Antique Authentication System

Abstract

Antique Authentication System An industrial-grade multimodal system for the authentication of antique patina and materials, integrating physicochemical testing with AI fusion decision-making to replace subjective expert judgment, and deliver reproducible and auditable authentication conclusions. Input Requirements: Trace sampling of 0.2 mg, plus four-modal data acquisition (AFM, GC-MS, IR/Raman spectroscopy, and microscopic imaging). Output Format: Three-category decisions (Authentic / Uncertain / Fake) coupled with tamper-proof blockchain-style audit reports. Core Modules and Methodologies AFM Module Nano-mechanical fingerprinting, extracting 12 key metrics (e.g., adhesion, hysteresis, slope, AUC, etc.). Zero-point detection of the second derivative is applied to identify contact points. Force-curve features are used to distinguish naturally aged patina from artificially forged counterparts. GC-MS Module Mass-to-charge ratio (m/z) ranging from 50 to 1000 is divided into 1024 bins, with the top 256 peaks extracted. Modern contaminants (e.g., phthalates, silicone oils, etc.) are detected as indicators of forgery. Spectroscopy Module Focuses on characteristic peaks at 1700, 1740, 2850/2920, and 1650 cm⁻¹. Peak intensity and intensity ratios are leveraged to determine oxidation and esterification characteristics. Imaging Module Processes 256×256 microscopic images to extract features including grayscale histograms, texture direction entropy, GLCM (Gray-Level Co-occurrence Matrix), Gabor filters, and wavelet transforms. These features enable the detection of unnatural textures and regular traces left by artificial processing. Fusion Engine Employs weighted voting and Bayesian posterior probability. The default weight allocation is 35% for AFM, 30% for GC-MS, 20% for spectroscopy, and 15% for imaging. Dynamic weight adjustment and drift detection are supported. Key Performance and Evidential Examples - Performance Metrics: Single-modal accuracy ranges from approximately 68% to 82%. The four-modal fusion achieves a reported accuracy of 94.3%, with a blind-test AUC of 0.96, a false positive rate (FPR) of less than 2%, and a false negative rate (FNR) of less than 3%. - Typical Cases: The system can distinguish authentic artifacts from the Palace Museum’s collection from modern forgeries (e.g., a comparison between a purported "Ming Dynasty Huanghuali piece" and auctioned items). It can also detect modern plasticizers, silicone oils, and other contaminants as evidence of forgery. - Self-Testing Capability: Built-in SelfTest functionality generates 100 synthetic samples for end-to-end stress testing. Support for Blind Test manifests ensures third-party validation. Audit and Engineering Design Evidence Chain Each test records sample metadata, raw data hash values, test results, and the hash value of the previous record, forming a tamper-proof chain to facilitate independent verification. Robustness Design When dependencies (Pillow, XGBoost, tifffile, pymzml) are missing, the system automatically falls back to robust strategies. Type cleaning is performed before data fusion to prevent runtime interruptions. Openness and Governance Released as open-source software under the MIT License, it provides synthetic datasets, pre-trained weights, and user documentation. It is recommended that configurations, models, and evidence be uploaded to a blockchain and integrated into CI/blind-test workflows. Deployment and Compliance Recommendations Short-Term Implementation 1. Conduct blind tests (N≥200) on a controlled sample library first. 2. Feed expert review results back into the system for rule fine-tuning or retraining. 3. Incorporate SelfTest and  ledger.verify  into the CI pipeline. Medium- to Long-Term Governance 1. Use KMS/HSM (Key Management Service/Hardware Security Module) to manage provenance secrets. 2. Establish a change control process (PR → CI → Blind Test → Deployment). 3. Monitor score distribution and review rates to trigger retraining when necessary. Industry Impact It is recommended to collaborate with auction houses, museums, and regulatory authorities to integrate scientific testing into transaction and supervision processes, thereby reducing information asymmetry and the cost of forgery. This system’s design covers a multi-dimensional evidence chain spanning nano-mechanics, chemical fingerprinting, and visual texture analysis. Equipped with engineering and auditing capabilities, it significantly enhances the objectivity and verifiability of antique authentication. To promote its application in real-world business scenarios, it is essential to incorporate the supplemented code and configurations into the repository, enforce mandatory self-testing in CI, conduct large-scale blind tests with expert review feedback, and prioritize the integration of KMS and evidence package signing processes.