Integrating Deep Learning Techniques for Enhanced Multi-Target Tracking in UAV Fire Control Systems

This paper presents a novel deep learning-based architecture designed to improve multi-target tracking for UAV fire control systems. The proposed approach integrates a ResNet-50-based Convolutional Neural Network (CNN) for robust feature extraction, a Long Short-Term Memory (LSTM) network for maintaining temporal consistency, and a re-identification module to manage occlusions effectively. These components work together to enhance tracking accuracy, robustness, and real-time performance. The architecture was evaluated on the UAV123 and MOT17 datasets, where it achieved a 15% increase in mean Average Precision (mAP) and a 10% reduction in ID switch rates compared to state-of-the-art methods such as YOLOv4-DeepSORT and Faster R-CNN. The results indicate that the model is highly effective in dynamic, real-world UAV tracking scenarios.