<p>The rapid growth of the Internet of things has led to the generation of large amounts of data by sensor nodes (SNs). It is a challenge for efficient data compression and accurate reconstruction due to the limited energy of SNs. Based on compressed sensing, this letter proposes a novel hybrid architecture RSTNet to address this problem. RSTNet consists of a compression module and a reconstruction module. The data from SNs is compressed by the compression module and then uploaded to the base station for reconstruction using the reconstruction module. Specifically, the residual sensing block in the compression module adaptively generates a measurement matrix to compress data. In the reconstruction module, a residual network and a swin transformer are introduced as the decoder to accurately reconstruct data. By conducting experiments on three different IoT datasets, RSTNet shows remarkable reconstruction results and residual sensing block can effectively help the model improve compression efficiency.</p>