·
·
Sunday, 4 October 2026dailybeirut.com
AI
New AI techniques analyze images, sound, and electromagnetic signals to identify counterfeit gold, achieving high accuracy rates in recent studies.
Distinguishing authentic gold from counterfeits no longer relies solely on jeweler expertise or traditional tests. Artificial intelligence technologies have entered the precious metals inspection sector by analyzing jewelry images, studying emitted sounds, and interpreting signals from non-destructive testing methods to detect subtle differences often invisible to the human eye.
A study published in the journal Algorithms in 2026 introduced a model named GoldFormer designed to differentiate between original and fake gold using visual data. The research addressed a complex computer vision challenge: identifying objects that appear nearly identical. Researchers utilized a database called GoldNet, comprising 2,127 real photographs of genuine and imitation gold pieces captured via smartphones under varying conditions, including natural, indoor, and low lighting, as well as diverse angles and backgrounds, without specialized imaging equipment.
The model focuses on minute image details such as surface texture, engravings, color, geometry, and overall shape. While these features may look similar to humans, deep learning algorithms can uncover patterns difficult for the naked eye to discern. In testing, GoldFormer achieved an accuracy rate of 95.02% during five-fold cross-validation, surpassing the approximately 89.8% accuracy recorded by human experts involved in the comparison. It also outperformed other computer vision models tested within the study.
However, this result does not imply that any photo of a gold piece sent to an AI model yields a final verdict on its karat or authenticity. The study tests image classification into original versus fake categories, rather than determining the internal chemical composition of each item through photography alone.
Research has extended beyond visual data. A 2025 study published in Applied Acoustics explored using sound to determine gold purity levels. Investigators analyzed audio recordings of various gold samples, extracting acoustic features known as MFCC delta-delta. They employed ten different machine learning algorithms, including Support Vector Machines (SVM), Random Forests, and XGBoost, to classify the gold according to purity grades.
The best-performing model, SVM, achieved 94.58% accuracy in classifying samples within the experimental categories of 14, 22, and 24 karats. The logistic regression model followed closely with 93.75% accuracy. The underlying concept suggests that physical material properties—such as density, elasticity, and internal structure—influence acoustic characteristics. Consequently, sound signals may carry information that machine learning algorithms can analyze and categorize.
Researchers view this method as a potential future tool for rapid, non-destructive gold inspection. Traditional methods may require significant time, specialized equipment, or risk damaging the sample, whereas other techniques like X-rays detect surface alloy structures without destruction.
A more advanced research direction combines artificial intelligence with Pulsed Eddy Current Testing (PECT), a non-destructive technology capable of capturing signals related to material properties. In a recent study, developers created a neural network named AC-RSN to process signals generated from inspecting gold samples of varying purity and thickness. The team collected 14,400 signals during experiments.
Results indicated that the model classified gold purity levels with 99.27% accuracy in the conducted trials. The neural network was used to reduce noise effects and variations caused by probe temperature while improving signal representation. In this context, AI functions differently than simple image capture; it acts as an analytical tool for data produced by specialized inspection devices. This suggests the future lies in integrating measurement hardware with AI algorithms rather than relying on AI in isolation.
Despite impressive results, AI cannot yet fully replace traditional verification. Studies confirm that artificial intelligence can detect patterns helpful in distinguishing original from fake gold via images, sound, or device signals. However, users cannot rely solely on a smartphone photo or audio recording to determine a gold piece's karat before purchasing it.
Authenticity and purity are distinct concepts. An algorithm might successfully recognize a piece as different from trained counterfeit samples, but precise determination of chemical composition requires appropriate measurement technologies. Currently, the most realistic application is AI serving as an intelligent assistant for gold inspections, analyzing large datasets to find hard-to-detect patterns, while specialized measurement and analysis devices remain the primary source for confirming composition and purity.
