Adversarially Robust and Explainable Vehicle Damage Claim Authentication Using Multi-Modal Forensics and LangGraph Workflow Orchestration
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
The growing digitalization of vehicle insurance claim workflows has significantly increased fraudulent submissions involving manipulated, AI generated, or reused damage images. Existing verification systems often rely on manual inspection and isolated AI modules, making them vulnerable to sophisticated tampering and adversarial perturbations. This research introduces a unified conceptual and empirical framework for adversarially robust and explainable vehicle damage claim authentication. The framework integrates five key components: (i) a Siamese network based vehicle identity verification module, (ii) a CNN based damage assessment network for real damage recognition, (iii) a multi modal image forensic engine combining PRNU noise analysis, Error Level Analysis (ELA), JPEG artifact profiling, and deep forgery detection, (iv) an adversarial robustness layer employing FGSM and PGD defense mechanisms, and (v) an Explainable AI (XAI) component utilizing Grad CAM, SHAP, and tamper region overlays for human-auditable transparency. All modules are orchestrated through a LangGraph inspired workflow that enables dynamic routing, modular execution, and complete auditability. Experimental validation on a dataset of$\mathbf{1 0, 0 0 0}$genuine and$\mathbf{6, 0 0 0}$manipulated images demonstrates$\mathbf{9 2. 4 \%}$vehicle identity verification accuracy, 94.7 % tamper detection AUC, 89.3% adversarial robustness, and 91.6% explainability usefulness, establishing the framework's reliability and scalability. The proposed hybrid conceptual experimental architecture serves as a foundation for real world deployment in intelligent, automation ready vehicle claims authentication systems. Furthermore, the proposed LangGraph orchestrated pipeline is the first unified system that combines vehicle identity verification, multi modal forensics, adversarial defense, and explainable reasoning within a single adaptive workflow. This integration enhances resilience against attacks while maintaining auditor level interpretability, making it highly suitable for real world insurance claim processing.
Fields of Study & Subjects
Digital Media Forensic Detection
Adversarial Robustness in Machine Learning
Advanced Steganography and Watermarking Techniques
Publication details
- Journal / source
- International Conference on Intelligent Systems for Communication, IoT and Security (ICISCoIS)
- Pages
- 1–6
- Publication date
- 8 January 2026
- Language
- EN