Abstract: Above-ground biomass (AGB) serves as a key for assessing crop photosynthetic product accumulation and ultimately determining final yield. Accurate monitoring of maize AGB is of significant importance for achieving precise crop management and ensuring food security. Unmanned Aerial Vehicle (UAV) remote sensing, with its advantages of high spatial resolution, strong maneuverability, and timely imaging, provides a more flexible and effective technical means for accurate AGB estimation at the field scale. An experiment was conducted at the Wangxingjian Family Farm in Dafeng District, Yancheng City, Jiangsu Province. To address the issue of soil background interference when directly extracting parameters from UAV multispectral imagery for maize AGB estimation, this study proposes a "segmentation-then-modeling" approach. Based on the processed UAV data, vegetation was first segmented from the soil background using the Green Normalized Difference Vegetation Index (GNDVI). Subsequently, five spectral band reflectance values (RBs), fourteen vegetation indices (VIs), and forty texture features (TFs) were extracted exclusively from the vegetation areas. Pearson correlation coefficients between each feature and AGB were calculated to select features, forming four distinct variable combinations—RBs, VIs, TFs, and their fusion (RBs+VIs+TFs)—as model inputs. Estimation models were then constructed using two machine learning algorithms: Random Forest (RF) and Partial Least Squares Regression (PLSR). The results demonstrated that: 1) When using single feature types, Vegetation Indices (VIs) provided the best predictive performance (e.g., for Fresh AGB with RF: R²=0.860, RMSE=72.408 g/m², rRMSE=22.874%, MAPE=25.574%; for Dry AGB with RF: R²=0.844, RMSE=28.161 g/m², rRMSE=21.166%, MAPE=23.171%). In contrast, Reflectance Bands (RBs) showed the poorest results (e.g., for Fresh AGB with PLSR: R²=0.526, RMSE=133.325 g/m², rRMSE=42.118%, MAPE=31.910%; for Dry AGB with PLSR: R²=0.423, RMSE=54.200 g/m², rRMSE=40.736%, MAPE=26.517%). The performance of Texture Features (TFs) was intermediate (e.g., for Fresh AGB with RF: R²=0.827, RMSE=80.458 g/m², rRMSE=25.417%, MAPE=34.689%; for Dry AGB with RF: R²=0.825, RMSE=29.881 g/m², rRMSE=22.458%, MAPE=27.428%). 2) The multi-feature fusion model (RBs+VIs+TFs) outperformed all single-feature models. Compared to models using only RBs, VIs, or TFs, the fused model for Fresh AGB showed R² increases of 0.356, 0.209, and 0.132, and RMSE reductions of 66.841 g/m², 44.287 g/m², and 30.347 g/m², respectively. For dry AGB, corresponding R² increases were 0.339, 0.179, and 0.012, with RMSE reductions of 19.413 g/m², 11.303 g/m², and 0.877 g/m². Optimal performance was achieved by the PLSR model for Fresh AGB (R²=0.882, RMSE=66.484 g/m², rRMSE=21.003%, MAPE=25.374%) and the RF model for Dry AGB (R²=0.844, RMSE=28.226 g/m², rRMSE=21.215%, MAPE=21.569%). 3) Overall, the RF algorithm performed better than PLSR, with a more pronounced advantage in single-feature scenarios (e.g., R² differences of 0.265 and 0.368 for Fresh and Dry AGB estimation using RBs, respectively). Although PLSR showed moderate performance with single features, its accuracy improved substantially with multi-feature fusion (RBs+VIs+TFs) for AGB prediction, even surpassing RF for Fresh AGB estimation. Multi-feature fusion narrowed the performance gap between the two algorithms.In summary, the proposed method effectively eliminates soil background interference and is particularly suitable for accurate AGB estimation in crops with sparse canopy coverage, thereby providing a scientific basis for UAV-based precision agriculture monitoring.