Uso de espectroscopia de emissão ótica com plasma induzido por laser (LIBS) associada à aprendizagem de máquinas para análise de resíduos de disparos de armas de fogo (GSR)

Abstract

In a situation involving an accident or assault with a firearm, it is possible to identify the person responsible for the shots using chemical markers. This is done by identifying elements already catalogued in conventional ammunition, such as Ba - barium, Pb - lead and Sb - antimony. However, ammunition such as NTA (non-toxic ammunition) does not have chemical markers as clear as conventional ammunition, requiring alternative methods to identify a possible shooter. Therefore, this work proposes an experimental protocol capable of providing such chemical markers, while still providing a margin of accuracy in identification and avoiding false positives. To this end, laser-induced breakdown spec- troscopy (LIBS) is used to provide the elemental signature of the samples, which in turn serves as information for training the Random Forest(RF) classification model, which is based on statistical information metrics. With this supervised machine learning method, it is possible to analyze which parameters are used to distinguish a shooter from a non- shooter, making it explainable. The samples were collected at a shooting range, under protocols established by SENASP (National Secretariat for Public Security), consisting of 2, 6 and 8 shots with NTA ammunition, and 1 and 6 shots with conventional ammu- nition. For comparison, samples were taken from 0 shots (samples from the hand of an individual who did not come into contact with the firearm). The results show chemical markers identified from Cu I, Cu II, Ti I and Mo I atomic transitions, the classification model was able to achieve an accuracy of over 90% and has a generalization power high enough to distinguish a shooter from a non-shooter regardless of the type of ammunition, conventional or NTA.