Abstract:[Objective] To address the low efficiency of manual surveys and the limited scale ability of existing remote sensing techniques in bund extraction, this study proposes a multi-source, multi-scale extraction framework of “macro-scale identification–micro-scale extraction” aimed at improving the accuracy and efficiency of farmland bund identification and parameter estimation.[Methods] At the county scale, satellite remote sensing imagery and digital elevation models (DEM) were integrated to establish an interpretation system for bund types (stone bunds vs. soil bunds) and to analyze their distributional responses to topographic factors. At the small watershed scale, high-resolution UAV-derived digital orthophoto maps (DOM) and digital surface models (DSM) were fused to extract geometric parameters of bunds, including length, width, height, and distribution coefficient. The extraction accuracy was evaluated using deviation error (DE), coefficient of determination (R2), and root mean square error (RMSE), and linear regression models were developed for parameter inversion.[Results] 1) Satellite remote sensing imagery demonstrated strong applicability for county-scale identification and characterization of farmland bunds. Validation in the Xiaziling watershed indicated that recognition accuracies for bund count and area both exceeded 91%. At the county level, more than 300,000 bunds were identified in Zhong County, with a total area of approximately 8 km2 and a bund coefficient of about 3%. 2) UAV imagery allows high-precision extraction of bund parameters at the watershed scale with absolute deviations of less than 2 m (length), 0.1 m (width), 0.3 m (height), and 2% (coefficient). The extraction accuracy ranked as: bund length > bund width > bund height > bund coefficient. Soil bunds consistently showed higher extraction accuracy than stone bunds. The regression models demonstrated a strong fit (R2 > 0.78; RMSE < 1.6). 3) In Zhongxian County, bunds are primarily distributed within elevation ranges of 300～600 m and slope ranges of 6°～15°, showing a‘first increasing then decreasing’trend with respect to both elevation and slope. Soil bunds have a wider distribution compared to stone bunds. Bund patches in valley areas are large and dense, whereas those in mountainous areas are closely spaced but fragmented, exhibiting linear or point-like patterns. [Conclusion] The proposed multi-source remote sensing method is suitable for both large-scale mapping of bund distribution and fine-scale parameter extraction. The regression models provide reliable estimates of bund metrics, offering technical support for ecological restoration and agricultural engineering planning in mountainous regions.