Advanced deep learning for automatic classification of fired bullets from standard-issue firearms.

Area of Science:

• Forensic Science
• Computer Science
• Artificial Intelligence

Background:

• Gun violence is a global issue with significant loss of life.
• Forensic firearm examination relies on subjective analysis, leading to inconsistent results.
• Automating bullet classification can enhance accuracy and reduce subjectivity in forensic investigations.

Purpose of the Study:

• To develop and evaluate a deep learning model for automated classification of fired bullet markings.
• To improve the accuracy and reduce subjectivity in forensic firearm examination.
• To compare the effectiveness of different image preprocessing strategies for bullet classification.

Main Methods:

• A dataset of 6000 fired bullets from six Chinese law enforcement firearm types was collected.
• Images were captured using the BalScan system and preprocessed using panoramic imaging, land engraved area (LEA) segmentation, and line segmentation.
• A pre-trained ResNet50 deep learning network was fine-tuned for image classification.

Main Results:

• The deep learning model achieved high classification accuracy across different firearm types.
• The LEA segmentation strategy significantly outperformed other preprocessing methods.
• The LEA segmentation approach achieved 97.2% accuracy for classifying bullets from six similar firearm types and 100.0% for distinct types.

Conclusions:

• Deep learning, particularly with LEA segmentation, offers a highly effective solution for automated fired bullet classification.
• This AI-driven approach can significantly improve the efficiency and accuracy of forensic firearm identification.
• The study provides a foundation for AI-driven advancements in forensic science and criminal investigations.