HYPERSPECTRAL IMAGING SYSTEM IN FORENSIC SCIENCE: BACKGROUND CORRECTION VIA NEURAL NETWORK

In recent years Hyperspectral Imaging System (HIS) has seen many technological developments that have led Forensic Science to use this powerful tool for crime resolution. Hyperspectral Imaging System is very advantageous technique as it is fast, portable, non-destructive, and integrates conventional spectroscopy with imaging, allowing to have both the spectral and spatial information of an object. Of the various biological traces found at a crime scene, the blood is the most important information carrier, as it provides useful information for solving a crime. DNA analysis makes it possible to trace the suspect or victim, while the chronology of events can be reconstructed by estimating the age of the blood. Hemoglobin is the main chromophore of the blood, and its spectrum dominates the blood spectrum in the visible region allowing to identify its main features. However the blood spectrum is influenced by the substrate on which it is deposited. Texture and colors of the materials can affect the bloodstain and its spectrum, degrading it or blending with it, such as dark and red colors, unlike light colored materials on which the blood is visible to the naked eye. This study aims to obtain a valid method for background correction, exploiting Multilayer Perceptron (MLP) neural network, generated through a Bayesian optimization approach. This new approach enables high performance background correction with a mean percentage error of 0.1% and a standard deviation of ± 0.05%.