The Internet of Everything is a necessary link for building smart cities, smart manufacturing, and other scenarios. And the demand for many forms of data such as high definition and high precision for the Internet of Everything puts great pressure on the storage and transmission of data. Starting from the end of the Internet of Things (IoT), this paper investigates a compressed sensing algorithm adapted to the end of the IoT sensing and computing system to reduce the power consumption and data communication volume of the end sensing and computing system and proposes an optimization algorithm for iteratively updating the measurement matrix and sparse dictionary. First, fix the sparse dictionary and use adaptive gradient descent to make the Gram matrix infinitely approximate the unit matrix and the matrix determined by the sparse dictionary structure to obtain the optimized measurement matrix; then, use the sample data obtained from this matrix to perform sparse dictionary learning and use the results as the dictionary input for the next round of measurement matrix; finally, improve the measurement matrix and dictionary performance through continuous iterative updating. Simulation experiments show that the measurement matrix and sparse dictionary obtained by this method have better performance for image signal acquisition reconstruction in a low sampling rate environment compared with traditional optimization algorithms.