Characterization of Gun Shot Residue (GSR) Particle in a Sample. A Statistical Approach

Osnat Israelsohn Azulay, DIFS, Toolmarks and Materials Lab., Jerusalem, Israel
Yigal Zidon, DIFS, Toolmarks and Materials Lab., Jerusalem, Israel
Tsadok Tsach, DIFS, Toolmarks and Materials Lab., Jerusalem, Israel
Yaron Cohen, DIFS, Toolmarks and Materials Lab., Jerusalem, Israel
Micha Mandel, Department of Statistics, The Hebrew University of Jerusalem, Jerusalem, Israel

One of the problems faced by the interrogation units during the investigation of shooting events is the question "Did a certain suspect fire a weapon before he was caught?"

The main method used nowadays by the forensic science to solve this question is particle characterization using a Scanning Electron Microscope (SEM).

In the process of firing weapons, a cloud of tiny particles is released from the firearm and spreads over the shooter and his immediate vicinity.

The tiny size of the particles requires that collecting a sample of particles from the suspect be done in a "blind" manner by pressing a sticky surface to the body from which they seek to collect and then testing this sample by SEM/EDX (Scanning electron microscopy, combined with energy-dispersive X-ray spectrometry).

The main disadvantage of the particle characterization method is the long time required to perform the test. The examination of one sample, can take many hours, due to the large number of intact particles (environmental), which are not GSR, that are found in the sample and to the time spent by the experts on a re-examination of all particles classified by the system as potential to be GSR, in attempt to confirm or reject the results.

An automated search of a sample at the SEM/EDS is evaluated for constituent elements that may identify the particle as being consistent with or characteristic of GSR. These output data serve as a first filter for classification to categories.

A second filter is the expert, when the particles composition classified and reported is reexamined and verified or rejected.
The expert's knowledge consist of considerations in identification relies on composition, morphology, coordinates etc…

These data are gained by the system during the processing but are not used for the first classification of the automated system.

The question dealt with here is: are there in the data properties that can be used as an intermediate filter and so reduce the time consumed by the expert?