Development of a new technique for road extraction and pavement surface condition mapping at primary level using worldview-2 satellite imagery

Road networks and their conditions hold fundamental meaning in development. Given the importance of these factors, accurate and comprehensive information on the condition of road infrastructures is necessary for effective pavement management, social network, and global economy. The main objective of this study is to develop a new spectral index for road detection and new technique for mapping road conditions. Field spectral data were collected using field spectrometer to determine the best bands on WorldView-2 (WV-2). The bands were selected based on significant wavelengths from visible and near-infrared to develop a new spectral index for extracting road networks, namely, road extraction index (REI). The accuracy of in the two classes of roads, namely, roads and non-roads, were 88% and 86%, respectively. These roads were extracted using REI from the two selected areas. Nevertheless, the proposed method has limitation in extracting several asphalt roads covered by trees or shadow. A novel spectral index that can detect shadows, namely, shadow detection index (SDI), was developed to improve REI. Two road extraction test areas are conducted on WV-2 to evaluate the propose method. SDI results were used to detect and remove shadow from WV-2 images and improve the accuracy of road extraction. Results show that the accuracy of REI increased up to 5% in the main and validation areas from 90.75% and 91.80% to 95.15% and 95.10%. Another main objective of this research is to extract road conditions using object-based image analysis (OBIA) and feature selection technique. Three different methods for feature selection, such as chi-square (CHI), random forest (RF), and support vector machine (SVM), were used. Results show that CHI, RF, and SVM respectively extracted 19, 17, and 20 out of 54 attributes of asphalt road condition using WV-2 satellite data. Results show that the accuracy of RF (71.06%) and SVM (75.75%) are less than the accuracy of CHI method at 83.79% in two different classes (i.e., High deterioration detected and Low deterioration detected). Finally, the CHI method is used to extract road conditions in seven study areas because this method, which has an accuracy of up to 91%, can potentially detect road conditions in different study areas using high-resolution images such as WV-2 in the primary level.

Preview Text FK 2017 30 - IR.pdf Download (2MB) | Preview

Actions (login required)

View Item