Abstract: The detection of subsurface target is an important application of low-frequency synthetic aperture radar (SAR). Landmines, as a typical kind of subsurface targets, are difficult to detect due to the weak echoes and the complexity of environments. Feature extraction is key to effective false alarm suppression, thus to successful landmine detection. A novel method to extract the scattering characteristics of targets is proposed to avoid the robustness and sensitivity issues of traditional approaches, based on the echoes reconstructed from regions of interest (ROI). Then, feature extraction and target discrimination are performed simultaneously based on the sparse time-frequency representation integrated with discriminative criteria, which can improve the accuracy of feature selection and simplify the algorithm flow. The results of the real experimental data demonstrate the feasibility and effectiveness of the proposed method.