Research and Application of Object Recognition Algorithm in Architectural Engineering Drawing Based on Deep Learning

Aiming at the current situation of mainly relying on manual identification, a method of computer automatic identification of electrical components in architectural and installation engineering drawings is proposed. Based on the integration of Structural Semantics, starting from the leading building components, in the reduced recognition space, combined with the characteristics of the current recognition level, the feature-based building symbol recognition method, symbol based grid recognition method, contour based node component recognition method, segment component recognition method based on association semantics, geometry based composite component recognition method and inheritance based information transmission method are adopted respectively, The subsequent recognition is completed under the constraints of the model, which can effectively eliminate the drawing noise and improve the recognition rate with a small amount of calculation. The proposed method for obtaining engineering quantity information is essentially based on the vectorization of construction drawings, identifying various graphic objects, classifying them by rotation, displacement and scale invariance, and then identifying and labeling semantics, associating planes, elevations and scales. Section diagram, so as to realize the acquisition of three-dimensional properties of architectural drawings and information of steel bars. This paper discusses the research on the automatic identification of non-orthogonal architectural drawings and the algorithm of construction quantities, and presents the algorithm formula for calculating construction quantities and the principle and block diagram of the automatic identification of architectural drawings.