Surface wave tomography using seismic ambient noise has been applied widely in subsurface characterization, the depth of which ranges from a few meters in urban underground engineering to tens of kilometers in delineation of crustal heterogeneities. Currently, multichannel analysis of surface waves with active and passive sources together with some similar techniques has been widely applied to obtain fine near-surface structures. However, it is still difficult to retrieve accurate dispersion curves at low frequencies with those methods to constrain structures at greater depths. To better retrieve surface wave dispersion data at lower frequencies using ambient noise data acquired by dense linear arrays, we use the extended range phase shift (ERPS) method, which can extract broadband dispersions from the ambient noise crosscorrelation functions (CFs). A broadband dispersion curve is obtained by merging a dispersion curve of medium-to-high frequencies with a dispersion curve of low-to-medium frequencies measured using receivers with different combinations, whereas both curves reflect the local property of the subsurface structure under a subarray with a relatively small aperture. Thus, the horizontal resolution of the inverted models is not compromised, whereas the inverted depth is improved considerably. We first perform a synthetic test to validate the ERPS method, and then apply the method in characterizing the Woxi polymetallic deposit in Hunan, China, where eight 10 km linear arrays using 467 seismic nodes were deployed to record ambient noise. The dispersion data within 0.5–10 Hz are extracted from the CFs using ERPS and then used to invert for the [Formula: see text] models down to at least 2 km. The inverted structures are found to be consistent with the previous geologic studies and known structures from underground mining, and the proposed passive-source imaging method is considered to be economical and practical for ore deposit explorations.