Placenta is closely related to the health of the fetus. Abnormal placental function will affect the normal development of the fetus, and in severe cases, even endanger the life of the fetus. Therefore, accurate and quantitative evaluation of placenta has important clinical significance. It is a common method to segment human placenta with semantic segmentation. However, manual segmentation relies too much on the professional knowledge and clinical experience of the staff, and it will also consume a lot of time. Therefore, based on u-net, we propose an automatic segmentation method of human placenta, which reduces manual intervention and greatly speeds up the segmentation, making large-scale segmentation possible. The human placenta data set we used was labeled by experts, which was obtained from prenatal examinations of 11 pregnant women, about 1,110 images. It was a comprehensive and clinically significant data set. By training the network with such data set, the robustness of the model will be better. After testing on the data set, the segmentation effect is basically consistent with the manual segmentation effect.