Research On Recognition Of Earthquake Damaged Road Based On High Resolution Remote Sensing Imagery

After the devastating earthquake, the most practical device for disaster reduction depends on timely information acquisition and effective assessment. In recent years, with timeliness, synchronism, less ground restrictions, remote sensing technology has become an important means of the relief and the assessment. With the improvement of image resolution, traditional pixel-based classification method has been far from requirements. Object-oriented classification can effectively overcome noise problems such as salt and pepper etc., so it will improve the classification accuracy. The accuracy of damaged road information extraction plays an important influence on the whole work result, therefore, how to improve the accuracy has an important practical value and theoretical significance.Based on the Demonstration of High Resolution Images Disaster Monitoring and Evaluation Information Service Application Projects, specifically, taken Ludian earthquake information extraction for an example, this paper uses object-oriented classification techniques to explore how to quickly extract damaged road information after earthquake. Firstly, this paper analyzed the image hierarchical network, scale selection, segmentation parameter optimization, etc., then verified and analyzed the scale effect through experiments, and proposed the best training segmentation-scale methods to establish rules about different resolution and types of road. Secondly, combining the two methods of threshold and fuzzy, this paper put on the optimization device about resolution and the entity roads, introduced a visualization realization rules to achieve effective combination of segmentation and classification. Finally, damaged road types would be quick interpreted by damaged road model. The paper verified the stability and accuracy of this method by using results of visual interpretation and confusion matrix theory.The experiment results proved that the object-oriented classification has a better accuracy which can provide strong support for the post-earthquake relief, fully embodies the merits and feasibility of this technology.