Physics-Aware Complex-Valued State Space Model with Scattering-Prior Feature Modulation for PolSAR Image Classification

2026-07-23 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Environmental Science & Earth Systems, Engineering & Applied Sciences · Depth: Expert, extended

Summary

CV-SSMNet is a novel physics-aware complex-valued state-space network designed for Polarimetric Synthetic Aperture Radar (PolSAR) image classification. This model addresses limitations in long-range spatial dependency modeling and the insufficient integration of physical knowledge in existing complex-valued networks. CV-SSMNet employs a complex-valued state-space model (CV-SSM) to capture extensive spatial dependencies while preserving polarimetric amplitude–phase coupling in the complex domain. It incorporates seven physically meaningful scattering priors—H, A, α, Ps, Pd, Pv, and Span—as FiLM-style modulation signals to adaptively recalibrate complex-valued features. The architecture combines multi-scale complex convolutions, branch-wise CV-SSM encoding, prior-guided recalibration, and lightweight global context aggregation. Evaluated on L-band Flevoland, San Francisco, Oberpfaffenhofen, and a P-band BIOMASS dataset, CV-SSMNet achieves competitive accuracy, including 97.56% OA on Flevoland, alongside improved regional consistency and boundary preservation, with inference times under 0.6 ms per sample.

Key takeaway

For AI Scientists and Machine Learning Engineers developing PolSAR image classification systems, you should consider CV-SSMNet's approach. Its integration of complex-valued state-space models with physics-aware scattering prior modulation significantly improves classification accuracy, regional consistency, and boundary preservation. This framework offers a robust method for incorporating physical knowledge into deep learning, enhancing interpretability and performance in GeoAI applications. Implement block-based data splitting to ensure reliable cross-region evaluation.

Key insights

Integrating complex-valued state-space models with physics-aware prior modulation enhances PolSAR image classification.

Principles

• Preserve complex-valued data structure for amplitude-phase coupling.
• Encode physical priors as conditional modulation signals.
• Combine local convolutions with long-range state-space modeling.

Method

CV-SSMNet uses multi-scale complex convolutions, branch-wise CV-SSM encoding with bidirectional scanning, and FiLM-style modulation from seven polarimetric priors to guide feature evolution before a complex-valued classification head.

In practice

• Apply complex-valued SSMs for long-range spatial dependencies.
• Use FiLM for physics-aware feature recalibration.
• Employ block-based data splitting to prevent spatial leakage.

Topics

• PolSAR Image Classification
• Complex-Valued Neural Networks
• State Space Models
• Physics-Aware Machine Learning
• Scattering Prior Modulation
• GeoAI

Code references

• Lucia-210/CV-SSMNet

Best for: Computer Vision Engineer, AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Physics-Aware Complex-Valued State Space Model with Scattering-Prior Feature Modulation for PolSAR Image Classification — CV-SSMNet integrates physics-aware complex-valued state-space models with scattering-prior modulation for enhanced PolSAR image classification.
• Physics-Aware Complex-Valued State Space Model with Scattering-Prior Feature Modulation for PolSAR Image Classification — CV-SSMNet integrates physics-aware complex-valued state-space models and scattering-prior modulation for enhanced PolSAR image classification.
• Physics-Driven Semantic Scattering Structure Understanding of Aircraft Target in SAR Images — SAR aircraft interpretation benefits from physics-driven semantic scattering structure understanding, associating responses with physical components and rigid-body topology.
• GeoDiff-SAR: A Geometric Prior Guided Diffusion Model for SAR Image Generation — Integrating 3D geometric priors with diffusion models significantly enhances SAR image generation fidelity and control.
• Learning with Semantic Priors: Stabilizing Point-Supervised Infrared Small Target Detection via Hierarchical Knowledge Distillation — Hierarchical knowledge distillation with VFM semantics stabilizes point-supervised infrared small target detection.
• DAPGNet: Dynamic Adaptive Physics-Guided Graph Diffusion Network for Hyperspectral Image Classification — DAPGNet enhances HSI classification by dynamically integrating spectral physical priors into graph topology construction and diffusion for improved pixel-relation modeling.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.