REAL-TIME WEAPON DETECTION AND BAHAVIORAL THREAT RECOGNITION USING AN INTEGRATED DEEP LEARNING APPROACH

With the increase in public violence, it is vital to have real-time monitoring tools that can detect
weapons and analyze the behavior of individuals who may be a threat. This document proposes a novel
integrated deep learning architecture that combines YOLOv11, Generative Adversarial Networks GANs,
and a convolutional long short-term memory CNN-LSTM structure for effectively identifying weaponry
in addition to examining threats from human behavior. The YOLO model is utilized for the rapid
identiϐication and localization of weapons, while GAN-generated synthetic data is used to enhance the
coverage of the training dataset and create a balanced dataset. CNNs extract spatial characteristics from
images, and by applying them to LSTMs, an understanding of behavior over time temporal analysis is
captured by modelling movement patterns to detect threats when they arrive. All results are based on
data collected from a combined dataset containing weaponnon-weapon conditions with varying levels
of illumination, obstructive objects, and population density. Results indicate superior performance of
98.3, 97.9, and 98.6, and an F1 score of 98.2, which is signiϐicantly better than the performances
obtained with standard approaches that employed only a single model. The combined framework allows
for real-time monitoring at low latency, allowing for easy integration into both smart city surveillance
and public safety systems. Results also indicate that the integration of identiϐication through detection,
augmentation through data generation, and temporal analysis enhances both the reliability of a
framework and its capability to identify threats in complex environments. Keywords Generative
Adversarial Networks GAN, you only look once YOLO, Convolutional Neural Networks CNN, Long ShortTerm Memory LSTM, Anomaly detection.