Targeting task planning for heterogeneous multi-UAV swarms, a collaborative method integrating improved density clustering with LKH-neural network optimization is proposed. To overcome limitations like high computational complexity and low path planning efficiency in multi-region scanning tasks, key innovations are introduced. Firstly, an improved density clustering algorithm balances task region clustering based on spatial distribution and attributes. Secondly, dynamic task cluster allocation balances workloads across heterogeneous UAVs. Finally, a neural network predicts critical LKH parameters to optimize paths and minimize flight time. Simulations show that, compared to K-means + LKH and genetic algorithms, the proposed method offers advantages in task completion time or efficiency as regions and UAVs scale, providing an effective solution for intelligent UAV swarm planning.