A Method Of Detecting Defects Of Electric Meter LCD Screen Based On LSD And CNN

With the widespread use of electric energy meters in the State Grid,its quality needs to reach a higher standard.In order to ensure a quality rate of electric energy meters at the factory,it is necessary to test their quality.At present,the common quality measurement of electric energy meter is mainly divided into two modes: manual and machine vision.Among them,manual quality inspection has some problems of low efficiency,high cost and low accuracy with increasing working time.The visual detection algorithm is mostly a traditional image processing algorithm,and a detection accuracy rate needs to be further improved,and as business increases,the target type of the algorithm detection needs to be increased.In order to solve the above problems,this paper combines deep learning technology,image processing technology,computer technology and database technology to design an automatic detection system for electric energy meter defects based on convolution neural network,which can quickly and accurately realize the quality detection of electric energy meter.Compared with the traditional image processing algorithm,it has higher accuracy and can accurately judge the complicated situation such as foreign object occlusion and black screen,and has the functions of LED light detection and electricity power recognition.The specific work of this paper is as follows:(1)An electric energy meter defect detection system based on convolution neural network is designed,which can realize state detection,character defect detection,electricity power value OCR recognition,barcode recognition and LED trip light detection of the energy meter LCD screen.Firstly,according to the requirements of the project,the system is divided into two modules: target detection and target positioning,and the character annotation configuration software,automatic data collection script,network training script,database and detection algorithm software call interface are designed.(2)In order to achieve target detection in the image,the target needs to be segmented to determine the area to be detected.The target in the positioning module in this paper mainly includes LCD screen,each character in LCD screen,and a LED trip light.In the LCD positioning module,the image is first pre-processed by grayscale,LSD line check,tilt correction,Canny edge detection,etc.,and then according to the manufacturer's standard LCD screen template,the corrected edge binary image is positioned accurately by correlation coefficient template matching method.In the character positioning module,the method of reading the character annotation text is adopted.According to the character position feature analysis,the character labeling configuration software is customized and developed.The character coordinate information text can be quickly generated by mouse drawing.After dividing the image of LCD screen of electric energy meter,the character coordinate information text can be read to locate the exact position of each character.In the development of the LED trip light positioning module,the position and shape characteristics were analyzed,and the approximate range was determined according to the relative position of the LCD screen,and then a Hough gradient method is used for accurate positioning.(3)The target detection module mainly includes LCD character defect detection of LCD screen,state detection of LED trip light,and electricity power value recognition.The character defect detection is implemented by convolution neural network.Each character is detected by convolution neural network in turn.According to the recognition result,the character of the location is judged whether to display,occlude and black screen.The bottom power value OCR recognition also uses the convolution neural network to accurately identify the current remaining total power,peak power,peak power,flat power,and valley power.The detection of the LED trip light uses the gray level average method to judge whether the LED lamp is on or not.The experimental results show that the character defect detection,electricity power value OCR recognition,LCD display status detection and LED trip light lighting detection achieved the project requirements.(4)At present,the system has been tested and used in the measurement center of the State Grid Henan Electric Power Company.According to the test results,the system meets the project acceptance criteria,which can achieve fast and accurate intelligent energy meter quality detection and ensure the qualified rate of the products.