Pointer instrument reading identification method based on neural network

Abstract

The invention discloses a pointer instrument reading identification method based on a neural network. The pointer instrument reading identification method comprises the following steps: 1) constructing and preprocessing a dial plate positioning data set; 2) designing a matched neural network model according to the characteristics of the recognition object and the application scene; 3) training theneural network model loading training parameters to obtain a dial positioning model; 4) inputting a to-be-identified image into the trained dial plate positioning model to obtain the position and category of the dial plate in the image, outputting and constructing a dial plate information extraction data set by the dial plate positioning model, performing preprocessing, loading training parameters to the neural network model, performing training on the dial plate information extraction data set, and obtaining a dial plate information extraction model after the training is completed; (5) inputting the dial plate positioning model output into the trained dial plate information extraction models to obtain dial plate range information, (6) extracting pointers and dial plate center position information in the dial plate positioning model output, and (7) calculating instrument readings. The pointer instrument reading identification method can obtain instrument reading with high precision onthe premise of guaranteeing timeliness.

Description

technical field [0001] The invention relates to the technical field of pattern recognition and artificial intelligence, in particular to a neural network-based pointer meter reading recognition method. Background technique [0002] Pointer instruments are widely used in many fields such as industrial production, social life and environmental monitoring, providing many conveniences for production work. At present, manual reading is a widely used reading method, but manual reading has defects such as low efficiency, large error, and poor stability, especially in some places that are not suitable for personnel to enter, such as high-voltage environments in substations, manual reading is no longer feasible. sex. In recent years, with the rapid development of machine vision and artificial intelligence technology, the automatic recognition technology of instrument readings based on image equipment has become possible. The use of automatic recognition technology can effectively av...

AI Technical Summary

Problems solved by technology

Methods based on traditional image processing mainly include subtraction method, Hough transform detection method and template matching method, etc., but these methods have some defects. For example, subtraction method not only needs to obtain the zero scale image of the instrument to be recognized in advance, but also needs to There is no change in the recognition background area. The Hough transform detection method can only detect dials and hands with relatively standard shapes. Template matching needs to obtain a template image of a single size, and a high contrast between the template image and the background is required. These traditional image processing methods The robustness is easily affected by factors such as illumination, shooting angle and image quality, and the application scenarios are limited

The method based on deep learning can locate and identify various types of instruments in different scenarios after training with a large number of data sets, which has high robustness, low requirements for application scenarios, and strong generalizability. The deep learning method requires a large amount of data sets and time for training, and due to its complex structure, it cannot guarantee real-time performance under limited hardware conditions

Images