SOLVING CLASSIFICATION ISSUE IN FORENSIC EXAMINATION: NEURAL NETWORK APPLICATION

Abstract

According to information technologies development their application area in human life
and activity becomes wider too. In particular, information technologies, information and expert systems have become widespread in forensic medical examination. Bruisings are the most
common investigation object during forensic medical examination carrying out. As part of their
study, it is supposed to establish the damage fact, its limitations statute, and, if possible, the
identification of the object that caused the damage.
This article highlights a convolutional neural network application for solving bruising classification issue according to one of harm to health signs – the hematomas limitations statute on
the victim’s body. There are five classes used. The starting work part included the tools choice,
images preparing and the convolutional neural network training using a data set that corresponds the victims’ injuries photofixing real results. TensorFlow version 2.8, Keras API version 2.3.0, and the Python programming language were chosen as development tools. The neural
network operation results were analyzed while different activation functions variants were
used, as well as with a different neurons number in the first fully connected layer.The trained
convolutional neural network efficiency with the best model is close to 90%.
The result of the work – the developed expert system module – makes it possible to attribute
bruises to one of the five specified categories, allowing through the images processing using
convolutional neural networks automating the injuries analyzing process received by the injured.
Using this module helps to increase the forensic medical examination conclusion objectivity
and reliability also.