Abstract: To further precisely locate forest tree information, segment forest areas and detect dynamic changes in forest resources in real-time, a semantic segmentation model of tree images based on an improved USE-Net convolutional neural network is proposed. Based on the U-Net network, the model adds SE attention module to the transition layer of the network to explicitly model the interdependence between forest tree feature channels, highlight specific forest tree segmentation features and suppress irrelevant regions. The experimental results indicate that the U-shaped structure and the introduction of the SE attention module make the improved USE-Net network have significant advantages in dealing with fuzzy forest boundaries and other aspects, and can accurately segment forest tree areas, which has theoretical value and applications in the field of intelligent scientific management of forest resources.