In order to improve the accuracy and efficiency of individual tree biomass estimation in broadleaf forests in complex stands, the biomass of Populus L. and Ulmus pumila L. trees was estimated by UAV LiDAR point clouds. The height and crown width of individual trees were extracted as the basic independent variables through individual tree segmentation, assisted by tree canopy parameters such as crown parameters such as crown area and crown volume to design different variable combinations. Biomass estimation models were constructed using the constant allometric ratio(CAR) model and random forest(RF), extreme gradient boosting(XGBoost), and categorical features gradient boosting(CatBoost), and the accuracy of different variable combinations and models in estimating individual tree biomass was compared and analyzed. The results show that the individual tree parameters of Populus L. and Ulmus pumila L. are fitted with the measured values, with R² values of 0.84 and 0.89 respectively. The extracted values have a good consistency with the measured values, and the estimation model can be constructed by using the extracted values. The results indicate that adding tree canopy parameters can effectively improve the accuracy of biomass estimation models. After the addition of tree canopy parameters, the predicting R² of Populus L. and Ulmus pumila L. increases by 0.02, 0.10, 0.06, 0.04 and 0.03, 0.10, 0.05, 0.07 respectively compared with before the addition. Compared with the CAR model, the machine learning models have better performance in estimating individual tree biomass. The RF models of Populus L. and Ulmus pumila L. have the best performances, with the fitting R² being 0.53 and 0.63 respectively. It is concluded that the UAV LiDAR data can estimate the biomass of individual tree through the extracted parameters of individual tree segmentation. The machine learning model adding tree canopy parameters can achieve better fit and higher prediction accuracy in estimating the biomass of individual trees.