TY - GEN
T1 - Methodology for Weapon Detection in Social Media Profiles using an Adaptation of YOLO-V5 and Natural Language Processing Techniques
AU - Baez-Velazquez, Alberto
AU - Hernandez-Suarez, Aldo
AU - Sanchez-Perez, Gabriel
AU - Toscano-Medina, Karina
AU - Portillo-Portillo, Jose
AU - Olivares-Mercado, Jesus
AU - Meana, Hector Manuel Perez
N1 - Publisher Copyright:
© 2022 IEEE.
PY - 2022
Y1 - 2022
N2 - Weapon identification has been a hot topic in the area of Object Recognition in recent years. However, its appli-cation has been virtually explored in social media. This work focuses on the detection of weapons in profiles that explicitly advocate their procession, both graphically and textually. This is a challenge, since access to a dataset is difficult; and once the samples are obtained, the dimensions and attributes of the images can vary significantly. In addition, the possession of a weapon does not imply that any offense or crime is being committed. To tackle these challenges, this manuscript presents a regularized adaptation of a Fast-Convolutional Neural Network (F-CNN) based on YOLO-V5, to merge and improve the results of the algorithm, along with a textual fingerprinting technique, to first corroborate if the intent of the post contains red flags of crime and violence. The results demonstrate that regularized adaptive models, mainly using Data Image Augmentation techniques, along with text classification, can provide better performance on unstructured data, such as those found in social media.
AB - Weapon identification has been a hot topic in the area of Object Recognition in recent years. However, its appli-cation has been virtually explored in social media. This work focuses on the detection of weapons in profiles that explicitly advocate their procession, both graphically and textually. This is a challenge, since access to a dataset is difficult; and once the samples are obtained, the dimensions and attributes of the images can vary significantly. In addition, the possession of a weapon does not imply that any offense or crime is being committed. To tackle these challenges, this manuscript presents a regularized adaptation of a Fast-Convolutional Neural Network (F-CNN) based on YOLO-V5, to merge and improve the results of the algorithm, along with a textual fingerprinting technique, to first corroborate if the intent of the post contains red flags of crime and violence. The results demonstrate that regularized adaptive models, mainly using Data Image Augmentation techniques, along with text classification, can provide better performance on unstructured data, such as those found in social media.
UR - http://www.scopus.com/inward/record.url?scp=85147540571&partnerID=8YFLogxK
U2 - 10.1109/ROPEC55836.2022.10018723
DO - 10.1109/ROPEC55836.2022.10018723
M3 - Contribución a la conferencia
AN - SCOPUS:85147540571
T3 - 2022 IEEE International Autumn Meeting on Power, Electronics and Computing, ROPEC 2022
BT - 2022 IEEE International Autumn Meeting on Power, Electronics and Computing, ROPEC 2022
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2022 IEEE International Autumn Meeting on Power, Electronics and Computing, ROPEC 2022
Y2 - 9 November 2022 through 11 November 2022
ER -