Abstract: Multi-beam water column detection has been widely applied owing to its advantages of high resolution, accuracy, and full coverage. However, due to the sidelobe effect and environmental interference, there is a lot of noise in the multi-beam water column image, which seriously reduces the accuracy of target feature extraction. In this study, based on the in-depth analysis of the distance and phase characteristics in multi-beam water column signals, a gas plume feature extraction method using density hierarchical clustering is proposed. Firstly, a detailed study of the principle of multi-beam imaging and the noise characteristics is carried out, and an adaptive local threshold segmentation method is employed to achieve overall denoising for effectively reducing the influence of noise. Subsequently, by using the density hierarchical clustering algorithm to accurately eliminate the remaining noise, the quality of water images can be further improved. The processed water column images show a significant enhancement in quality while retaining the contour information of plume bubbles, thus the effectiveness of the proposed method is validated. Through this research, a feasible technical approach is provided for target feature extraction in the field of multi-beam water column detection.