Intelligent Recognition Technology Of Circuit Drawings Based On Convolutional Neural Network

With the wide application of mapping software,power grid enterprises have gradually accumulated a large number of circuit engineering drawings.But the quality issues such as consistency,accuracy and timeliness of data have become one of the main contradictions widely faced by current power grid enterprises.How to realize the digitization and informatization of existing power drawings,and the fast information extraction and retrieval query of existing drawings has become an urgent problem to be solved.In order to solve these problems,this thesis analyzes the imaging characteristics and fixation laws of electric power drawings in detail and summarizes the influencing factors of drawing recognition.On this basis,a detection and recognition framework based on deep learning is used to accurate recognition of drawings.The primary focus is outlined as follows:(1)A dedicated dataset for circuit drawing recognition is established.In response to the problem that circuit drawing data is scarce and the drawing recognition task is difficult to carry out learning,we firstly collect real data from the plant stations,secondly filter the data to ensure the richness of the data,and then adopt the manual annotation means to annotate the data to complete the construction of a high-quality circuit drawing dataset.These are the foundation for the subsequent development of text recognition and element recognition tasks.(2)Deep learning optical character recognition(OCR)technology based on FCOS algorithm is proposed.The classical anchor-free algorithm FCOS for target detection is used to recognize the text of the power drawings after data pre-processing and augmentation.It can accurately determine the category and location information of the text in the power drawings and automatically recognize the text content according to the corresponding location text information.The text content is automatically recognized according to the corresponding position text information.(3)A circuit drawing element recognition technology based on YOLOv5 algorithm is presented.Combined with the previous work,the text label data in the power drawings are extracted out using a pre-trained OCR model,so that the drawings retain only the equipment and circuit information.Meanwhile,the YOLOv5 model is improved to retain the characteristics of YOLOv5 recognition model such as fast speed and good effect.Aiming at the problem that the size of power drawings is inconsistent and the unified reduction of drawings will lead to the failure of some equipment identification,a sliding window segmentation scheme is adopted to divide a drawing into multiple pictures for training,and then the results are combined.The experimental results show that the proposed algorithm can effectively improve the accuracy of equipment identification and realize the equipment specialization detection of small and medium-sized targets in power drawings.(4)Based on the aforementioned work,the circuit drawing recognition system is implemented.The algorithm modules are programmed and integrated,and a set of identification process architecture that meets the actual application of the power station is designed to complete the construction and implementation of the entire system.After the analysis of experimental results,it can be seen that the algorithm and the designed system can effectively meet the actual working needs of the power station and achieve the expected goals,so as to improve the digital efficiency of drawings in the power system.