Publicação
Gun model classification based on fired cartridge case head images with siamese networks
| Resumo: | The identification of the firearm model that triggered the firing of a bullet is an important forensic information that, historically, has been done by trained examiners through visual inspection using microscopes. This is an extensive and very time-consuming process that requires the examiners to be trained to identify and compare the fired cartridges. This paper proposes an automated objective method for binary classifying pairs of fired cartridge head images as belonging to the same or different classes, using siamese neural networks (SNNs). With this technique, an accuracy of up to 70% was reached by using firing pin mark images as the input of the SNN. For the training and optimization of the network this paper also analyses and presents different image preprocessing approaches. |
|---|---|
| Autores principais: | Valentim, S. |
| Outros Autores: | Fonseca, T.; Ferreira, J.; Brandão, T.; Ribeiro, R.; Nae, S. |
| Assunto: | Siamese neural networks Image preprocessing Firearm model classification |
| Ano: | 2021 |
| País: | Portugal |
| Tipo de documento: | documento de conferência |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | ISCTE |
| Idioma: | inglês |
| Origem: | Repositório ISCTE |
| Resumo: | The identification of the firearm model that triggered the firing of a bullet is an important forensic information that, historically, has been done by trained examiners through visual inspection using microscopes. This is an extensive and very time-consuming process that requires the examiners to be trained to identify and compare the fired cartridges. This paper proposes an automated objective method for binary classifying pairs of fired cartridge head images as belonging to the same or different classes, using siamese neural networks (SNNs). With this technique, an accuracy of up to 70% was reached by using firing pin mark images as the input of the SNN. For the training and optimization of the network this paper also analyses and presents different image preprocessing approaches. |
|---|