Explosive Detection - Raman Spectrum Recognition

Uses Deep Convolutional Neural Networks for classification of chemicals present in an explosive from their Raman Spectrum.

Steps

1. Data Preprocessing
   • Smoothening by Savitzky Golay filter
   • Derivatization of spectra
   • Normalization
2. Principal Component Analysis (PCA) for dimentionality reduction. (Optional)
3. Deep Neural Network (Multi-layer Perceptron architecture) for classification.

Hardware and Software used

Hardware	Specs
Processor	Intel i7
RAM	4 GB
HDD	1 TB
GPU	12GB NVIDIA Tesla K80 GPU

Software	Details
Operating System	Linux
Development Environment	Google Colab, Jupyter notebook
Language and Libraries	Python and libraries (Pandas, Scikit-learn, Matplotlib), Tensorflow, Keras

Dataset Used

• Spectra of chemicals including Sulphur, Acetone, Urea, DNT, DMSO, AN, Ethyl aclcohol, Nepthalene, HMX, PNBA etc.
• Data for Open-souce distribution: RRUFF Dataset consisting of 3700 spectrum samples.

Reference

Liu J, Osadchy M, Ashton L, Foster M, Solomon CJ, Gibson SJ. Deep convolutional neural networks for Raman spectrum recognition: a unified solution. Analyst. 2017;142(21):4067-74.