Neural network based hyperspectral imaging for substrate independent bloodstain age estimation

Abstract

Being able to determine the age of a bloodstain can be a key element in a crime scene investigation. Many techniques exploit reflectance spectroscopy because it is very versatile and can be used in the field with ease. However, there are no methods for estimating bloodstain age with adequate uncertainty, and the problem of substrate influence is not yet fully resolved. We develop a hyperspectral imaging based technique for the substrate-independent age estimation of a bloodstain. Once the hyperspectral image is acquired, a neural network model recognizes the pixels belonging to the bloodstain. The reflectance spectra belonging to the bloodstain are then processed by an artificial intelligence model that removes the effect of the substrate on the bloodstain and then estimates its age. The method is trained on bloodstains deposited on 9 different substrates over a time period of 0-385 h obtaining an absolute mean error of 6.9 h over the period considered. Within two days of age, the method achieves a mean absolute error of 1.1 h. The method is finally tested on a new material (i.e., red cardboard) never used to test or validate the neural network models. Also in this case the bloodstain age is identified with the same accuracy.

Keywords: Bloodstain age measurement; Hyperspectral imaging; Neural networks; Reflectance spectra measurement.