A Method of Electricity Meter LCD Screen Defect Detecting Based on Convolutional Neural Network

In order to quickly and accurately detect the display defects of electric meter LCD screen, this paper proposed an electric meter LCD screen defect detecting method based on convolutional neural network (CNN). First, a horizontal straight line of LCD screen frame is found by LSD line detection method for tilt correction. Second, the LCD area is positioned by normalized correlation matching of the corrected image. Then, the accurate positions of the meter characters are located by the characters position information generated by a template annotation tool. Finally, a CNN is used to perform character defect detecting and OCR recognizing on the segmented meter characters. The experimental results show that the accuracy of the method of positioning and detecting the LCD screen characters is about 99%. At the same time, the CNN has an OCR recognition function, which can accurately identify the meter LCD screen characters.