Abstract: To solve the problems of sensitivity to parameter setting and poor separation accuracy in applying density clustering algorithm in underdetermined blind source separation, we propose an optimized density clustering algorithm based on an improved salp swarm algorithm. First, we use the wavelet threshold noise reduction to de-noise the noisy observation signal to remove interference points and improve the performance of density-based spatial clustering of applications with noise algorithm (DBSCAN). We then apply the salp swarm algorithm incorporating the firefly perturbation strategy to find the domain radius of DBSCAN, solve the insensitivity of the algorithm to parameter settings, improve the algorithm's robustness, and obtain the optimal mixed estimation matrix. Finally, the source signal is reconstructed by the L₁-norm minimization algorithm. Simulation results show wavelet threshold noise reduction preprocessing can effectively reduce interference points. Compared with the traditional density clustering algorithm, the proposed algorithm has a better estimation of mixed matrix and better separation accuracy.