Abstract: As oil fields enter the middle-late period of development, the accuracy of reservoir description becomes increasingly high. Simply utilizing seismic attributes is no longer sufficient to meet development needs. Geostatistical inversion, as a commonly used inversion method, can combine seismic inversion results with stochastic simulation principles. It has achieved certain results in shallow river facies. However, its application in middle and deep delta facies reservoirs is relatively limited. To achieve better characterization of thin reservoirs, the sparse Radon transform is first performed in the time-frequency domain using an iterative convergence threshold algorithm to improve the quality of pre stack data; To sovle the issue of difficulty in compensating for low frequencies in seismic signals, a compressive sensing algorithm is adopted to simultaneously compensate for both low and high frequency information, thereby improving the bandwidth and main frequency of seismic data; Finally, sedimentary facies and net to gross ratio information are incorporated into the geological statistical inversion process to constrain the model establishment and inversion results. The practical application results show that the predicted results are consistent with the actual drilling, which can meet the distribution characteristics of thin reservoirs in the middle and deep layers, and confirm the reliability of this method research.