Multigroup classification of firing pin marks with the use of a fully connected neural network

The article discusses the problem of classifying the images of firing pin marks using a fully connected neural network. The purpose of this work was to study the effectiveness of using clone images of firing pin marks with modified features in order to improve the quality of training of fully connected neural networks, as well as to evaluate the accuracy of multigroup classification of firing pin marks made by different weapons using this type of network. The scientific novelty of the work is the formation of clone images of firing pin marks in order to increase the number of objects in the training sample and to artificially increase the feature diversity of objects of each class. The conducted studies have shown that the classification accuracy of the analyzed objects reaches approximately 84% in the case of a fixed value of the classifying criterion and 94–98% when classified according to the three largest signals on the output neurons. The work is of interest to developers of software for automated ballistic identification systems.