Local Perception-Based Intelligent Building Outline Aggregation Approach with Back Propagation Neural Network

摘要

With the analysis of the characteristics of back propagation neural network (BPNN) and map generalization of building outlines in both urban and suburban areas, a new approach that is based on local perception of map contexts for building aggregation has been studied. The method called the local perception-based intelligent building outline aggregation approach with BPNN technique consisted of two BPNNs. \(\text {BPNN}_{1}\) was developed to generate initial aggregated building outlines. A circular detector coupled with a set of mapping rules was designed to detect buildings from raster maps. Once trained, \(\text {BPNN}_{1}\) produced initial aggregated building outlines. Due to the existence of unwanted nodes forming small steps along the outlines, \(\text {BPNN}_{2}\) was created to remove the unwanted ones. Here, a square detector was designed and a set of refining rules formulated. Together, \(\text {BPNN}_{1}\) and \(\text {BPNN}_{2}\) intelligently delineated individual buildings or a group of buildings. The performance of the approach has been assessed and the generalized results were cartographically satisfactory.