Cell Recognition Of Drawings Based On Neural Network Size Label

With the development of computer science and its application in the engineering fields, there is a revolution in the method of reserving and reusing engineering drawings. To transform the paper-based engineering drawings into the electronic files, which can be viewed / edited by the computer and transmitted via network systems, is a fundamental requirement of modern engineering technology. Many researchers worldwide conduct research projects on this topic and get great success on the vectorization of raster files. One of the bottlenecks in the further research on topological verification and 2D/3D geometric model reconstruction is the recognition of dimension-sets and information marks in the engineering drawings.This thesis provides a new methodology to resolve this problem -- to encipher the dimension-sets based on the corresponding topological structures and to recognize the individual types with Artificial Neural Networks (ANN). First, according to the rules defined by Machine Drawing (GB.4458.4-84), the thesis discusses the classification of dimension-sets,and then analyzes the fundamental topologic elements of dimension-sets and works out the way to encipher these basic structures. Combining the codes of these fundamental topological structures, there are complete code series of dimension-sets and manufacturing information marks. Finally, the thesis provides an ANN structure to encipher the marks and another ANN structure to recognize the corresponding codes. The thesis also provides solutions to dimension-sets reconstruction, such as arrow matching and comprehended dimension-mark decomposing.