Study On Automatic Recognition Technology Of Representative Notation In Power Engineering Drawing

With the incessant expansion and application of Geographic Information System, the problem of data quality givs rise to more and more atention. At the present, scanning cartographic digitization was regarded as a kind of popular data gathering measure of G1S; its influence to G1S data quality is also attached importance. However,the automation recognition technology of the engineering drawing is considered as key technology by international people is a difficult problem. In this thesis the common electric notation is selected as research object, the job of classifying and recognition is completed after extracting electric notation invariant moments. Accordingly pels tracing and function approximation are avoided ,and vectorization process is simplified effectively.At first ,every arithmetic of image transformation and image filter is researched and its disposal image is analyzed .According to the experiment results ,the appropriate pretreatment method that should be choosed by combining with image itself is present.DWT transformation is applied to compress and filter image .Secondly, the existing segmetation methods are classified and summarized, the edge detection methods are researched, especially the threshold image segmentation methods are analyzed. And self-adapt threshold image segmentation is advanced, comparing to Histogram and OSTU, this method has excellent segmentation effect to the electric drawings .At last, accroding to the charecteristic that HU invariant moments possess RTS invariability to continuum image but not scatter image, the relative betterment is used to make it RTS invariant to scatter image. The invariant moments are extracted and used to neural network training swatch, in this way,the NN model is used to classify and recognise the correlative notation.The results shows that the model possesses perfect performance; the recognization rate is above 90%.Accrording above research results The functions such as image pretreatment image segmentation characteristic extractioan training neural network and image recognition are realized with VB6.0 and Matlab6.5.