In the actual engineering process

landslide identification relies on manual interpretation

has problems such as low efficiency and strong subjectivity.To solve these problems

takes the areas along typical power line projects as the application scenarios and adopts Deep Learning algorithms to identify the landslide areas.This method not only improves the accuracy of landslide identification but also verifies its feasibility in engineering applications.This paper constructs a landslide dataset by integrating open-source images and high-resolution aerial survey data.In this paper

multi-level features are extracted by using the image encoder of the improved SAM

and then high-precision segmentation results are generated through the cross-feature fusion decoder.The research results show that the method adopted in this study performs well in terms of the identification accuracy of landslide boundaries

with the identification accuracy reaching 90.31%.Deep Learning methods reduce manual intervention and can provide reliable technical support for the identification of geological disasters in power line engineering.