With the rapid development of the theory and algorithms for sparse recovery in finite dimension, compressive sensing (CS) has become an exciting field that has attracted considerable attention in signal processing, such as sub-Nyquist sampling systems, sound imaging and reconstruction, wavelet denoising, compressive sensor networks, and so on. Moreover, the broad applicability of CS framework has already inspired some notable investigation in the context of array processing. The problem of acoustic source identification can be investigated from a limited number of measurements delivered by a microphone array as a basis pursuit problem, which has been developed in the context of compressive sensing, and the CS beamforming can be proved to be better than the conventional beamforming even in its near-field focusing version based on spherical waves. Focused beamforming is a typical method used to localize the position of acoustic sound sources in the near field of the measurement array, and can be a jointly reconstructed source powers and positions. Many super-resolution focused beamforming approaches have been developed to overcome the Rayleigh resolution limit of conventional focused beamforming. Especially, turning to the compressive sensing (CS) framework, we are able to exploit the inherent sparsity of the underlying signal in space domains to achieve super-resolution for the focused beamforming even in a noisy and coherent environment with few snapshots.Prior research has established CS as a valuable tool for array signal processing, but it is mainly from a theoretical point of view, and its application to underwater acoustic sources localization has been developed only for very limited scenarios. In this paper, we present an underwater noise sound source near-field localization method based on a sparse representation of vector sensor array measurements. By utilizing the sparsity approach, the new localization methods can jointly reconstruct source powers and positions, and enforce sparsity by imposing penalties, based on the l1-norm, to improve the integrated performance. By comparing with other source localization methods, such as the conventional focused beamforming, MVDR focused beamforming, and the maximum likelihood focused beamforming, the performance of compressive focused beamforming and the typical focused beamforming using pressure or vector sensor array is analyzed in detail, especially under noisy conditions, and coherent sources. Simulation and experimental results demonstrate that this new approach has a number of advantages over other source localization techniques, e.g. increased resolution, improved robustness to noise, limitations in data quantity and correlation of the sources, as well as lower levels of background interference. It is feasible to apply the proposed approach for effectively localizing and identifying underwater noise sound sources.