Abstract: Aiming at the performance degradation of most source number estimation algorithms under the condition of complex environment and small number of snapshots, a source number estimation method based on two-dimensional dominant mode rejection (DMR) whitening filter and agglomerative hierarchical clustering (AHC) is proposed. Firstly, the signal covariance matrix is reorganized by means of a method of enhancing the covariance matrix, and the signal information is further extracted to replace the orig- inal signal eigenvalues of original signal. Secondly, the one-dimensional sample features are mapped to a two-dimensional plane using a local density method and enhancing the eigenvalues, so as to achieve dimensionality increase. Finally, each eigenvalue is calculated by an AHC algorithm, while focusing on the distribution of the eigenvalues of the noise subspace, and noticing the fact that the distribution of the reconstructed noise eigenvalues is denser than that of the pre-reconstructed noise eigenvalues; and an improved algorithm for estimating the types of information sources is obtained by classifying the latter, using the minimum distance as the criterion for describing the intra-cluster similarity. The simulation results show that the proposed algorithm exhibits better estimation performance in the white noise environment, and is not sensitive to the number of snapshots; meanwhile, it can also better estimate the type of sources in the color noise environment. In the snapshot experiment, the proposed algorithm has better estimation performance, reaching an estimation accuracy of more than 90 % while maintaining estimation stability under the condition of small snapshots.