Convolutional Neural Network for Headstamp Recognition

Abstract

Weapons and armor are transported all over the world after manufacture. One way to connect the place of manufacturing to a cartridge is by using the headstamp, a stamp on the bottom of the casing. This information could, for example, be used to obtain information in conflicts or to get more insight into the exports of ammunition of a country. In this bachelor thesis, I trained a convolutional neural network to automatically detect and identify the characters of headstamp codes in order to allocate armors all around the world to their country of origin. The trained CNN had an accuracy of 71.33%. This result implies that a CNN can be used for the recognition of characters of a headstamp code. Further research has to be done in order to automatically recognize the characters in images of headstamp codes and automatically use this information to obtain the manufacturer of the cartridge.