A New Road Extraction Method from Satellite Images Using Genetic Programming
Abstract
Extracting roads from satellite images is an important task in the field of computer vision with a wide range of applications. However, efficient road extraction from satellite images remains a complex challenge due to issues such as data labeling and the diversity of road features. Existing methods often struggle to balance accuracy, robustness, and interpretability. Genetic programming (GP) is based on a flexible and interpretable structure that is robust and does not require a large amount of data support. We position the road extraction problem as a binary semantic segmentation task and introduce GP algorithms. First, an approach for extracting pixel neighborhood features is proposed, and features from multiple images in the DeepGlobe road extraction dataset are extracted. Then, an advanced feature construction method based on GP is employed. Finally, these advanced features are utilized for training classifier and classification to achieve road extraction. We have validated the effectiveness of the approach on the DeepGlobe road extraction dataset. The results demonstrated that the proposed approach exhibits superior performance compared to traditional classification methods and multilayer perceptron (MLP) in terms of accuracy, generalization, and interpretability. This study provides a valuable reference for the integration of GP into the domain of road extraction from satellite images, showcasing their potential to enhance the accuracy and efficiency.
