Aiming at the problem that the DeepLabv3+ network does not fully use the multi-scale feature information generated by the backbone network and the excessive atrous rate in the atrous spatial pyramid pooling (ASPP) leads to poor segmentation results for small targets, the FPDA DeepLabv3+ network is proposed. First, the backbone network was replaced with MobileNetv3 to reduce the number of model parameters, and then the ASPP connection method was changed to a dense connection to capture a larger receptive field. Finally, the Feature Pyramid Fusion Network (FPFN) was used to integrate the high-resolution features effectively. The experimental results show that the mIoU and mPA of the FPDA DeepLabv3+ network on the semantic segmentation data set AeroScapes verification set reached 82.98% and 91.36%, respectively, which fully verified that the network could effectively improve the prediction results on the data set AeroScapes.