The investigation of target capture in urban environments has emerged as a prominent area of research in recent years. The dense road networks in urban areas prevent significantly for the pursuers to capture evaders. To address this issue, this paper proposes the so-called Deep Q-Network (DQN) with State Decomposition (DQNSD), which decomposes different types of state information based on the DQN framework. And it also introduces a mutation strategy to optimize the algorithm. Finally, through simulation, it is demonstrated that the algorithm proposed in this paper outperforms the original algorithm in terms of performance and effectiveness.