The application of artificial neural networks for object identification on topographic maps

The process of manual vectorization of objects present on raster maps is time-consuming but necessary due to the usefulness of vector data in GIS analyses. For this reason, the authors attempted to shorten this process by identifying these objects automatically. Artificial neural networks (ANN) were employed for this purpose, with input data consisting of classified pixel color values that make up the raster map. Based on this data, the network predicted four classes of objects: trees, single homesteads, secondary roads, and rivers. The best-tested variants allowed for the detection of over 80% of objects while keeping false detections (overinterpretation) below 10%. The proposed methodology significantly automates the process of object detection on maps, which is essential for expediting tasks related to their vectorization.