The multi-UAV target coverage problem is vital in disaster relief and industrial inspection applications. Traditional centralized planning methods face challenges in scenarios with numerous targets and UAVs, especially in dynamic environments such as sudden changes in target numbers during missions. Neural network-based heuristics, which use offline-trained models for rapid online decision-making, offer a promising solution. In this paper, we introduce an attention neural network for distributed decision-making to achieve effective target coverage. This network enables agents to make implicit situational judgments by processing the relative positions of targets and other agents, and dynamically plan step goals. Through imitation learning from a Vehicle Routing Problem (VRP) heuristic solver in small-scale scenarios, our network learns to assign weights and applies them to larger-scale scenarios. We validate that our models outperform baseline approaches in coverage reward variation over time.