Abstract: The number and acquisition time of distributed sensors in complex electromagnetic environments are very limited, which seriously affects the accuracy of reconstructing electromagnetic radio environment maps (REM). This article proposes a sensor layout and optimization method for REM reconstruction. This method is based on compressed sensing theory and utilizes the principle that the smaller the correlation of the sensing matrix, the higher the reconstruction accuracy. The gradient descent method is used to optimize and obtain the optimal expression of the measurement matrix. In addition, a greedy matching based position optimization algorithm is designed for sensor position selection to address the issue of the optimal measurement matrix not being directly mapped to the sensor position. Finally, simulation verification and performance evaluation are conducted on the REM data of campus scenes. The results show that under sparse sampling conditions of 5%-60%, the reconstruction performance of the sensor layout scheme proposed in this paper is superior to other layout schemes, with an average absolute error improvement of about 20%-50%. It can be used to assist in efficient collection of spectrum data and accurate construction of REM s in real environments.