Classification of secondary explosives with a 1D convolutional neural network technique using terahertz time-domain...

Area of Science:

• Spectroscopy
• Machine Learning
• Materials Science

Background:

• Terahertz time-domain spectroscopy (THz-TDS) is a key technique for explosive identification.
• Integrating THz data with machine learning enhances the speed and accuracy of explosive molecule classification.

Purpose of the Study:

• To investigate the use of THz-TDS for identifying premium explosives.
• To evaluate the performance of various machine learning algorithms for classifying explosives based on THz spectral data.

Main Methods:

• Collected terahertz absorption spectra and refractive index data for RDX, HMX, TNT, PETN, and Tetryl (0.2-3 THz).
• Applied Principal Component Analysis (PCA) for feature extraction.
• Utilized supervised machine learning algorithms (SVM, KNN, RF) and a 1D-CNN for classification.

Main Results:

• Supervised machine learning models achieved over 90% prediction accuracy.
• A 1D-CNN model outperformed traditional methods, reaching prediction accuracies greater than 95%.
• THz-TDS combined with 1D-CNN demonstrated high efficiency and practicality for explosive identification.

Conclusions:

• THz-TDS is a powerful spectroscopic tool for explosive detection.
• Machine learning, particularly 1D-CNN, significantly improves the classification accuracy of explosives using THz data.
• The combined approach offers a practical and efficient solution for rapid explosive identification.