Compressive sensing based matching pursuit algorithm can estimate the channel state information of communication system with shorter pilot sequences. It has the advantages of lower computational complexity and less number of pilots. However, the estimation accuracy is relatively low. Assigning different weights to the estimated channel according to the estimation error can improve the estimation performance under low Signal Noise Ratio (SNR) conditions. Nevertheless, it will reduce the estimation accuracy under high SNR conditions. To address this issue, an improved weighted matching pursuit algorithm is proposed. The information of weight value is obtained by training the channel data. Adjusting the weight values adaptively with the iterations can further improve the estimation accuracy. The simulation results show that the proposed algorithm can improve the estimation performance under both low and high SNR conditions.