Abstract: Agricultural mechanization is a key pillar for advancing agricultural modernization and ensuring national food security, while the level of farmland machinery adaptability (i.e., the degree to which farmland facilitates agricultural machinery operations) constitutes a crucial foundation that determines the efficiency and cost of mechanized operations. However, existing research has mostly focused on demonstrating the necessity of machinery adaptability or conducting cost-benefit analyses after farmland consolidation, lacking a systematic and refined evaluation methodology. This gap makes it difficult to scientifically and precisely guide the planning and implementation of large-scale farmland machinery adaptability improvement projects. To address this research gap, this study is based on multi-source medium- and high-resolution remote sensing data and the deep learning semantic segmentation network HRNet (High-Resolution Network) to automatically extract the spatial distribution information of farmland parcels and roads. The model employs a multi-task learning strategy to simultaneously predict farmland extent, parcel boundaries, and roads, and further refines parcel boundaries using a conditional probability-based post-processing method, thereby providing a fine-scale data foundation for farmland machinery adaptability assessment. On this basis, combined with a 30 m digital elevation model (DEM) and paddy/dryland distribution data derived from a random forest classifier, a comprehensive evaluation system is established from four dimensions: parcel shape, parcel size, surface evenness, and road accessibility. This system includes seven specific indicators. The weight of each indicator is determined by a combined weighting method integrating subjective judgment and the objective CRITIC (Criteria Importance Through Intercriteria Correlation) method, and the parcel-level Farmland Machinery Adaptability Index (FMAI) is subsequently calculated. The method is applied and validated in Dangyang City, Hubei Province. The results show that the HRNet semantic segmentation model performs well in both parcel and road extraction, achieving an area F1 score of 0.92 for farmland and 0.79 for roads, and an edge F1-score (F1-edge) of 0.85 for parcel boundaries and 0.86 for roads. A total of 468,000 farmland parcels and 9,747.23 km of roads are identified in the study area. The spatial distribution of farmland machinery adaptability exhibits significant heterogeneity: high-adaptability areas (FMAI > 77.3) are concentrated in the southeastern Juzhang River alluvial plain, scattered across the central plains and low hilly areas, while the western hilly and mountainous regions generally show low adaptability levels. The grading results indicate that the area proportion of grades 1 and 2 (good adaptability) together accounts for 67.3% of the total farmland area, but their parcel number proportion is only 38.6%, reflecting that fragmented parcels generally have lower adaptability grades. Township-scale attribution analysis reveals that, except for Caobuhu Town, irregular parcel shape and small parcel size are common constraints limiting machinery adaptability across all townships. In hilly and mountainous townships (e.g., Wangdian Township and Miaoqian Township), insufficient road accessibility (road accessibility score 62～70) and poor surface evenness (evenness score 78～85) impose additional constraints. The evaluation method developed in this study can effectively and finely characterize the parcel-level farmland machinery adaptability and its limiting factors, providing scientific data support and a decision-making basis for dynamic monitoring of large-scale farmland machinery adaptability, formulation of differentiated consolidation strategies, and planning of high-standard farmland construction.