Abstarct: One of the important issues in the daily life is the security of society. In recent years, dangerous and unusual behavior detection systems are being developed and used. One of the major issues with recognizing unusual and dangerous behaviors and preventing undesirable events is the automatic detection and differentiation of sounds such as explosions, thunderstorms, bullets and so on. In this thesis, we determine the type of weapon from the sound signal generated by the firing. For this purpose, we use a multi-laxyer perceptron neural network and features such as real-time casts, linear prediction predictors, and collet-scale coefficients on the Mel scale to identify and distinguish guns. The weapons used in this dissertation include 6 types of lumbar guns, which are the sound signals used by the industry. The obtained results of the proposed method show that the approach of using the multilxayer perceptron neural network and extracting appropriate features is an appropriate approach in detecting the type of weapon.