| The utility pole nameplate is used to mark its properties and number information,and it is an important reference mark for the power department to manage the grid lines.After investigation,the current power grid information collection work is mainly completed by traditional manual mapping.The collection process consumes high human and material resources.With the development and in-depth application of image recognition technology in the power industry,the demand for constructing intelligent power acquisition system is becoming more and more urgent.Replacing traditional surveying with computer vision technology can reduce the manual dependence of data collection.It is also an effective method to improve the efficiency and accuracy of information acquisition in the power grid.However,it is found that the domestic power industry has not yet formed a unified target detection and recognition engine.Especially in the field of grid information collection,there are few relevant research results retrieved.And the algorithm model can not meet the current business needs.Therefore,OCR technology has potential challenge research space in the application of pole information collection.Based on the above problems and requirements,this paper researches and analyzes the text features of the utility pole nameplate image,combines OCR technology to establish a text detection and recognition network model,and builds a prototype system for automatic identification.Finally the automatic detection and recognition of the text in the utility pole nameplate image is realized.The research results of this article are as follows:(1)Data pre-processing of utility pole nameplate images.The project team collects actual images and combines the data enhancement method to enrich the sample features.Manually labeled data is used to adjust network model parameters.Experiments show that the improved data samples can effectively improve the generalization ability of text detection and text recognition models.(2)A text detection model based on improved EAST for utility pole nameplate is researched and implemented.This model uses a non-maximum suppression algorithm to effectively integrate the text features into the full convolution feature learning network.It deepens the convolutional network hierarchy by reducing the feature parameters to improve the network’s feature learning ability,and optimizes the loss function to improve the network’s feature classification ability.Experiments show that the improved EAST algorithm has better detection effect than the original network model.(3)A text recognition model based on CRNN for utility pole nameplate is researched and implemented.The long-short term memory network in this model can more fully learn the sample spatial information features.It also uses connectionist temporal classification to eliminate text cutting steps,thereby completing the text feature extraction task end-to-end.Experiments show that the model has faster recognition efficiency under the premise of ensuring recognition accuracy.It is more suitable for industrial applications.(4)A automatic identification prototype system for utility pole nameplate is designed and implemented.Combined with the actual business scenarios and application requirements,the system designs multiple functional modules to connect the power grid GIS platform,and then completes the front-end and back-end system development.To sum up,through data pre-processing and feature labeling work,this paper establishes a text detection and recognition network model based on EAST and CRNN algorithm.It also designs an end-to-end automatic identification prototype system for utility pole nameplate.The system is expected to be used in the intelligent data collection and management system of State Grid in the future,which provides a feasible way for grid companies to build GIS big data platforms. |
