Related Experiment Video
Updated: Nov 28, 2025
07:05
Applying Hyperspectral Reflectance Imaging to Investigate the Palettes and the Techniques of Painters
Published on: June 18, 2021
2.7K
Blood Stain Classification with Hyperspectral Imaging and Deep Neural Networks
Kamil Książek1, Michał Romaszewski1, Przemysław Głomb1
1Institute of Theoretical and Applied Informatics, Polish Academy of Sciences, 44-100 Gliwice, Poland.
Sensors (Basel, Switzerland)
|November 25, 2020
Summary
Deep learning models show promise for hyperspectral imaging (HSI) substance classification in forensics, especially in challenging inductive scenarios. While performance varied, specific deep learning architectures significantly outperformed traditional methods in classifying blood and similar substances.
Area of Science:
- Forensic Science
- Computer Vision
- Machine Learning
Background:
- Hyperspectral imaging (HSI) generates high-dimensional data valuable for non-invasive substance classification.
- Deep learning neural networks offer potential for processing complex HSI datasets.
- Forensic science can benefit from accurate substance classification at crime scenes, such as identifying blood stains.
Purpose of the Study:
- To evaluate the performance of various deep learning models for hyperspectral blood stain classification.
- To compare deep learning approaches against traditional methods like Support Vector Machine (SVM).
- To assess model performance in both transductive and inductive classification scenarios.
Main Methods:
- Experiments were conducted using 1D, 2D, and 3D Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Multilayer Perceptrons (MLPs).
- Hyperspectral Transductive Classification (HTC) and Hyperspectral Inductive Classification (HIC) experimental setups were employed.
- Model evaluation included t-SNE and confusion matrix analysis to detect undertraining.
Main Results:
- In the transductive case, deep learning models showed comparable performance to MLP and SVM, with Overall Accuracy ranging from 98-100% on easier datasets and 74-94% on more difficult ones.
- In the more challenging inductive case, selected deep learning architectures achieved higher Overall Accuracy (57-71%), outperforming non-deep models by up to 9 percentage points.
- Per-class error analysis revealed significant model dependency, with 3D CNNs and RNNs showing the best performance in inductive scenarios.
Conclusions:
- Deep learning models offer significant advantages over traditional methods in inductive hyperspectral classification for forensic applications.
- The optimal deep neural network architecture for hyperspectral data remains an open research problem, requiring tailored approaches.
- Accurate classification of blood and blood-like substances using HSI is feasible, but model selection is critical for real-world forensic applications.
