employee
The processes of digitalization inevitably affect the trends in the development of criminalistics, forensic examination and forensic science activity. It seems that special expert systems based on artificial intelligence will have great potential in these areas. Their functioning requires reliable machine learning methods and using neural networks. The article discusses fully connected and Siamese neural networks that have been used as machine learning models in solving problems of assessing the similarity of digital images of firearm marks on bullets and cartridge cases. Correlation methods (congruent matching cells, correlation cells, congruently matching profile segments) are discussed. These methods make it possible to construct a space of compared marks features and determine the dependence of features on each other. The "random forest" and k-nearest neighbors methods were used for machine learning to compare digital images of firearm marks on bullets and cartridges cases, and the author analyzes the results of their application. The results of studies on the preparation of training data by means of artificial formation of clone images with different orientations and with distorted partial signs of traces are presented.
ARTIFICIAL INTELLIGENCE, MACHINE LEARNING, NEURAL NETWORKS, CORRELATION METHOD, TRAINING DATA, CRIMINALISTICS, FORENSIC EXAMINATION
1. Hare E., Hofmann H., Carriquiry A. Algorithmic approaches to match degraded land impressions. Law, probability and risk, 203–221, 2017. Available from: https://doi.org/10.1093/LPR/MGX018. (In Eng.).
2. Sharma A., Kaur M. Automatic segmentation for separation of overlapped latent fingerprints. International journal of computer science and engineering, 484–490, 2018. Available from: https://doi.org/10.26438/ijcse/v6i7.484490. (In Eng.).
3. Liu Y.-Y., Welch D., England R. (et al.). Forensic STR allele extraction using a machine leaning paradigm. Forensic science international: genetics, 102194, 2020. Available from: https://doi.org/10.1016/j.fsigen.2019.102194. (In Eng.).
4. Wang S., Jia S. Signature handwriting identification based on generative adversarial networks. Journal of physics conference science, 042047, 2019. Available from: https://doi.org/10.1088/1742-6596/1187/4/042047. (In Eng.).
5. Park S., Carriquiry A. An algorithm to compare two-dimensional footwear outsole images using maximum cliques and speeded-up robust feature. Statistical analysis and data mining, 188–199, 2020. Available from: https://doi.org/10.1002/ sam.11449. (In Eng.).
6. Fedorenko V. A., Sorokina K. O., Giverts P. V. Classification of firing pin marks images by weapon specimens using fully connected neural network. Izvestiya of Saratov University. Economics. Management. Law, 184–190, 2022. Available from: https://doi.org/10.18500/1994-2540-2022-22-2-184-190. (In Russ.).
7. Song J. Proposed "NIST Ballistics Identification System (NBIS)" based on 3D topographic measurements on correlation cells. AFTE Journal, 184–194, 2013. (In Eng.).
8. Song J. Proposed "Congruent Matching Cells (CMC)" method for ballistic identification and error rate estimation. AFTE Journal, 177–185, 2015. (In Eng.).
9. Sorokina K. O., Fedorenko V. A., Giverts P. V. Estimation of similarity of images of cartridge stop marks by the correlation cell method. Information technologies and computing systems, 3–15, 2019. Available from: https://doi.org/10.14357/ 20718632190301. (In Russ.).
10. Chen Z., Chu W., Soons J. A. (et al.). Fired bullet signature correlation using the congruent matching profile segments (CMPS) method. Forensic science international, 10–19, 2019. Available from: https://doi.org/10.1016/j.forsciint.2019.109964. (In Eng.).
11. Chen Z., Song J., Soons J. A. (et al.). Pilot study on deformed bullet correlation. Forensic science international, 110098, 2020. Available from: https://doi.org/https://doi.org/10.1016/j.forsciint.2019.110098. (In Eng.).
12. Carriquiry A., Hofmann H., Xiao Hui Tai, VanderPlas S. Machine learning in forensic applications. SIGNIFICANCE, 29–35, 2019. Available from: https://doi.org/https://doi.org/10.1111/j.1740-9713.2019.01252.x. (In Eng.).
13. Fedorenko V. A., Sorokina K. O., Giverts P. V. Analysis of the traces on the discharged bullets by the congruent matching profile segments method and the k-nearest neighbors. Information technologies and computing systems, 70–82, 2021. Available from: https://doi.org/10.14357/20718632210108. (In Russ.).
14. Giverts P., Sorokina K., Fedorenko V. Examination of the possibility to use Siamese networks for the comparison of firing marks. Journal of forensic sciences, 2416–2424, 2022. Available from: https://doi.org/10.1111/1556-4029.15143. (In Eng.).
15. Fedorenko V. A., Sorokina K.O., Giverts P. V. Multi-group classification of firing pin marks using fully connected neural network. Information technologies and computing systems, 43–57, 2022. Available from: https://doi.org/10.14357/20718632220305. (In Russ.).
