Abstract: To address the limitations of traditional phase unwrapping (PU) approaches in differential interferometric synthetic aperture radar (DInSAR) technology which performs poorly in unwrapping low-coherence deformation interferograms in mining areas, an L²-norm PU approach driven by the matrix pencil model is proposed. Using the improved matrix pencil local frequency estimator, the wrapped phase gradients of deformation interferograms are accurately extracted in four directions: horizontal, vertical, and the two diagonals. These gradients are then integrated with the four-direction weighted least squares PU algorithm, utilizing the Gauss-Seidel iteration method to compute the unwrapped deformation phase of the mining surface. It not only preserves the advantages of high unwrapping efficiency and smooth phase surfaces but also effectively mitigates noise interference during the PU process. Experimental results based on mining subsidence simulation data demonstrate that the proposed approach ensures high continuity in the unwrapped phase, minimizes residuals and unwrapping errors, and improves PU accuracy by approximately 73.5%, 46.2%, and 74.1%, respectively, compared to the minimum-cost flow, statistical-cost network-flow, and traditional four-direction weighted least squares PU methods. Additionally, three coal mining collapse sites in Yongcheng Mining Area of Henan Province were selected as experimental subjects to carry out tests. Experimental results based on Sentinel-1A data demonstrate that the proposed approach provides more accurate, complete, and smooth unwrapped deformation phase compared to commonly used PU algorithms in engineering applications, such as network-flow and max-flow/min-cut methods. This new approach significantly enhances the accuracy and noise resistance of the L²-norm unwrapping model.