Abstract: Above-ground biomass (AGB) is an important parameter reflecting the growth status of individual trees. Timely and accurate estimation of single-tree AGB is essential for monitoring the restoration effectiveness of restored forest areas in open pit mines. This study aimed to investigate the potential of feature-level fusion of multispectral and single-tree structural feature information (SFI) to improve the accuracy of single-tree AGB estimation. First, a canopy height model was constructed for the study area based on tilted image data and combined with the watershed algorithm to realize single-tree segmentation, obtaining the spectral information extraction window and single-tree SFI, and at the same time using multispectral data to extract a total of 57 indexes of canopy spectral texture information. Second, the AGB estimation model was constructed by combining three algorithm models, namely, back propagation neural network, random forest, and support vector machine, based on each type of feature variable and their combinations. Finally, the model fitting effect before and after data fusion is tested to screen out the algorithmic model with optimal accuracy. In order to avoid the limitation of model parameters, particle swarm and genetic algorithms are used to further optimize the optimal model. The results of the study show that although the estimation model based on a single variable type can characterize the size of single wood AGB, the estimation accuracy has some limitations. The coefficient of determination (R²) and mean absolute error (MAE) of the SFI-based AGB model were R²=0.406-0.628 and MAE=0.895-1.439 kg, respectively; and the AGB model based on multispectral data had R²=0.355-0.627 and MAE=0.712-1.497 kg. The data fusion significantly improved the estimation of the single-timber AGB. accuracy, and among the three machine learning models, the RF model had the best estimation effect, in which R²=0.740 and MAE=0.535. After the algorithm optimization, the model based on the GA-RF algorithm had the best mono-wood AGB estimation effect, in which R²=0.910 and MAE=0.420 kg, and compared to the model based on mono-wood SFI combined with RF, the R² was improved by 44.91% to 74.66%, and MAE decreased by 53.07%～62.73%. This study can provide a feasible method for monitoring the growth status of single logs using UAV platforms in the restored forest area of open pit mines.