R-CNN-based intelligent electric meter numerical value identification method

Abstract

The invention discloses an R-CNN-based intelligent electric meter numerical value identification method. The method comprises the steps of offline training: collecting a large number of ammeter pictures shot by a camera, preprocessing ammeter picture samples by using an image preprocessing method, and adding labels to the samples to facilitate network learning, the labels being set as reading values of a dial plate to obtain a network for identifying the dial plate; extracting a position area of an ammeter dial, and generating a series of candidate areas through a selective search algorithm; inputting a zoomed picture of each candidate area into an R-CNN network for feature extraction, identifying whether the candidate area picture is a target area, namely an ammeter dial area, through anSVM network, identifying dial information through a multilayer R-CNN network and a full connection network, and outputting a finally identified ammeter reading. The recognition method provided by theinvention can quickly and accurately recognize the reading of the electric meter, is very small in limitation on the electric meter image, and is good in practicality.

Description

technical field [0001] The invention belongs to the technical field of intelligent reading of electric meter values, and in particular relates to an R-CNN-based intelligent electric meter numerical recognition method. Background technique [0002] As a measuring instrument, the electric meter is widely used in various fields in social production and life, and plays a pivotal role. At present, the reading of the electric meter is mainly completed by manual interpretation. However, this method is greatly affected by human factors, and has poor reliability and low efficiency. [0003] In recent years, with the continuous development and improvement of digital image processing technology, the identification of electric meters has also made great progress. Aiming at the automatic interpretation of electric meters, researchers have conducted researches using machine vision technology, which has expanded the applicable scope of the automatic identification system and can effective...

AI Technical Summary

Problems solved by technology

However, traditional network learning methods based on convolutional neural networks (Convolutional Neural Network, CNN) still have problems such as slow detection speed and low recognition accuracy in electric meter recognition.

On the one hand, the traditional CNN cannot accurately identify the location of the target, that is, the area where the meter dial is located in the meter image, which requires manual preprocessing of the image before smart meter recognition

On the other hand, in the traditional CNN-based electric meter recognition method, the network model training function is single, and the training takes a long time, making the recognition efficiency low

Moreover, the actual test environment is complex, for example: there are various forms of dials, uneven lighting, reflections from the glass on the surface of the meter, the optical axis of the camera is not perpendicular to the plane of the dial, and the dial is tilted, etc. These factors will affect the meter. Image recognition is difficult, and most recognition methods will produce large errors or even fail

Images