A Recognition Method of Geared Digital Meter Using Projection Regression

Abstract

The invention relates to the technical field of image processing in electric power meters, and relates to a gear-type digital meter recognition method using projection regression. Including extracting and sharpening the edge of the digital area; performing multi-angle projection on the digital area to obtain multi-angle projection features; using the artificial neural network to train the weight matrix according to the pre-set digital image data set, and obtaining the regression function; extracting the edge of the actual image After sharpening, the multi-angle projection features are extracted and put into the regression function to obtain the number recognition result; this method solves the problem of multiple numbers appearing at the same time, and converts the number recognition problem to the regression position problem, thereby accurately identifying numbers.

Description

technical field [0001] The invention relates to the technical field of image processing in electric power meters, and relates to a gear-type digital meter recognition method that converts the problem of recognition and classification into the problem of regression position to obtain continuous readings, specifically a gear-type digital meter that uses projection regression recognition methods. Background technique [0002] There are many digital instruments in substations. With the prolonged use of instruments and the images taken by observers or robots, it is very easy to have multiple digital displays in one area, making it impossible to judge the numbers; most of the existing methods are to directly classify the digital extraction features. However, due to the limitations of the classification principle, this kind of method cannot obtain accurate instrument readings, which easily leads to misjudgment. Therefore, it is necessary to invent a method that can accurately inte...

AI Technical Summary

Problems solved by technology

[0003] The traditional methods of three-line method, eight-line method, and water drop method analyze and classify the digital structure, and the feature extraction is not enough.

[0004] As a result, the numbers cannot be recognized well, and most of the existing methods cannot solve the ambiguous state; in the existing classification algorithms, because the special rotation of the numbers of the gear instrument is not considered, the recognition basis is not enough, and the recognition results are low.

[0005] In "A Printed Number Recognition Method Based on Intersection Features" proposed by Dai Jing et al., the main steps are divided into two steps: the first step is to scan the digital image after binarization and store it in a matrix to obtain the left zero and right one Key points such as intersection points, the second step is put into the RBF feedforward neural network to obtain the recognition result; this method cannot solve the problem of digital ambiguity, and cannot give accurate recognition results;

[0006] In Chinese patent CN2018102101062, the digital image histogram and digital closed features are used to use the area ratio feature of the digital box and the number, the geometric relationship feature, and the height-to-width ratio of the number and the number box as feature quantities to screen out the squares in the binarized image. The frame and the corresponding digital area; but this method has insufficient feature quantity and cannot solve the problem of accurate interpretation in multiple digital states;

[0007] From the above analysis, it can be seen that most of the digital structure features are not enough, and the structural information is not fully utilized. There are many types of features, complex calculations, and difficult to obtain; and most of them use classification algorithms. Traditional classification methods are difficult to solve digital ambiguity. Credible recognition results

Images