Transforming Remote Sensing (RS) classification result from the raster to vector format (R2V) is a common task in Geographic Information Systems (GIS) and RS image processing. R2V acts as a bridge connecting GIS and RS data integration, and is an important module in many commercial software packages such as ENVI and ArcGIS. While considering inconvenience and inefficiency existed in current R2V algorithm, it still has some room to improve. In this paper some technologies and skills are addressed to improve R2V, including sub-image dynamical separation, fast edge tracing, segment combination and partial topology construction. A new method of two-Arm chain edge tracing is introduced. The improved algorithm has so me advantages: It can transform all types of RS classification only once, and build complete topology relationship; The shared edge between two polygons is recorded only once, the diagonal pixels with same attribution are connected automatically; It is scalable while processing large dimension image,it runs fast and enjoys a significant advantage in processing large RS images; It is convenient to edit and modify the vectorised map because of its complete topology information. Based on case study, the preliminary results show its some advantages over Envi and ArcGIS.