Hybrid Anomaly Detection for Bullion Coin Authentication Leveraging Acoustic Signature Analysis
1 st Krzysztof Siwek, 2 nd Tran Hoai Linh, 3 rd Tomasz Gryczka, 4 th Maciej Stodolski · 2026
Read Hybrid Anomaly Detection for Bullion Coin Authentication Leveraging Acoustic Signature Analysis by 1 st Krzysztof Siwek, 2 nd Tran Hoai Linh, 3 rd Tomasz Gryczka, 4 th Maciej Stodolski (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-27803.txt.
Introduction
Siwek and three colleagues propose a non-harmful check framework for bullion coins based on sound tests and deep neural networks. Each coin's sound print, set by metal, mass, and shape, is captured through tapping. A two-model setup pairs a spectrum coder for oddity spotting with a coin-type sorter. A dynamic oddity threshold and data boosts address noise and limited data range. The combined system achieves high care telling real coins from top fakes across varying settings and tools.
Catalog record
| CREATOR | 1 st Krzysztof Siwek, 2 nd Tran Hoai Linh, 3 rd Tomasz Gryczka, 4 th Maciej Stodolski |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 41239 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.27803
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Hybrid Anomaly Detection for Bullion Coin Authentication Leveraging Acoustic Signature Analysis 1 st Krzysztof Siwek, 2 nd Tran Hoai Linh, 3 rd Tomasz Gryczka, 4 th Maciej Stodolski Abstract The verification of bullion coin authenticity is essential for maintaining integrity within the precious metals market; however, the increasing sophistication of counterfeits has rendered traditional inspection methods insufficient. This paper proposes a non-destructive verification framework based on acoustic frequency analysis and deep neural networks. The methodology leverages the unique acoustic fingerprint of a coin, a physical signature determined by its material composition, mass, and geometry, captured through mechanical excitation. We implement a synergistic dual-model architecture consisting of an autoencoder that reconstructs the spectrum for anomaly detection and a deep learning classifier for coin type identification. To address the challenges of environmental noise and limited dataset diversity, a dynamically calculated anomaly threshold and data augmentation techniques were employed. Experimental results demonstrate that the integrated system achieves high precision in distinguishing authentic specimens from high-quality counterfeits, maintaining stability across varying recording conditions and devices. Beyond bullion authentication, the study highlights the scalability of the proposed non-destructive testing method for assessing the safety of critical components in the automotive and aerospace industries. I Introduction Since the dawn of the first civilizations, the exchange of goods has relied on materials with relatively stable value, such as grain, salt, or ore. However, these early media suffered from significant drawbacks, they were often perishable, cumbersome to transport, and difficult to standardize. Over time, precious metals, specifically gold and silver, emerged as the superior choice due to their durability, divisibility, and ease of transport. The transition to precious metals as a primary currency took various forms, but the true turning point in monetary history was the introduction of coins [ 2 , 7 ] . These standardized units of weight and purity featured increasingly complex imagery, symbols, and graphic elements on their surfaces. This standardization reduced the need for constant, exhaustive testing of the payment medium during every transaction, allowing coins to become a universal measure of value and a reliable method for storing wealth. I-A The persistent challenge of counterfeiting Unfortunately, as coins became the dominant form of payment, counterfeiting followed almost immediately. By the Middle Ages, as trade expanded, forgery became a systemic problem. Counterfeiters employed a variety of techniques to mimic the appearance of genuine currency while minimizing actual precious metal content. Common methods included mixing precious metals with cheaper alternatives such as copper (debasement), casting coins from inexpensive lead or copper alloys, and covering them with a thin layer of gold or silver (plating). Despite severe legal penalties, often including the death penalty, counterfeit money continued to circulate. This ongoing threat to the monetary system forced the development of early verification methods, such as weighing, measuring dimensions, and visual inspections. reverse obverse Fig. 1: One-ounce Australian Kangaroo Silver Coin More advanced techniques included the