Road Detection in Urban Areas Using Random Forest Tree-Based Ensemble Classification

The rapid growth in using remote sensing data highlights the need to have computationally efficient geospatial analysis available in order to semantically interpret and rapidly update current geospatial databases. Object identification and extraction in urban areas is a challenging problem and it becomes even more so when very high-resolution data, such as aerial images, are used. In this paper, we use Random Forest Classifier tree based ensemble to enhance the extracting accuracy for roads from very dense urban areas from aerial images. Both the spatial and the spectral features of the data are used for pre-classification and classification. Comparisons are made between the RF ensemble and other ensembles of statistic classifiers and neural networks. The proposed method is tested to aerial and satellite imagery of an urban area. The result shows that the RF ensemble enhances the overall classification accuracy for roads by 8 %. Also, it demonstrates that the approach is viable for large datasets due to its faster computational time performance in comparison to other ensembles.