State Space Models Meet Remote Sensing: A Survey

A new survey paper provides a comprehensive review of State Space Models (SSMs) in remote sensing, analyzing their rapid adoption and architectural evolution since their introduction to the field [1]. The paper, posted to arXiv on June 24, 2026, synthesizes the progress of SSM-based approaches, which offer linear computational complexity and strong capabilities in capturing long-range dependencies [1]. The authors note that SSMs have gained traction in remote sensing because they address specific challenges, including dense visual predictions, multi-modal data, and temporal data analysis [1]. These demands have driven significant advancements in customized architectures [1]. The review covers most relevant studies since SSMs first appeared in the remote sensing literature [2]. The researchers conducted a structured literature search for over 300 papers between March 2024 and December 2025, querying databases such as Google Scholar, arXiv, and IEEE Xplore using keyword combinations that included “state space model” and “Mamba” alongside “remote sensing” [3]. Papers were included only if they targeted remote sensing data or tasks and used an SSM or Mamba-derived block as a primary modeling component, not a minor auxiliary module [3]. For each included work, the team extracted task type, data modality, scanning strategy, architectural design choices, and key reported metrics [4]. The survey arrives as deep learning architectures in remote sensing face well-documented trade-offs. Convolutional Neural Networks suffer from limited receptive fields, while Vision Transformers grapple with quadratic computational complexity that hinders scalability for high-resolution data [5]. SSMs, particularly the Mamba architecture, have been positioned as an alternative that combines linear computational scaling with global context modeling [5]. A separate systematic review of approximately 120 Mamba-based remote sensing studies has catalogued innovations across micro-architectural advancements—such as adaptive scan strategies and hybrid SSM formulations—and macro-architectural integrations, including CNN-Transformer-Mamba hybrids and frequency-domain adaptations [5]. The new survey aims to serve as a foundational resource by offering a multi-dimensional analysis of SSM applications across remote sensing tasks and discussing architecture design advancements [1]. The authors also identify key challenges and future opportunities, outlining potential research directions for how SSMs can continue to evolve to meet the demands of remote sensing applications [3]. An open-source repository tracking related works is maintained at GitHub [2].