Abstract: With the continuous evolution of integrated sensing and communications (ISAC) technology and the rapid development of the low-altitude economy, there is an increasingly urgent demand for continuous and accurate sensing of highly maneuverable targets in low-altitude heterogeneous networks. To address the challenges of sensing continuity and high precision in complex electromagnetic environments, this paper proposes a federated learning-driven adaptive handover mechanism for sensing nodes in low-altitude ISAC networks by leveraging existing communication infrastructure. First, a hierarchical hybrid-frequency base station deployment and a multi-transmit single-receive sensing architecture based on separate transmission and reception are employed to construct the foundation for cooperative sensing. Second, in view of the highly dynamic spatial trajectories of aerial targets, a long short-term memory network is utilized to predict target motion. Furthermore, a federated learning framework is introduced to collaboratively train a global “node scheduling decision model” among distributed nodes. This model takes environmental states and target prediction information as input and outputs an optimal transceiver node combination strategy in real time. Finally, through a dynamically elected central node, rapid adaptive handover of sensing nodes is achieved, dynamically aligning with service requirements and enabling persistent target tracking. Simulation results demonstrate that, during full-course tracking of high-speed maneuvering targets, the proposed algorithm significantly outperforms conventional sensing schemes in terms of perception coverage rate and resource utilization efficiency, verifying the effectiveness and superiority of the proposed approach.