With the rapid development of artificial intelligence, human- computer interaction systems have attracted much attention, and wireless end-side sensing devices have become the focus of research. Although traditional Wi-Fi sensing has a wide range of application scenarios and many advantages, its sensing data acquisition and access often face problems. In this paper, we explore a new approach based on federated incremental learning for improving wireless sensing techniques. Federated incremental learning combines the characteristics of incremental learning and federated learning to effectively address the degradation of model performance due to the need to recognize new sensory data in the case of distributed data storage and the uneven distribution of old and new categories of sensory data. The method enables local nodes to obtain the classification capability for new category data through the global server with- out directly accessing the original data of the new category. The central server performs global federated incremental training and model aggregation by selecting nodes, and then distributes the model to all sensory endpoints so that all nodes are able to recognize the new category of sensory data. Experimental results show that after about 400 rounds of communication, the global federation incrementally trained model achieves 82.5% classification correctness on the old task and 77.3% on the new task, which improves the performance of the global sensing data recognition model compared to the traditional FedAvg method.