This paper provides a method for route optimization of transmission line based on remote sensing images recognition and the principles of route optimization. This method has the following advantages: Using deep learning technology to automatically identify remote sensing images, and for the problem that the route selection area involves multiple ground features of different sizes and scales, taking the U-Net model as the basic model, introducing deformable convolution and FPN, design an improved U-Net model that can automatically adjust the detection position and fuse multi-scale information according to the shape and size of the target. Due to the higher degree of intelligence, the work intensity of designers is reduced; By quantifying the impact of various constraints on the project cost, the project cost of each route can be calculated more accurately, therefore, the precision is higher and the lack of personal experience of the designer is avoided; The A* algorithm is used to solve the route of the smallest project cost with the smaller search range and higher efficiency.