To segment crop plant seedlings accurately in natural environment, a segmentation network model based on regional semantic and edge information was presented. Firstly, the U-Net network was used as the backbone network, and the side depth supervision mechanism was used to guide the backbone network to perceive the plant edge information when extracting features. Then, based on atrous spatial pyramid pooling, the feature fusion module was built to fuse the semantic information in the backbone network and the edge information extracted by the edge perception module. The fused feature map would have enough detail information and strong semantic information. Besides, combined with the loss of edge perception and the loss of feature fusion, the joint loss function was defined for the overall network optimization. The experimental results showed that the proposed model can achieve the pixel accuracy of 0.962 and the mean intersection over union of 0.932. Compared with the U-Net, SegNet, PSPNet and DeepLabV3 models，the mean intersection over union of the used model was about 0.07 higher. Therefore, the proposed model can achieve good segmentation effect and generalization ability for crop plant seedlings in natural environment, which can provide important basis for plant location, target spraying, growth recognition and other applications.