Development of Crime Scene Intelligence Using a Hand-Held Raman Spectrometer and Transfer Learning

Area of Science:

• Forensic Science
• Analytical Chemistry
• Artificial Intelligence

Background:

• Accurate classification of ignitable liquids is crucial for arson investigations.
• Current standard forensic testing (gas chromatography-mass spectrometry) is time-consuming and requires laboratory access.
• Rapid field classification of gasoline evidence is a significant challenge in forensic science.

Purpose of the Study:

• To develop a novel intelligent analytical platform for the rapid, field identification and classification of gasoline evidence.
• To utilize Raman spectroscopy combined with artificial intelligence for enhanced crime scene intelligence.
• To overcome the limitations of traditional laboratory-based forensic analysis for ignitable liquids.

Main Methods:

• Collected Raman spectra from reference gasoline samples with varying octane numbers using a hand-held Raman spectrometer.
• Applied continuous wavelet transformation (CWT) to convert spectral data into image presentations for AI development.
• Employed transfer learning with a pretrained convolutional neural network (CNN), GoogLeNet, for classification model training.
• Compared the CNN model's performance against six conventional machine learning algorithms.

Main Results:

• The transfer learning-based CNN model demonstrated superior performance across multiple benchmarks, including accuracy, precision, recall, F1 score, Cohen's Kappa, and Matthews correlation coefficient.
• The developed classifier maintained significant accuracy (73% and 53%) even with weathered gasoline samples (50% and 25% weathered, respectively).
• Wavelet transforms and transfer learning effectively processed complex Raman spectral data without manual feature engineering.

Conclusions:

• Wavelet transforms combined with transfer learning provide a robust method for processing and classifying complex Raman spectral data.
• The developed nondestructive, automated, and accurate platform can accelerate crime scene intelligence by analyzing chemical signatures detected by hand-held Raman spectrometers.
• This intelligent platform offers a promising alternative to traditional methods for field identification of ignitable liquids in arson investigations.