This invention discloses a collaborative inversion method for multiple attributes of individual trees based on transfer learning, relating to the field of forest ecological
remote sensing monitoring technology. The method acquires
remote sensing images and training samples, generates a
canopy height model based on
LiDAR point clouds, constructs and trains a single-tree segmentation model based on the U-Net++ architecture, and fine-tunes the
encoder weights of the single-tree segmentation model by transferring them to a tree height prediction model and a
forest type classification model. Finally, the
remote sensing images are input into the three models to obtain the single-
tree canopy boundary, predicted tree height, and
forest type, and the integrated output is the multi-attribute inversion result for individual trees. This method addresses the problems of complex data dependencies, non-reusable features, low inversion accuracy and efficiency, and easy error accumulation inherent in traditional independent inversion of single-tree attributes.