Construction of UAV trajectory prediction model based on 5G communication and image recognition technology

Accurate drone trajectory prediction is essential for low-altitude safety amid rapid industry growth.Current systems face challenges like poor small-target detection and tracking accuracy.This study presents a complete technical solution featuring a 'perception-computation-decision' architecture, with 25 5G-A ISAC base stations and 46 HD sensors covering 50 km 2 .Algorithm improvements include enhanced YOLOv11 for small targets, a JPTrack module to minimise identity switches, and an LSTM-Attention prediction model.In 30 km 2 urban tests, the system achieved 93.5% detection recall, 82.3% MOTA, and fewer than 15 identity switches per 10 minutes.Trajectory prediction errors were 0.8 m (1 s), 2.1 m (3 s), and 5.3 m (10 s), representing a 46.7%-56.3%improvement over conventional methods.End-to-end latency remained under 5 seconds with efficient resource use.