In view of the flaws of prolonged computation time and limited resolution in traditional deconvolution algorithms during practical applications, an improved beamforming algorithm—DAMAS2-GFISTA—has thus been proposed, which incorporates momentum restart and adaptive strategies. The effectiveness and applicability of the proposed algorithm can be verified based on numerical simulations and comparative measurements conducted in both single-source and dual-source scenarios. The simulation results show that the proposed algorithm can achieve clearer sound source separation and imaging reconstruction with sound sources in close proximity or in the presence of coherent interference. The main lobe width is compressed by about 70% compared to CBF. Furthermore, under the same number of iterations, the computational speed is increased by about 1.64 times and 1.37 times respectively compared to DAMAS2 method in the single and double sound source experiments, exhibiting an enhanced stability and computational efficiency in complex scenarios with multiple sound sources. In practical experiments, whether under single or dual sound source conditions, the proposed method is characterized with excellent positioning accuracy and energy focusing effect, so as to obtain clearer and more accurate sound source images in a shorter calculation time, thus verifying its practicality and superior performance in complex sound field environments.