An Efficient WiFi CSI-Based Multi-Task Modeling Method for Indoor Activity Recognition and Localization: LBA-TCN

With the rapid development of the Internet of Things (IoT) and wireless sensing technologies, contactless perception has become a key enabler for intelligent environments. WiFi Channel State Information (CSI), due to its advantages such as obstacle penetration, low cost, and no need for additional hardware, has been widely applied in tasks including activity recognition, localization, and vital sign monitoring. In this context, how to efficiently utilize CSI data for joint multi-task perception has become an important research focus in the field of wireless intelligent sensing. This paper proposes a multi-task deep learning model, LBA-TCN (Lightweight Bahdanau Attention Temporal Convolutional Network), which integrates multi-scale convolution, temporal modeling, and attention mechanisms for simultaneous activity recognition and indoor localization. The model employs three convolutional branches with different receptive fields to extract multi-scale spatial features and incorporates a Temporal Convolutional Network (TCN) to capture temporal dependencies in CSI sequences. A lightweight additive attention mechanism is further designed to enhance the representation of key temporal features. Experimental results show that LBA-TCN demonstrates strong stability and generalization in multi-class recognition tasks, verifying its potential in WiFi-based multi-task indoor perception applications.