A Multi-UAV Task Scheduling Framework Based on Convolutional Neural Networks

This study proposes an enhanced convolutional neural network (CNN)-based task scheduling framework for multi-UAV cooperative networks engaged in joint target search and tracking missions. To address the computational complexity inherent in traditional tree search methods for task allocation, we develop a novel neural-augmented optimization architecture that integrates CNN-based node cost prediction with systematic search space pruning. The core innovation lies in employing deep convolutional networks to estimate the minimum cost bounds of decision nodes, enabling efficient pruning of non-optimal branches in the search tree while preserving optimality guarantees. Furthermore, we establish a comprehensive simulation environment incorporating various state-of-the-art artificial intelligence algorithms for performance benchmarking. Simulation evaluations confirm the superior computational efficiency of our CNN-augmented scheduling framework when benchmarked against conventional branch-and-bound methods, while simultaneously demonstrating marked improvements in mission success rates compared to contemporary reinforcement learning-based schedulers. The proposed methodology provides a viable solution for real-time mission planning in dynamic multi-UAV systems, effectively balancing computational efficiency with effectiveness.