System of Portable Firearm Analysis Device (PAFAD) / Saadi Ahmad Kamaruddin and Nor Azura Md. Ghani

Forensic ballistics involves the analysis of bullets and bullet impact to determine information which could be used as evidences in a court or other parts of a legal system. Firearm and tool mark examinations (ballistic fingerprinting) involve analysing firearm, ammunition, and tool mark evidence in order to establish vital recognition, whether a certain firearm or tool was used in a crime. In Malaysia, there are not many experts in the arena of ballistics forensics. Therefore, this system is designed to recognize and classify firearms based on bullets found in crime scenes using Artificial Neural Network (ANN). It was found that using the supervised features of firing pin impression images, 100% of the images were correctly classified according to the firearms used. This means that neural network method can learn and validate the numerical features of firing pin impression images with high precision and faster result execution. Due to this the Royal Malaysian Police Force (RMP) may increase their efficiency and expertise in firearm detection relying on this system with confirmed accuracy and lesser effort. There are a few automatic identification system that have been invented, namely IBIS, CONDOR, ALIAS, FIREBALL and EVOFINDER which assist the investigators in making links to crime cases by comparing every characteristic on the bullets and the cartridge case images that have been captured in the current crime to the earlier evidences in the database (Smith, Cross & Variyan, 1995; Smith & Li, 2008; Geradts, Bijhold, Hermsen & Murtagh, 2001). Usually such task uses up great time constraint since it involves comparison with an enormous amount of existing evidences in the records established earlier and also the extensive amount of firearms to be matched. Problems normally surface whenever the investigators make wrong judgement using only visual inspection, also commonly named as human errors. It is also burdensome for these investigators as it is long-established fact that manual inspection necessitates careful planning and expertise. Therefore, the firearm cataloguing using numerical characteristics of the combined firing pin impression image in the feed forward back propagation neural network (BPNN) has become focal point of this project due to the artificial neural networks’ (ANNs) efficiency and effectiveness in the areas of clustering and making classifications (Widrow, Rumelhart & Lehr, 1994; Saadi, Nor Azura, Liong & Aziz, 2010). This has been proven through the ANNs’ adaptability to robotic, finance, medicine fields as well as social studies. In the initial study of this system, a two-layer feed forward BPNN was used to distinguish five types of pistols. The two layers in the network consist of a computationally calculated hidden layer and a target output layer. In our previous studies - Saadi et al. (2010, 2011, 2012, 2013, 2014), firing pin impression has been proven to be adequate and efficient in firearm classification with 100% overall correct classification rate. The classification result observed in our research outperformed the result produced by Leng & Huang (2012). This means that the system developed by us for firearm recognition is expected to assist many practitioners who are involved in forensic ballistics.