Jewelry rock discrimination as interpretable data using laser-induced breakdown spectroscopy and a convolutional LSTM...

Area of Science:

• Archaeological Science
• Materials Science
• Artificial Intelligence

Background:

• Accurate classification of ancient jewelry rocks is crucial for historical and cultural studies.
• Traditional methods for rock identification can be time-consuming and may lack precision.
• Developing advanced analytical techniques is essential for characterizing materials from historical sites like Shahr-e Sokhteh.

Purpose of the Study:

• To classify jewelry rocks (agate, turquoise, calcites, azure) from Shahr-e Sokhteh using a deep learning approach.
• To interpret the layer-by-layer effectiveness of the Convolutional Neural Network long short-term memory (CNN-LSTM) architecture.
• To quantitatively determine major chemical elements in jewelry rocks and investigate data interoperability.

Main Methods:

• Utilized a Convolutional Neural Network long short-term memory (CNN-LSTM) deep learning architecture.
• Employed interpretable deep learning-assisted laser-induced breakdown spectroscopy (LIBS) for feature extraction and analysis.
• Applied the Lasso method to investigate data interoperability.

Main Results:

• Achieved excellent performance in classifying various jewelry rocks based on their historical periods and styles.
• Demonstrated the effectiveness of CNN-LSTM in self-adaptively obtaining LIBS features and quantitative chemical data.
• Confirmed high accuracy in the discrimination process, validating the proposed methodology.

Conclusions:

• Laser-induced breakdown spectroscopy (LIBS) effectively combines with deep learning algorithms for jewelry rock classification.
• The CNN-LSTM approach offers a highly accurate and suitable method for material discrimination.
• This study provides a novel, interpretable deep learning framework for analyzing historical artifacts.