Research And Design Of Reading System Of Pointer Meter Based On Image

Pointer meters are widely used in daily life and industrial fields.How to efficiently and accurately automatically read the numbers indicated by the pointer meters becomes very important.The meter reading is mainly divided into four aspects: meter detection,dial correction,pointer positioning,and display calculation.At present,due to the interference of illumination changes,pointer shadows,and irrelevant information on the image of the meter,the positioning of the meter pointer is still a difficult point in the current algorithm research.This paper aims at this difficulty through image processing and deep learning.Related technologies have been studied,and the specific research work is as follows:1.Aiming at the problem of poor pointer positioning effect caused by interference factors such as illumination changes and pointer shadows,a pointer meter reading method based on single-parameter homomorphic filtering and global contrast enhancement is proposed.A single-parameter exponential homomorphic filter is designed to enhance the adaptability of the meter image to light changes;the global contrast image enhancement algorithm is used to highlight the dial information area;the pointer area is extracted through threshold segmentation,connected domain processing,and morphological processing,And then realize the pointer centerline positioning;finally determine the pointer deflection angle through the direction judgment mechanism.Simulation experiments show that the proposed algorithm can accurately extract the pointer area under the conditions of different illumination changes and the presence of pointer shadows,and the average absolute error of reading is 0.011,the average quoted error is 0.20%,the average time for each picture is 0.243 s.2.Aiming at the problem of YOLOv3 in the pointer meter detection process,which has complex model parameters and large calculations,MobileNetv3 is used to decompose the standard convolution into deep convolution and dot convolution to reduce the number of parameters.MobileNetv3 is used as the backbone network of YOLOv3.The lightweight YOLOv3-MobileNetv3 network detects the target image of the meter;for Unet++ in the process of extracting the pointer,the imbalance of the instrument training sample causes the prediction result to be completely black.The Dice coefficient loss function is used to intersect the prediction result and the real label.The characteristic of maximum parallel ratio is used as the loss function of Unet++ to solve the problem that the prediction result is completely black;the direction judgment mechanism is used to determine the deflection angle of the pointer straight line,and the angle method is used to complete the calculation of the meter display.The simulation experiment shows that the average time of meter detection is 0.036 s,the pointer extraction accuracy is 95.47%,which has been significantly improved,and the average absolute error of reading is 0.008,the average quoted error is 0.13%.3.In order to apply the reading algorithm of the indication of the pointer meter,based on the reading method of the indication of the pointer meter proposed in this paper,an automatic reading system of the indication of the pointer meter is designed and implemented.The system is mainly composed of four modules: visual reading,image display,log recording,and one-key reading module.The system uses the detection algorithm and segmentation algorithm proposed in Chapter 4 to detect and extract the meter and pointer respectively.Through the friendly visual interaction module,not only can the processing process of reading the reading be viewed in real time,it is also convenient for users to operate and manage,and simulate the experimental results.It shows that the average time consumed by the system in this chapter is0.07 s,which can meet the actual application requirements of reading the meter.