Research On Reading Recognition And Remote Meter Reading System Of Word Wheel Water Meter

With the development of computer technology,image recognition technology is gradually being applied in people’s lives.As an important carrier of information,images cover a lot of information,so the wide application of image information recognition technology is an inevitable trend of future development.So far,most of the research on image recognition technology has focused on document recognition,automobile license plate recognition,etc.,but less research on water meter reading recognition.At this stage,the word wheel type mechanical water meter is still widely used in residential life and industrial production,but manual meter reading is time-consuming and costly,and manual meter reading in an industrial environment will have great safety hazards,so for this type of water meter The research of reading recognition technology is particularly necessary.The water meter image is captured by the camera,and the meter reader no longer needs to go to the spot to watch the meter,which ensures safety and reduces manpower consumption.The main tasks completed in this paper are as follows:(1)The image size is adjusted by bilinear interpolation,and the weighted average method with the best experimental effect is used for gray-scale processing.According to the noise mechanism in the image,Gaussian filtering is used to remove the noise from the image,and finally the pre-processed water meter image is obtained.(2)The pre-processed image is further positioned to obtain the water meter character wheel area.Through the effect comparison,the Canny algorithm with better processing effect is selected to detect the edge of the image.Based on the principle of Hough transform,the circle detection method is used to remove the pixels in the non-target area,and on this basis,the straight line detection method is used to correct the image tilt and capture the positioning of the character wheel area.In addition,the arc coordinate calculation method in the classic Hough circle detection algorithm is improved to reduce the calculation time.(3)In order to achieve the purpose of highlighting the characteristics of the characters,through comparison,the Niblack binarization method with the best effect is used to separate the characters and the background.On this basis,the four-neighbor labeling method is used to analyze the connected domain of the image,and then the characters are segmented and classified.According to the projection principle,a curved surface restoration method is proposed to correct the character shape.The template matching recognition method is adopted for different types of characters,and corresponding templates are made to realize character recognition.According to the processing and recognition results,the feasibility of the character image segmentation and recognition method used is verified.(4)Design and implement a remote meter reading system for water meters that automatically takes pictures of water meters,automatically transmits images,automatically recognizes readings,and facilitates data export.The water meter image acquisition device is built with the ESP32-cam camera module,and the captured images are sent by mail through the POP3 protocol.The image cloud processing platform built on the cloud server receives image emails and recognizes them,uses SQLite as the storage database to store meter reading data.The cloud processing platform and the software installed on the local computer use TCP protocol for meter reading data transmission,and the data is displayed,modified and exported with the human-computer interaction interface in the software.(5)According to standard experiment requirements,collect water meter image samples to test the recognition accuracy and recognition speed of the algorithm.In addition,the single variable method is used to test the reading recognition effect of the image in the case of insufficient light supplementation and water droplets and sediment covering the dial.Through the test,the recognition accuracy of the water meter image reading is 93.4%,the recognition accuracy of the whole character and the half character are 99%and 96.4%,respectively,and the algorithm has a certain degree of anti-interference.