Transformer based Compressed Data Recovery for Efficient Data Collection in Wireless Sensor Networks

Data collection is challenging in wireless sensor networks (WSNs) since energy consumption remains a significant constraint. Although energy consumption has increased, most data collection methods incur excessive computation overhead. Compressive sensing is used to minimize the number of sensing samples with a suitable sparse basis. A novel Transformer neural network-based compressed data recovery (TCDR) method has been proposed for extracting temporal correlation structures from compressed signals and reconstructing them with high accuracy. The compressed signal is forwarded to the base station through multi-hop communication. The base station uses an attentionbased Transformer Neural Network (TNN) to recover the compressed signal. This model provides greater energy saving performance than the other baseline models considered for evaluation, which are 5.23 percent, 6.82 percent, 10.26 percent, and 11.47 percent, respectively.