In platoon autonomous driving, a leader vehicle (LV) leads the entire platoon moving forward by instructing the follower vehicles (FVs) to drive cooperatively. To gather sufficient environmental data as the prior knowledge, the LV employs the FVs to perform diverse data collection tasks. Each FV determines the participation willingness and its effort level according to an incentive mechanism presented by the LV and decisions of the social partners in the same platoon. We develop a game theoretic approach to model the strategic interactions among the LV and FVs. After that, the equilibrium solution is derived based on the backward induction method and information collection of the FVs. We further propose a deep reinforcement learning approach to effectively seek the equilibrium solution without the complete information. In our scheme, we apply a multi-agent deep deterministic policy gradient algorithm to train each game player as an agent with the continuous action space. Finally, numerical results are provided to demonstrate the effectiveness and efficiency of our scheme.