Method Of Chart Recognition Based On Converlutional Neural Network YOLOv5 Model

It’s crucial to grasp the latest industry data in Real-time,which can provide an effective reference for the choice of work direction.Among the many data acquisition methods,charts make readers to obtain information intuitively and quickly.Therefore,how to obtain the chart in as little time as possible is a topic worthy of study.The traditional chart recognition technology based on data analysis,time series,digital image processing and other methods has problems such as high professional requirements for the application personnel,the recognition accuracy greatly affected by the preprocessing method,and long image processing time.In recent years,deep learning technology has made outstanding achievements in the field of image processing.In order to overcome the problems in traditional chart recognition technology,this paper proposes a new method for chart recognition based on convolutional neural network YOLOv5 model.This method is suitable for scenes with high demand of real-time and accuracy.The work of this thesis is mainly divided into two parts: sample data generation and performance optimization.(1)Image recognition training data sets are often produced by manual labeling methods,but the time cost of this method is too high.In order to solve this problem,we propose to use PPTX protocol analysis method for document generation to get closer to the real data;the data set uses the document page as the background to achieve the effect of enrich background;use the XML analysis method to obtain the chart position and size information,automatically mark documents.(2)The document chart recognition task is a large target recognition task and requires a high recognition speed.Therefore,this subject chose YOLOv5(You Only Look Once version 5)for chart recognition.YOLOv5 has strong real-time performance and high accuracy of large target detection.It performs well on the data set of this subject.With this method as the core,the paper designs and implements a chart recognition system.Users can upload documents and automatically obtain the charts in the document.