AUTOMATION OF ROAD FEATURE EXTRACTION FROM HIGH RESOLUTION IMAGES

Road feature detection from remotely sensed images is crucial for maintaining an up-to-date and reliable road network, essential for transportation, emergency planning, and navigation. While convolutional neural networks have shown promise in automating this process, existing methods often trade off accuracy for complexity. This study aims to develop an accurate road extraction method without sacrificing computational efficiency. We propose a semantic segmentation neural network combining transfer learning and U-net architecture with minimal complexity. Post-processing techniques are employed to enhance output quality. Our method achieves an F1 score of 0.83 and 95.57% accuracy, outperforming other models on the Massachusetts dataset. This approach demonstrates superior performance and reduced network complexity compared to existing methods.