Intelligent technologies for target detection and engagement by UAVs based on artificial intelligence and simulation data

The subject matter of the article is the methodology for developing integrated intelligent systems that enable automated detection, classification, tracking, and engagement of targets by unmanned aerial vehicles (UAVs) using artificial intelligence, computer vision, and simulation modeling technologies. The goal is to design an architecture and experimental environment for evaluating the efficiency of deep learning models, particularly the YOLO family, in real-time operation across different hardware platforms, considering resource and power constraints, as well as to quantitatively assess improvements in key system characteristics. The tasks to be solved are: developing a microservice-based system architecture using Kubernetes and ROS2; creating specialized datasets for model training; integrating the AirSim, ArduPilot SITL, and Mission Planner simulators into a unified testing environment; and conducting comparative performance analysis of various YOLO versions on platforms ranging from low-power ARM/NPU to high-end GPU. The applied methods are simulation modeling, automated neural network training, model quantization, inference optimization using TensorRT, and statistical analysis of the obtained metrics. The following results were obtained. A unified experimental methodology and a hardware–software platform were developed, providing a complete research cycle—from data generation and annotation to model performance evaluation. Experimental tests demonstrated that YOLOv8n and YOLOv11n offer the best balance of accuracy and speed on low-resource platforms. The most effective onboard solutions were identified as the Google Coral Dev Board Micro and NVIDIA Jetson Orin Nano. A simulation testbed was deployed that successfully models UAV flight, target acquisition, and engagement processes. Conclusions. The results confirmed the feasibility of using simulation data and a microservice-based approach for developing autonomous intelligent UAV control systems, demonstrated measurable improvements in performance, computational efficiency, and energy consumption, as well as provided practical guidelines for selecting the optimal “model–platform” configuration. The scientific novelty of the obtained results lies in the creation of a comprehensive research environment that integrates modern artificial intelligence tools, simulation modeling, and hardware optimization into a unified reproducible structure, enabling objective evaluation of target detection and engagement algorithms in real time.