The delay-and-sum beamforming method, which has been used in previous studies, is a useful technique to identify noise sources in a high-speed train. However, the method does not provide detailed noise source position in the low-frequency range. Thus, the lowfrequency noise sources of a high-speed train were not identified. Therefore, in this paper, a resolution improvement study was conducted to identify low-frequency noise sources in a high-speed train. The relation between an actual noise source and a beamforming output was first derived using the point spread function. The deconvolution approach of the relation was used to obtain an actual noise source, which was then transferred to an inverse problem. To solve the ill-conditioned inverse problem, an optimization problem was applied and the iteration method was conducted to solve the problem. The algorithm developed in this research was verified with a moving vehicle test. After the successful verification test, the method was applied on a high-speed train. The improved noise map obtained in this research provides the detailed noise sources of a high-speed train.