| In industrial production,for the safety and stability of equipment operation,the instrument is often used to monitor the operation of equipment.The traditional instrument reading method is manual reading.Which not only has high error,low efficiency,and high labor cost,but also brings great obstacles to manual meter reading when there are dust,radiation,and other factors in the production site.Therefore,it is great significance to study the industrial instruments readings automatically.This thesis mainly focuses on the automatic reading research of the common pointer instruments and digital display instruments in industrial instruments.It systematically researches the algorithms of instrument detection,instrument tilt correction,instrument panel preprocessing,character detection,character recognition,etc.involved in the process of industrial automation reading.Based on summarizing previous studies,this thesis uses digital image processing technology combined with deep learning to propose two pointer instrument reading methods and one digital instrument reading method.The specific content is as follows:1.A pointer instrument reading scheme based on the minimum bounding rectangle is designed.The complex background of the pointer instrument makes it difficult to extract,a scheme of first detection and then correction is adopted.First,YOLOv3 object detection algorithm is used to extract pointer dial from the complex background,then the feature point detection algorithm Simple Pose is used to detect the rigid feature points in the dial,and the frontal view of the dial is obtained through perspective transformation.To solve the problem of interference in the front view of the dial,a series of preprocessing methods such as contrast enhancement,image denoising and morphological processing are used to preprocess the pointer panel.Finally,the minimum bounding rectangle of the pointer is extracted,and the automatic reading of the pointer instrument is completed by constructing the mapping relationship between the Angle and the reading.Experimental results show that the relative error of the proposed scheme on the test set is less than 1.5%,and the algorithm has a high recognition accuracy,which can be applied to the automatic recognition of pointer instrument.2.A pointer instrument reading method based on improved PSPNet is proposed.To solve the problem of too much invalid information interference on pointer panel surface,a scheme based on semantic segmentation is proposed.Firstly,the semantic segmentation algorithm PSPNet is improved in structure,and the semantic segmentation accuracy is improved by integrating ECA and joint pyramid up-sampling.Then,the improved PSPNet is used to segment the pointer instrument,and the experiment shows that the average accuracy of the improved PSPNet is 1.91% higher than that of PSPNet.Finally,the semantic segmentation graph of the pointer dial is expanded into a rectangle graph by coordinate transformation,and the distance method is used to complete the reading of the pointer instrument.The experimental results show that the relative errors on the test set are all less than 2.5%,which has certain practicability.3.A digital instrument identification scheme based on deep learning is designed.The preprocessing steps of the existing digital instrument identification scheme are complicated and depend too much on the precision of character segmentation.A two-stage reading method based on line detection algorithm Pixellink and line recognition algorithm CRNN is adopted.It avoids the complicated processing steps of the traditional method and has a certain anti-interference ability to the complex background.The experimental results show that the accuracy of the proposed scheme is 99.3% in the digital instrument of substation,and it has strong robustness.4.A reading recognition system for industrial instrument is designed and developed.The system includes pointer instrument reading recognition module,digital instrument reading recognition module,has a certain practicability. |
PDF Full Text Request