Road Extraction from U2Net High Resolution Remote Sensing Images with Attention Constraints

In order to solve the problem that Convolutional Neural Networks(CNN) have poor modeling ability for global information and may degrade performance when processing target objects at different locations or scales, this paper proposes an attention-constrained U2Net (CBU2Net) high-resolution remote sensing image path extraction method. Based on the original U2Net model, spatial and channel attention mechanisms are introduced. Network model was tested on the Massachusetts road data set, and the accuracy rate, recall rate, Intersection over Union(IoU) and F1-score reached 96.72%, 98.71%, 96.55% and 98.25%, respectively. Compared with the mainstream methods U2Net and U-Net, IoU increased by 0.04% and 0.34%, and F1-score increased by 0.03% and 0.18%, respectively. The experimental outcomes demonstrate that the introduced approach exhibits commendable efficacy, successfully enhancing the precision of road segmentation.