Remote Sensing Image Road Extraction Method: U-Net Model Integrating Attention Mechanism And Residual Modul

This article mainly focuses on studying how to efficiently and accurately extract road information from high-resolution remote sensing images.With the development of remote sensing technology,remote sensing images have been widely used in various fields.Roads,as an important component of urban transportation and infrastructure,have significant implications for urban planning and traffic management.Therefore,road extraction based on remote sensing images has received significant attention.However,due to the complexity of roads and their similarity to other objects,traditional road extraction methods have limitations in terms of weak anti-interference ability,low automation level,and poor extraction results in high-resolution remote sensing images.Therefore,this article proposes a model called Res-AU-Net which combines attention mechanisms and residual modules to extract roads from high-resolution remote sensing images using deep learning techniques.This model aims to overcome the challenges faced by traditional methods.The proposed model is based on the U-Net network model and uses a residual network called Res Net50 to overcome the problem of network degradation caused by the depth of the network.It efficiently extracts abstract deep semantic features by deepening the network and thus improving the model’s training speed and robustness.A focus attention mechanism is employed to enhance road features while suppressing irrelevant features,thereby retaining more contextual topology relationships of roads and increasing the model’s focus on road regions.Furthermore,the model is optimized using dilated convolution to achieve reduced parameters and enhanced efficiency.An improved loss function is also proposed,and the Adam optimizer algorithm is used to increase the model’s accuracy and stability.Finally,the more advanced GELU activation function is employed to improve the model’s accuracy.Experimental results on the Massachusetts Roads dataset demonstrate that the proposed road extraction model has a high extraction accuracy on high-resolution remote sensing images,with overall accuracy,precision,recall,and F1-score of98.7%,89.51%,85.71%,and 87.57%,respectively.When compared with several classical semantic segmentation networks,the proposed approach has better extraction results in terms of road extraction integrity and accuracy,as well as tree,building,and shadow occlusion.Additionally,to extract roads from high-resolution remote sensing images conveniently and efficiently,a remote sensing image road extraction system based on the proposed Res-AU-Net model is designed and implemented.This system uses the Py Qt5 graphical program framework to design a graphical interface and is developed and implemented using Python.It significantly reduces the complexity of road extraction and provides convenience for relevant practitioners.In summary,the proposed Res-AU-Net model combining attention mechanisms and residual modules has high accuracy and robustness in the road extraction task and has a wide range of application prospects.