Research On Feature Detection Technology Of Smart Meter Based On Machine Vision

As the service terminal of the smart grid,smart meters provide enterprises and home users with the measurement,monitoring and feedback of the daily electricity required.Before the smart meters leave the factory and during the repair process,appearance characteristics inspections are required to ensure that the product quality is qualified.At present,the feature detection of smart meters mainly relies on manual operation.The efficiency of manual quality inspection is low,the ability to adapt to the environment and the ability to solve complex problems is poor,and it cannot work continuously for a long time.Therefore,this article combines image processing technology and deep learning technology to design an automatic detection system for the appearance of smart meters.The key technologies used in feature detection such as shape template matching algorithm,improved Canny edge detection algorithm,and average gray value of regional pixels are designed.The algorithm,MLP neural network classifier recognition,convolutional neural network algorithm,etc.have been studied in depth,and the main research contents are as follows:(1)According to the smart meter feature detection requirements,the overall architecture of the smart meter feature automatic detection system is designed;the image acquisition system is built,the light source lighting method,the model of the industrial camera and the lens,and the data transmission interface method are determined,and the connection with the upper level is achieved.The communication control of the machine;the construction of the experimental platform for the automatic detection of the characteristics of the smart meter has been completed.(2)In order to realize the detection of multiple basic features of smart meters,different feature detection algorithms are studied.Aiming at the problem of incomplete information edge contour extraction in the process of parameter information detection,an improved algorithm is proposed to improve the adaptive threshold selection of Canny edge detection algorithm through Otsu automatic threshold segmentation method;for the detection of LED indicator status,an improved algorithm is proposed.An LED indicator detection algorithm based on the average gray value of regional pixels.The detection is completed by analyzing the difference of the pixel gray value in different states of the indicator.Experimental results show that both algorithms achieve accurate and efficient detection of corresponding features.(3)In the battery recognition process,when the LCD screen is backlit or the shooting light is dark,the battery information recognition is not clear,and there is a problem of misjudgment of characters,a battery recognition algorithm based on MLP neural network classifier is proposed.By collecting a large number of characters that are prone to misjudgment as data samples,a suitable MLP classifier is created to recognize and classify special characters to complete accurate identification of electricity.Experimental results show that the algorithm effectively improves the accuracy and efficiency of power identification.(4)In order to realize the detection of the lack of information on the LCD screen of the smart meter,a detection algorithm based on image processing technology is first proposed.Through the module division of the LCD screen,the combination of OCR character recognition technology and character threshold segmentation algorithm completes the information missing detection.Experimental results show that the overall detection accuracy of the algorithm is low,and the detection efficiency is poor;in order to solve the problems of the above methods,a missing detection algorithm based on deep learning technology is proposed.Based on the Le Net-5 network architecture,the network parameters and algorithms are optimized,the Dropout layer is added to avoid over-fitting,and the curve of accuracy and loss rate is obtained through network training.Experimental results show that the model significantly improves the detection accuracy and efficiency of the lack of information on the LCD screen,enhances the generalization ability of the model,and meets the feature detection requirements.(5)According to the requirements of smart meter feature detection,the system software development platform adopts C# and Halcon mixed programming methods to develop a smart meter feature detection system,which embodies the related algorithm of feature detection,and realizes the statistics of human-computer interaction and detection results.Smart meter parameter information,LED indicator light,power level,lack of display information and other features can be tested.It has achieved good results in a large number of experiments and applications in actual production,and has promotion value.