Research On Point Symbols Recognition In Color Topographic Map

The point symbol recognition in color topographic map is an important part of map pattern recognition which is the key technology to improve the efficiency of map automatic digitization.It is the focus and difficulty of the current research.To improve the accuracy of point symbol recognition is of great significance to development map automatic digitization.This thesis expounds the research significance and development status of map point symbol recognition,and summarizes the basic theory of color topographic map.It briefly introduces the color segmentation algorithm and linear feature extraction algorithm of color topographic map,and focuses on the map point symbol recognition.In this thesis,a point symbol recognition algorithm based on nonlinear mapping and Generalized Hough Transform and a point symbol recognition algorithm based on deep learning and connected domain features are given,and a point symbol recognition software based on the QGIS open source platform is developed.This thesis presents a point symbol recognition algorithm based on nonlinear mapping and Generalized Hough Transform for the feature of point symbol's gray histogram in color topographic map.Based on the extraction of linear feature and the traditional Generalized Hough Transform,this algorithm introduces the gray histogram feature of the point symbol.It applies the non-linear mapping to the results of the traditional Generalized Hough Transform algorithm and the point symbols.Then,it selects and corrects the initial recognition results to improve the accuracy of the point symbol recognition algorithm by using template matching of the non-linear mapping results.In addition,this thesis presents a point symbol recognition algorithm based on the deep learning and connected domain features,which aims the connected domain features of point symbols in color topographic map.The algorithm introduces the connected domain features of the point symbol in the black layer of the color topographic map.By filtering all the connected domains in the black layers,it obtains the connected domains of all the suspected point symbol area.The external rectangle of the suspected point symbol extracts the corresponding image blocks in the original color topographic map,and then thisalgorithm identifies and selects the suspected point symbol image blocks by using a pre-trained model which is trained by a convolution neural network named Le Net in Caffe.After identifying the image blocks,this algorithm obtain the final recognition result.By introducing the connected domain features and ignoring its category,this algorithm improves the efficiency of the point symbol recognition.According to the research results of point symbol recognition,a point symbol recognition software based on QGIS open source platform is developed.The software is developed in Visual Studio 2010 environment and depends on Open CV library.In this thesis,a point symbol recognition algorithm based on nonlinear mapping and Generalized Hough Transform and a point symbol recognition algorithm based on deep learning and connected domain features are applied to the color topographic map.The point symbol recognition algorithm is fully experimented.And the accuracy of the point symbol recognition in color topographic map is more than 90%.In addition,a series of point symbol recognition functional modules based on the QGIS open source platform are developed and they have been used in actual collection.