Research On Industrial Control Logic Diagram Recognition Technology Based On Deep Learning

Industrial control logic SAMA(Scientific Apparatus Makers Association)diagram is often used in the design of industrial process control system,when the control system design is completed,it needs to be imported into the special simulation modeling analysis platform for calculation and verification.As the drawings are drawn by Visio software,they cannot be directly imported into the simulation modeling and analysis platform,and it is inefficient and time-consuming to verify the industrial control logic SAMA diagram manually.Logic diagram.In the process of PDF file identification and conversion,the identification of logical elements of I?C SAMA diagrams is the key to the identification and conversion process,and the text attributes and line segment connection relationships in the drawings need to be attached to the existence of logical elements to have practical significance.The main research contents of this paper:(1)According to the characteristics of PDF logic diagram files provided by the design unit and the technical indicators of the conversion and identification system,the overall scheme of the conversion and identification system for industrial control logic diagrams is designed to meet the needs of the industrial control logic diagram conversion and identification system is divided into four modules: PDF file conversion module,logic diagram element identification module,information integration output module,XML file inspection module,so as to output the simulation platform in line with the format specification of XML file;analyze the shortcomings of the existing logic diagram element identification and detection methods.(2)Image pre-processing and sample expansion for the existing industrial control logic diagram data set.In this paper,we select the mask technology to extract the element information part of the industrial control logic diagram,so as to remove the edge grid of the drawing template,and use the threshold segmentation method to remove the irrelevant line information to improve the image quality;we adopt the traditional image method to expand the industrial control logic diagram data set,and carry out the element annotation on the data set after the image pre-processing.(3)Based on the fact that the original scheme uses NCC-based template matching to identify logical elements is not satisfactory,the Faster RCNN algorithm is introduced to identify logical elements,and its original network is improved to make it more advantageous for the recognition of industrial control logical SAMA diagram elements.In this paper,by replacing the VGG16 network in the original algorithm with Res Net101 network,introducing the residual value module to ensure the detailed features of the deep network,and then analyzing the data set of industrial control logic diagram by K-Means clustering algorithm to improve the scale and proportion of anchor frames in the region generation network,improving the non-extreme value suppression algorithm in the region generation network to avoid the situation of mistakenly deleting candidate frames due to the close distance of logic diagram elements,and introducing the association of similar industrial control logic diagram elements for the problem of easy false recognition.To improve the recognition rate of similar logical elements by introducing the method of associating the inherent properties of the text of logical elements.(4)The framework of Tensorflow is used to build a test platform based on Windows system,and the efficiency of the improved method is compared with the original one by using the control variable experiment method.The results show that the average accuracy of the improved scheme can reach 96.3% for the recognition of industrial control logic diagram elements,which greatly improves the efficiency of industrial control logic diagram element detection.