神经网络技术在工程图纸识别中的应用

THE APPLICATION OF THE NEURAL NETWORK IN THE ENGINEERING DRAWING RECOGNITION

提出一种应用神经网络技术识别工程图纸扫描图象中图形符号的方法，这种神经网络分类器不仅使识别系统具有自适应性、可拓展性等特点，而且可适应工程图中符号处于粘连、相交、退化等复杂情况下的识别，对目前工程图识别中的最困难问题提出了一个有效的解决方法。

This paper describes the recognition method of the graphics symbols in the engineering drawings by using the neural network technique. This neural network classification technique not only has the advantage of the self-fitness and the opening property,but also has the ability of recognizing the adhered,intersected,and degraded symbols.The effective resolution method is supposed for the most difficult problem in the engineering drawing recognition. We use the mathematical quadrate and the distance functions between the outline and the center point as the feature vectors of the local linked area graphics which serve as the input value of the neural network.These feature vectors have the invariant properties of different size,movement and direction in the same class graphics.The identified strings and the symbols in the engineering drawings are usually adhered to the line or intersected with the graphics due to the paper scanning sampling as well as the personal high density graphics drawing,and some of the engineering symbols are needed to be expressed on the line,so the scanning and ratiocinating recognition algorithm based on known lines can get better results. On the other hand,bacause there are so many symbols in the paper drawings, the neural learning and recognizing method has the characteristics of robust fitness.The neural network classifier has the obvious advantage in which we can create the classified edges for the subsection line segments or quadratic curve segments,and that general classifier can only create the the classified edges of super-plane or super-quadratic curve line.Because the classified edge of the neural network classifier can approach to any shape edge of the graphics in the image,so the accuracy of the engineering drawings classification and recognition could be raised much more.The architecture structure drawings automatic recognition and calculation play an important role in the architecture biding and construction.This research project is consigned by HongKong Yaulee Architecture Company and developed by Multimedia Research Institute,Nanjing University.