unfolding Network to predict the K-barriers for Intrusion Detection Using a Wireless Sensor Network

Abstract

Wireless Sensor Networks (WSNs) are critical in surveillance and intrusion detection systems, where reliable detection of unauthorized movement across a protected region is essential. A key concept in this domain is k‑barrier coverage, which ensures that any intruder path intersects with at least k disjoint sensor coverages, increasing detection reliability and resilience. Designing efficient and adaptive models for dynamic k‑barrier prediction poses significant challenges due to node mobility, resource constraints, communication limitations, and real‑time requirements. Recently, Dynamic Path‑Controllable Deep Unfolding Networks (DPDUNs) have emerged as promising architectures that combine model‑based optimization with data‑driven deep learning. These hybrid systems interpret traditional iterative optimization algorithms as network layers, enabling improved adaptability in dynamic environments. When integrated with spatial‑temporal models, reinforcement learning, and attention mechanisms, DPDUNs can effectively predict k‑barriers, adapt to changing network topologies, and optimize intrusion detection paths. This survey systematically reviews 30 studies from 2020 to 2023 that investigate methods and architectures related to deep unfolding networks, hybrid deep learning, temporal models, optimization techniques, and AI‑centric intrusion detection frameworks in WSNs. Major trends, advantages, limitations, and open research challenges are discussed. The review reveals that hybrid AI approaches combining deep unfolding with spatio‑temporal learning and optimization algorithms achieve superior performance in dynamic k‑barrier prediction. Despite notable progress, challenges related to scalability, deployment complexity, resource constraints, and real‑time adaptability persist, highlighting the need for lightweight and distributed learning models tailored for large‑scale WSNs.