Non-Destructive Trace Detection of Explosives Using Pushbroom Scanning Hyperspectral Imaging System

Area of Science:

• Spectroscopy
• Chemometrics
• Imaging Science

Background:

• Trace detection of explosives is critical for security.
• Existing methods like Raman spectroscopy have limitations.
• Hyperspectral imaging (HSI) offers a contactless, non-destructive approach.

Purpose of the Study:

• Investigate HSI potential for trace explosive detection.
• Evaluate the accuracy of a Support Vector Machine (SVM) model.
• Compare full spectrum vs. selected wavelengths for SVM performance.

Main Methods:

• Collected hyperspectral images (400-1000 nm) using a BaySpec sensor.
• Applied image processing, Principal Component Analysis (PCA), and first derivative for feature selection.
• Developed SVM classification models using full and selected spectral ranges.

Main Results:

• Selected 22 optimal wavelengths out of 144 using PCA.
• Achieved 81.11% accuracy with SVM on the full spectrum.
• Attained 77.17% accuracy with SVM on selected wavelengths, reducing computational load.

Conclusions:

• HSI combined with SVM is a viable tool for trace explosive detection.
• Optimizing wavelengths can maintain high accuracy with reduced computation.
• This non-destructive technique offers advantages over traditional methods.