Since the multi-dimensional classical multiple signal classification (MD-MUSIC) algorithm requires a huge amount of computation for multi-dimensional grid search, an improved ant colony optimization (IACO) algorithm with across-neighborhood search (ANS) capability is therefore proposed in this paper. The scheme uses the elite reverse learning strategy to construct the initial solution population, and the optimization method of ant colony is dynamically adjusted by introducing global ANS and Gaussian kernel function local search. Finally, the nonlinear global optimal solution of the MD-MUSIC estimation method is obtained. The experimental results indicate that the new method effectively reduces the calculation without losing the estimation accuracy. Moreover, the algorithm has faster convergence performance and better stability.