Automatic road identification and labeling in Landsat 4 TM images

A conceptually parallel road detection method in Landsat 4 TM images is proposed. The goal is to detect roads at three different levels: major roads, local roads, and minor roads. This road network information is used for the evaluation of detected potential oil/gas pads, since these pads seldom occur on major roads but are often located at the end of minor access roads.

The proposed method is composed of two phases: low-level road detection and high-level road labeling. Experimental results from several images show that the proposed method can detect roads reasonably well in the low-level phase and is useful in pad evaluation. Future research includes combining multi-band information in road detection and determining thresholds in a more systematic way.