Mangroves are important marine ecosystems, and their canopy structure parameters are important bases for evaluating the growth status and carbon sequestration capacity of mangroves. Currently, mangrove canopy parameters are mainly acquired using airborne LiDAR. The main drawbacks of airborne LiDAR are its high cost and limited accessibility. Meanwhile, the application of UAV oblique photogrammetry in this field is relatively limited. Therefore, the mangrove restoration area in Hainan was taken as the research object. A mangrove instance segmentation dataset was independently constructed. The YOLACT instance segmentation model was innovatively applied to the estimation of mangrove canopy structure. An efficient monitoring scheme for mangrove canopy structure was established. Comparative experiments were carried out with traditional methods including watershed segmentation, point cloud distance discrimination clustering, and layer stacking. The results show that UAV oblique images can realize fine identification of individual trees in the mangrove restoration area. The overall accuracy of the four segmentation algorithms is all greater than 0.80, with F-scores around 0.90. Among them, the YOLACT model performs the best (OA=0.93, F=0.96). In the experiments of estimating canopy structure parameters, the YOLACT model exhibits the optimal fitting effect and the smallest error, and the mean absolute error percentage was reduced by approximately 2%~12% compared with other algorithms. Overall, the YOLACT model adopted in this study shows higher reliability and application potential in individual mangrove tree identification and canopy structure estimation. This provides a technical reference for mangrove monitoring and management.