A Non-Destructive Computational Assaying framework uses a fine-tuned MobileNetV2 architecture to analyze microscopic surface textures. The system, developed by Ayush Jain and hosted on Harvard Dataverse, reportedly achieves 92.5% accuracy in differentiating genuine .925 sterling silver and flags 98% of fraudulent hallmarks. The dataset, last updated in April 2026, is derived from a specialized 'Jeweler’s Macro-Library'.
Use Cases
- Train a CNN to authenticate sterling silver purity based on microscopic surface morphologies.
- Develop a hallmark fraud detection system based on casting texture analysis.
- Deploy a material verification model to consumer-grade edge devices like smartphones.
- Mitigate financial risk in the bullion value chain using AI-driven verification.
Strengths
- Reported model performance metrics: 92.5% accuracy for silver purity and 98% for fraudulent hallmark detection.
- Framework is designed for deployment on accessible, consumer-grade edge devices.
- Dataset is hosted on an authoritative academic platform, Harvard Dataverse.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count and file formats are unknown, which may limit suitability assessment.
- Data may reflect bias inherent to the specific 'Jeweler’s Macro-Library' source.
Provenance
- Source
- Harvard Dataverse
- Collection Method
- Images likely collected to form a specialized 'Jeweler’s Macro-Library' for training a convolutional neural network.
- Time Range
- null
- Freshness
- Last updated 2026-04-13 15:12:47; freshness should be verified.
- Geography
- null