A Method Of Detecting Defects Of Smart Meter LCD Screen Based On CTPN

The appearance defects of smart meters are an important part of their quality inspection.At present,the common detection methods for appearance defects of smart meters mainly include manual and machine vision.Manual quality inspection not only takes a long time and high cost,but also has low detection efficiency and high rates of missed detection,so,it does not meet industrial inspection requirements well.A common method for machine vision is template matching method.However,due to the large number of domestic smart meter manufacturers,there are some differences in appearance and size of energy meters produced by different manufacturers.Therefore,this method often fails to match template.At this time,factory needs to stop working and remake the template,which is very time-consuming.In order to solve the above problems,this paper designs a deep learning-based energy meter display defect detection system.This system can accurately and quickly achieve the quality inspection of the smart meter.Compared with the traditional template matching method,this paper no need to make a template of LCD screen,and it can achieve a very good missed detection effect.It also has the function of defect detection of appearance nameplate information and defect detection of LED trip lights.The content of this article is as follows:(1)In order to improve the detection accuracy of the system,an image preprocessing step is performed on the detection image,including image graying,LSD straight line detection,and tilt correction.(2)In the LCD character leakage detection module,according to character characteristics of smart meter,the character area is divided into two types: ordinary character area and digital tube character area.Firstly,use CTPN method to detect all text in the LCD screen,and then Ordinary characters are recognized by CRNN,and connected-domain detection is performed on the digital tube area.Finally,it is determined whether the LCD screen characters are missing or not based on the overall detection results of the two areas.(3)In the appearance nameplate module,the template matching method is used to determine whether nameplate information is displayed normally by comparing the similarity between test picture and template picture information.This paper designs a character labeling software,which uses the software to label the nameplate information area to determine its coordinates.The software can be used to quickly make templates and set detection thresholds.(4)In the LED trip lamp module,direct positioning of the LED trip lamp was abandoned,and the template matching method was used to locate the position of the two characters of"??",and then the coordinates of the trip lamp were determined based on its relative coordinates with the trip lamp.Finally,the trip light is accurately detected.System to test the smart meter LCD without making templates,can unite with different manufacturers,different sizes of watt-hour meter for testing,and pictures of some of the poor quality of smart meter(such as shadow images)there is still a good detection effect,for digital tube area,this method can effectively detect the defect images,and have the picture of the scratches;For the detection of external nameplate information,this method can calibrate the detection items simply and conveniently.For the detection of LED trip light,this method can accurately locate the trip light area and realize the subsequent detection.At present,the system has been tested and used by Guangdong Electric Power Company.The detection accuracy rate of LCD screen has reached 92.7%,the detection accuracy rate of appearance nameplate has reached 98%,and the detection accuracy rate of LED trip lights has reached 96.2%.The appearance of the display shows defects are detected quickly and accurately to ensure the factory pass rate of the energy meter.