displacement method based on Archimedes’ Principle, which allowed assessment of material density but required specialized tools and knowledge to execute [ 11 ] . I-B Sound analysis for verification One of the most enduring and intuitive verification techniques is the ”ping test”: listening to the sound a coin makes when dropped or struck against a hard surface. A genuine coin produces a clear, bell-like ring, whereas counterfeits made from inferior alloys often sound dull or muffled. Historically, this method relied entirely on the experience of the merchant, whose ear had to be finely tuned to the specific resonance of authentic coinage [ 22 ] . However, as a standalone technique, it lacked the precision necessary to detect high-quality, modern counterfeit (fake) coins. Modern technology has transformed this intuitive practice into a rigorous scientific discipline. Advances in microphone sensitivity and signal processing algorithms now enable precise recording and analysis of a coin’s acoustic signature. Every coin has a unique sound-frequency profile defined by its composition, mass, and shape [ 1 , 8 ] . I-C Neural networks for bullion authentication Bullion coins (also known as investment coins) are ideal candidates for this type of acoustic analysis. Typically minted from high-purity gold, silver, or platinum, their value is derived from their metal content rather than their face value or numismatic rarity. Because bullion markets are often perceived as stable stores of value during periods of economic uncertainty, preserving market confidence is essential. In addition, modern bullion coins are manufactured with exceptional precision, resulting in negligible differences in mass or geometry between authenticated and counterfeit specimens [ 5 ] . The spectrogram shown in the figure was created by striking a coin with a hard object. The coin was held on both sides at a point on the axis of rotation passing through the center. Analysis of the spectrogram reveals several distinct peaks at resonant frequencies, located primarily in the lower part of the spectrum, i.e., below 10 kHz. These peaks are located around 3500 Hz, 3750 Hz, 8700 Hz, and 15500 Hz. The resonant peaks appear as horizontal, brighter lines and are a symptom of sustained oscillations at these frequencies after the initial strike [ 3 ] . The intensity of the resonant modes decreases over time, as seen in the gradual, slow disappearance of the bright horizontal lines towards the right of the spectrogram. Immediately after the strike, the highest signal energy is observed (approximately 0.1s), where the entire frequency range is excited by the strike. After the impact pulse fades, only resonant modes dominate the spectrogram. This behavior is typical of the natural vibrations of coins, where the resonance frequencies correspond to the natural vibration frequencies, which in turn result from the coin’s material properties and shape. Fig. 2: Mechanical impact spectrogram By leveraging Artificial Neural Networks (ANNs) to process these acoustic signals, we can move beyond human subjectivity. ANNs are uniquely capable of identifying the complex, nonlinear patterns within a coin’s frequency spectrum that indicate its true material integrity. This research aims to develop a fast, simple, and effective method for bullion verification, providing investors with a robust tool to navigate the precious metals market with certainty. II Metodology At the foundation of this methodology is treating acoustic response as a principal means of assessing a coin’s physical characteristics. The recorded sound is considered an intrinsic acoustic signature, distinctive to each coin type. In this study, the acoustic signal is produced by mechanically striking the coin with a rigid object. This external excitation induces the coin’s natural vibrational modes, which are directly governed by its metallurgical composition and geometric accuracy. By capturing and analyzing these vibrations, the system obtains high‑resolution data that forms the basis for reliable coin verification. TABLE I: Resonance peaks in the acoustic signal of the 1 oz Australian Kangaroo silver coin Peak number Frequency [Hz] Relative Amplitude [dB] 1 3770.00 0.00 2 3505.62 -0.36 3 8648.75 -6.78 4 15258.12 -22.18 Table I shows the observed resonance frequencies in the acoustic signal, which were extracted using the Fourier transform (FFT) from the signal shown in the previous spectrogram. By detecting local maxima in the power spectrum, accounting for minimum peak separation in the frequency domain, and applying a prominence threshold, it was possible to extract only the dominant resonance peaks from the signal [ 14 ]