Classification of Hyperspectral Images of Explosive Fragments Based on Spatial-Spectral Combination

Area of Science:

• Forensic Science
• Remote Sensing
• Computer Vision

Background:

• Current explosive fragment detection methods often overlook multi-band spectral information, limiting their effectiveness.
• Hyperspectral imaging offers high spectral resolution, enabling detailed analysis for fragment classification.

Purpose of the Study:

• To develop and validate a spatial-spectral joint method for classifying explosive fragments using hyperspectral imaging.
• To leverage deep learning for enhanced feature extraction and classification accuracy.

Main Methods:

• Collected hyperspectral images of explosive fragments in simulated laboratory scenes.
• Employed a deep learning framework combining Convolutional Neural Network-Bidirectional Long Short-Term Memory (CNN-BiLSTM) for spectral classification and U-Net for spatial segmentation.

Main Results:

• The spatial-spectral joint method achieved an overall classification accuracy exceeding 95.2%.
• Demonstrated accurate identification of explosive fragment targets through combined spatial and spectral analysis.

Conclusions:

• Hyperspectral imaging combined with deep learning is feasible for explosive fragment classification in controlled environments.
• Future work will focus on validating this method in outdoor environments using airborne hyperspectral imaging.