A Robust Dimension Reduction Technique for Hyperspectral Blood Stain Image Classification

This study emphasizes the potential for hyper-spectral imaging in identifying and classifying blood stains in forensic science without physical sampling of crucial evidence. The chemical processes currently used for blood identification and classification can affect DNA analysis, making it necessary to explore novel approaches. Developing algorithms for blood detection is difficult due to the high dimensionality of hyper-spectral imaging and the scarcity of training sample data. This issue is addressed with a new hyperspectral blood detection data set. The proposed work emphasizes 8 dimensionality reduction methods as a preprocessing technique on hyperspectral data. Evaluation of these methods is done using state-of-the-art fast and compact 3D CNN and Hybrid CNN models. The experimental results and analyses demonstrate the challenges of blood detection in hyperspectral data and provide recommendations for future research in this area. Furthermore, this paper highlights the significance of Factor Analysis as a statistical tool for identifying underlying factors that explain patterns and relationships among observed variables.