Transformer Nameplate Text Detection And Recognition Based On Deep Neural Network

In recent years,due to the rapid development of intelligent industry and Internet of Things technology,it is critical to promote the transformation of digital economy industry,strengthen the construction of measurement capacity,and consolidate the measurement foundation for high-quality development.The intelligent power electronic equipment nameplate detection and recognition system is one of the core cornerstones of the intelligent ubiquitous IoT measurement system.This article focuses on the common power electronic equipment-transformer nameplate text,and conducts research on the transformer nameplate text detection and recognition system based on deep neural networks.This system can actively detect the position of the nameplate text on the transformer image and recognize the text,providing important information such as the production date,production location,and equipment parameters of the transformer for the equipment database.The main work content and research results of this article include the following points:To address the issue of text detection and recognition in transformer images,a transformer text detection network DBU based on deep neural networks and a transformer text recognition network "mv3_fc_ctc" based on fully connected feature encoding are proposed to complete the task of transformer text detection and recognition in this paper.The DBU network trained the model on the text detection common dataset ic15 and tested it,with an average detection accuracy of 85%.The "mv3_fc_ctc" network trained the model on the common text recognition dataset ic17wlt and tested it on the self-made transformer dataset,with an average recognition accuracy of 75.54%.The two networks ensure network performance while also meeting engineering requirements for inference speed and other indicators.FPGA hardware accelerated deployment:through further tailoring the internal structure of the above text detection and text recognition network to adapt to the model deployment of the FPGA platform,generate the FPGA configuration file corresponding to the network model,and complete the acceleration of the transformer text detection and text recognition network model on the FPGA platform.Aiming at the problem of hardware acceleration of deep learning network,the instrument transformer text detection network and text recognition network are built through Paddle platform to adapt to the model deployment of FPGA platform,generate the FPGA configuration file corresponding to the network model,and complete the acceleration of instrument transformer text detection and text recognition network model on the FPGA platform.Through the above contents,this paper constructs a text detection and recognition system of transformer based on a deep neural network,improves the accuracy of text detection and recognition network on the transformer nameplate text,deploys the network model to the FPGA hardware acceleration platform,and designs the corresponding upper computer software.The above results show that the transformer text detection and recognition system built in this paper can detect and recognize the transformer nameplate text in the real environment,and the network model can display the recognition results in the upper computer software after FPGA hardware acceleration.