A reading method of two-pointer instrument based on deep learning

Abstract

The invention discloses a two-pointer instrument reading method based on deep learning, which belongs to the technical field of image recognition, and specifically includes the following steps: performing instrument detection on a target image, and using a trained yolo v3 target detection model to convert the instrument frame from the original image Segmentation; Hough circle detection is performed on the instrument image, and ellipse connected domain screening is performed at the same time, and the circular dial of the instrument is obtained by combining the two methods, and the dial is corrected; the corrected dial image is denoised, grayscaled, and image enhanced , binarization, morphological closed operation processing; according to the initial information of the pointer on the dial image, coarsely filter the connected domain of the pointer, and make a circumscribing rectangle for the connected domain of the pointer. This deep learning-based two-pointer meter reading method has good accuracy and versatility in pointer area extraction and scale recognition under complex backgrounds, and can meet the actual application requirements of substations.

Description

technical field [0001] The invention relates to the technical field of image recognition, in particular to a method for reading a two-pointer instrument based on deep learning. Background technique [0002] In substation inspection and identification, there are many kinds of pointer instruments, which occupy a large proportion in the identification task; the instrument automatic identification technology based on digital image processing technology can improve the efficiency of instrument detection and reduce human error. At this stage, there are three common detection methods for pointer instruments: the step size method, the circle gray scale detection method and the Hough transform method. At present, there are many research results of single-pointer recognition algorithms, most of which are based on manual calibration of key points, combined with traditional image processing methods, the robustness is poor, and instrument recognition is susceptible to factors such as var...

AI Technical Summary

Problems solved by technology

At present, there are many research results of single-pointer recognition algorithms, most of which are based on manual calibration of key points, combined with traditional image processing methods, the robustness is poor, and instrument recognition is susceptible to factors such as various lighting, specular reflection of the instrument glass panel, and dust. Impact

In addition, in the instrument detection, the existing technology, one is through the method of feature matching, which is easily affected by the environment and complex background, and the speed of instrument matching is slow; the second is the detection method using deep learning, but there are often a few errors. In the case of double-pointer instrument recognition, most of the existing technologies use feature matching or template matching to determine the positions of scales and pointers on the dial, but this method has poor versatility and recognition efficiency. Accuracy is greatly affected by image noise

Images