Fractures play a crucial role in controlling hydrocarbon migration pathways and the spatial configuration of accumulation zones. Accurate and detailed characterization of fracture systems is essential for improving the efficiency of oil and gas exploration and development. However, conventional seismic attribute-based fracture identification methods are highly sensitive to data quality, and often suffer from limited resolution and low stability. These limitations restrict their ability to precisely delineate the geometric structure and spatial distribution of fractures, particularly in areas with complex geology or low signal-to-noise ratios. To address this challenge, this study proposes a high-precision fracture identification method based on multi-scale seismic data derived through matching pursuit decomposition. The matching pursuit algorithm exhibits excellent time-frequency localization capability, allowing seismic signals to be decomposed into frequency components that emphasize different structural scales. This process generates large, medium, and small scale seismic datasets, which better highlight fracture features of corresponding sizes. Based on the decomposed multi-scale seismic data, gradient structure tensor attributes are computed at each scale to extract fracture-related discontinuities. These attributes are then integrated through weighted fusion to form a comprehensive multi-scale fracture attribute volume that captures both the overall structural trends and fine scale fractures. Subsequently, the optimal surface voting algorithm is employed to enhance the continuity and clarity of the identified fractures. It refines fracture representation by scanning orientation and dip angles, selecting candidate fracture surfaces, and applying a voting mechanism to suppress noise and highlight true fracture features. Validation on synthetic models and real seismic data demonstrates that the proposed method significantly improves the spatial resolution and geometric integrity of fracture interpretation. Compared to traditional attribute-based approaches, this method achieves superior accuracy in locating fracture positions and delineating fracture boundaries, particularly for small scale fractures that are otherwise difficult to detect. The methodology presented in this work offers a robust and effective tool for fracture detection and imaging in structurally complex regions. It has strong potential for wide application in the fine scale interpretation of fractured reservoirs and supports improved decision-making in hydrocarbon exploration and production.