In response to the accuracy and efficiency issues of power system path planning, this paper proposes a novel power system path planning method based on the DeepLabv3+ network, which views the power system path planning problem as an image segmentation problem. On the basis of the DeepLabv3+ method, the spatial attention mechanism is initially introduced to focus on more important information of the image, which is conducive to the restoration of image detail information. Subsequently, a feature reconstruction module is designed by introducing asymmetric convolution and depthwise separable convolution to reconstruct the feature information and obtain the output of the network. Next, a new loss function is devised to optimize the features of different scales of the network. Finally, the test was conducted based on the Vaihingen dataset, and compared with the methods of SegNet, UNet, DANet, and DeepLabv3+, which showed that the proposed method had higher accuracy. The effectiveness of each module was verified through ablation experiments.
