Automatic Stones Classification through a CNN-Based Approach.

Area of Science:

• Geology
• Computer Science
• Artificial Intelligence

Background:

• Stone classification is crucial for geological surveys and cultural heritage.
• Existing methods may lack efficiency and accuracy for diverse stone types.
• Calabrian quarries possess unique geological characteristics requiring specialized identification tools.

Purpose of the Study:

• To develop an automated system for classifying stones from Calabrian quarries.
• To implement a hybrid deep learning and machine learning approach for stone recognition.
• To compare the performance of various machine learning classifiers in conjunction with deep learning feature extraction.

Main Methods:

• Utilized Convolutional Neural Networks (CNNs) for feature extraction.
• Employed Transfer Learning (TL) to leverage pre-trained models.
• Integrated CNNs with machine learning models: Softmax, Multinomial Logistic Regression (MLR), Support Vector Machine (SVM), k-Nearest Neighbors (kNN), Random Forest (RF), and Gaussian Naive Bayes (GNB).
• Developed a two-stage hybrid model combining Deep Learning (DL) and Machine Learning (ML).
• Established an image acquisition process to create a comprehensive stone typology database.

Main Results:

• The proposed two-stage hybrid model demonstrated effectiveness in stone classification.
• Performance comparison of different DL and ML combinations provided insights into optimal configurations.
• The system successfully aids in the identification of stone typologies within the Calabrian region.

Conclusions:

• The developed automated system offers a robust solution for classifying Calabrian quarry stones.
• The hybrid DL-ML approach, particularly with Transfer Learning, shows significant potential for geological material recognition.
• This technology can enhance geological resource management and the preservation of local stone heritage.