Spatial crowdsourcing (SC) recruits online workers to perform geographically relevant tasks based on their real time locations, which exposes sensitive location information. Workers’ heterogeneous privacy needs make uniform protection ineffective and degrade service quality. To address these issues, this paper proposes a deep reinforcement learning-based personalized privacy protection scheme for spatial crowdsourcing. First, the proposed scheme constructs a personalized privacy protection model based on spatiotemporal features, which dynamically assigns personalized privacy protection intensity to different users by quantitatively analyzing the association between factors such as location, semantic sensitivity, access frequency, and access duration with individual privacy needs. Secondly, deep reinforcement learning algorithms are employed to adjust the model dynamically according to changes in the environment, thereby achieving an adaptive balance between privacy protection and task completion efficiency, and a bidirectional privacy-preserving mechanism combining Local Differential Privacy (LDP) and homomorphic encryption is designed to protect the privacy of both workers and requesters. Again, to achieve a more comprehensive task allocation strategy, this scheme proposes a multi-factor optimized task assignment mechanism that combines parameters such as worker reputation and task response latency. Finally, compared with the existing schemes, the proposed scheme can maintain a high accuracy rate for task assignment while ensuring personalized privacy protection, thereby improving the efficiency and reliability of the crowdsourcing system.