With the rapid development of Internet of Things (IoT) technology, IoT devices are playing an increasingly important role in modern society. However, as IoT applications become more widespread and deeply integrated, the demand for higher information transmission rates is continuously increasing. This poses new challenges to the design and performance of IoT systems. Compressed Sensing (CS), an emerging signal processing technique, enables the reconstruction of signals at low sampling rates by performing random sub-sampling and compression. This technology has already achieved considerable success in traditional communication and has shown significant potential in various fields. The primary focus of this paper is to optimize the transformation basis in the principles of Compressed Sensing. Most existing methods rely on predetermined transformation bases, such as the Fourier matrix, wavelet transform matrix, or some sparsity basis that is effective for a specific application scenario. The focus of this paper is on using an adaptive dictionary learning algorithm, which, by training on historical data from a particular usage scenario, ultimately produces a sparsity basis that is tailored to the custom application scenario. This approach helps to some extent in overcoming the limitations associated with the selection of sparsity basis.