Abstract: In response to the low efficiency of manual and semi-automatic pruning operations in dwarfing apple trees during the winter dormancy period,based on the U-Net network model, an improved U-Net network model was constructed by combining VGG16 with U-Net. Using VGG16 as the upsampling feature extraction network, the attention mechanism SEnet was used to enhance the image feature extraction ability, improve segmentation accuracy, and then fuse with the downsampling extracted image features to achieve the end-to-end image segmentation effect. The results showed that the MIoU and MPA of the SE2 network model （improved U-Net network model） on the test set were greater than those of the original U-Net network model;in the SE2 network model, when r=8, the MIoU of the test set, MPA of the test set, F_score of the training set, and F_score of the test set were all the highest, with values of 89.59%, 94.17%, 0.942 806, and 0.944 506, respectively; the performance of the SE2 network model （r=8） was validated on the test bench, and it was found that the segmentation performance of the SE2 network model （r=8） was good.

Key words: improved U-Net, network model, winter dormancy period, dwarfing apple trees, pruned branches, segmentation method