In the process of path planning during UAV data collection, some traditional deep reinforcement learning algorithms suffer from poor path planning, slow convergence speed and high energy consumption. Therefore, for the above problems, a novel approach called RDER-DDPG is proposed, incorporating a continuous action deep reinforcement learning framework and double experience replay with retrace mechanism. By storing superior empirical information and adjusting the sampling ratio between replay buffers, the algorithm achieves faster convergence, higher average reward, and shorter flight trajectories. The retrace mechanism enables effective obstacle avoidance learning and improves path planning efficiency while reducing UAV energy consumption. Experimental results demonstrate significant improvements in task execution success rates and energy consumption compared to conventional DQN and DDPG methods. The proposed RDER-DDPG approach enhances the effectiveness of UAV data collection path planning in IoT networks.