Research And Application Of Dial-Reading Image Recognition Technology

In the current industrial production,a large number of old pointer instruments and digital instruments without data communication interface are still used in industrial production,and the data recording needs to be completed manually.Manual data acquisition has high intensity,low efficiency,large investment and high cost in replacing intelligent instruments.In some special environments,there are still some potential safety hazards.In addition,at this stage,on-site manual data acquisition can not meet the current rapid development of production and scientific research needs.Therefore,using machine vision to recognize the instrument image will be more and more applicable.In this paper,two kinds of instruments commonly used in practical applications are taken as the research objects,and the preprocessing of instrument images,the detection and positioning of target dials and image correction,and the identification methods of instrument readings are studied.Analyze the factors that affect the recognition of instrument image readings,such as complex background,lighting conditions,etc.,and study the image recognition method of dial readings.The research aspects of the article are:(1)Research on Image Preprocessing Methods.By analyzing the obtained low light image,the illuminance enhancement algorithm of low light image based on Retinex net neural network is studied,so that the low light image can be recognized later.By detecting the fuzziness of the original instrument image,the instrument image that cannot be recognized later due to image fuzziness is eliminated.By studying the bilateral filtering algorithm and Retinex algorithm,the influence of noise in the instrument image is removed and the part of interest is highlighted.(2)Research on target detection methods.Through the analysis and research of the traditional algorithm,it is found that the traditional algorithm is vulnerable to the interference of background,illumination and other factors.In order to overcome these interferences,yolov4 tiny’s target detection algorithm is finally used and optimized.This method can accurately detect the instrument panel in complex background without being disturbed by environmental factors.For the acquired instrument image with certain tilt deformation,in order to identify it accurately in the future,the perspective transformation algorithm is used for processing.(3)Research on the Recognition Method of Pointer Meter Readings.By marking and filtering the connected domain of the binary image of the instrument panel image,the scale and pointer region are extracted.Through the processing of these two areas,the scale and pointer straight line are fitted to obtain the information such as the center and radius of the instrument.On this basis,the indication of the instrument is read according to the angle method.Experiments show that the image recognition results are basically consistent with the pointer meter readings,and the recognition accuracy reaches the accuracy of the meter itself,indicating that the recognition algorithm is effective and has practical value.(4)Research on reading recognition method of digital instrument.Through the processing of the obtained binary image of the character region,the position information of each character is obtained.The preliminary work of recognition is completed through the segmentation and normalization of the character.Finally,the recognition is completed by BP neural network.Experiments show that the accuracy of image recognition results is high,which shows that the recognition algorithm is effective and has practical value.