CN-121982690-A - Pointer instrument intelligent reading method and system based on deep learning

CN121982690ACN 121982690 ACN121982690 ACN 121982690ACN-121982690-A

Abstract

The invention discloses a pointer type instrument intelligent reading method and system based on deep learning, which belong to the field of intelligent monitoring of power equipment; acquiring an instrument positioning model based on the image verification set and the instrument positioning data set; the intelligent reading method and system for the pointer type instrument based on the deep learning effectively improves the efficiency and accuracy of pointer reading in the pointer type instrument by acquiring text information based on an instrument positioning data set and a preset semantic recognition model, acquiring a feature map based on the instrument positioning data set, acquiring text classification data based on the text information, the feature map and the preset multi-mode model, acquiring a dial understanding model based on the text classification data and the instrument reading data set, acquiring an original instrument photo, and acquiring pointer type instrument reading information based on the original instrument photo, the instrument positioning model and the dial understanding model.

Inventors

GE JIAFEI
SUN QIZHEN
WANG XIJUN
WANG HAOYU
TANG ZHANWEN
ZHU ZIHE
FAN SIQI
GUO MENGXUAN

Assignees

广东电网有限责任公司广州供电局

Dates

Publication Date: 20260505
Application Date: 20251211

Claims (10)

1. The intelligent pointer instrument reading method based on deep learning is characterized by comprising the following steps of: Acquiring a plurality of pointer instrument images; Dial frame selection and content labeling are carried out based on a plurality of pointer instrument images, and an image verification set, an instrument positioning data set and an instrument reading data set are obtained; Performing target positioning training on a preset target detection model based on the image verification set and the instrument positioning data set to obtain an instrument positioning model; acquiring text information based on the instrument positioning data set and a preset semantic recognition model; Performing visual coding based on the instrument positioning data set to obtain a feature map; Performing feature fusion based on the text information, the feature map and a preset multi-mode model to obtain text classification data; carrying out semantic recognition training on a preset semantic recognition model based on the text classification data and the meter reading data set to obtain a dial understanding model; And acquiring an original instrument photo, and performing instrument identification reading based on the original instrument photo, an instrument positioning model and a dial understanding model to acquire pointer instrument panel reading information.
2. The intelligent reading method of pointer type meter based on deep learning as claimed in claim 1, wherein the dial frame selection and content annotation are performed based on a plurality of pointer type meter images, and the image verification set, the meter positioning data set and the meter reading data set are obtained, comprising: performing definition qualification screening based on a plurality of pointer instrument images to obtain an image verification set; Dial plate frame selection is carried out based on the image verification set and a preset image marking tool, and an instrument positioning data set is obtained; and carrying out content labeling and labeling classification based on the instrument positioning data set, and obtaining an instrument reading data set.
3. The intelligent reading method of pointer instrument based on deep learning as claimed in claim 2, wherein the dial frame selection based on the image verification set and a preset image marking tool, to obtain the instrument positioning data set, comprises: for any one verification image in the image verification set, acquiring meter positioning data based on the verification image so as to acquire a meter positioning data set based on meter positioning data corresponding to all verification images in the image verification set; wherein the obtaining meter positioning data based on the verification image comprises: Acquiring dial annotation frames, frame center coordinates and annotation parameter information based on the verification images and a preset image annotation tool; And carrying out size unification and format normalization based on the dial marking frame, the frame center coordinates and marking parameter information to obtain instrument positioning data.
4. The method for intelligent reading of pointer instrument based on deep learning according to claim 1, wherein the obtaining the original instrument photograph, and the instrument identification reading based on the original instrument photograph, the instrument positioning model and the dial understanding model, the obtaining the reading information of the pointer instrument panel comprises: acquiring an original instrument photo; Acquiring dial area images based on the original instrument photo and an instrument positioning model; acquiring key information of the instrument based on the dial area image and the dial understanding model; And carrying out pointer identification and reading calculation based on the dial area image and the instrument key information, and obtaining pointer instrument panel reading information.
5. The intelligent reading method of pointer instrument based on deep learning as claimed in claim 4, wherein the steps of performing pointer identification and reading calculation based on the dial area image and the key information of the instrument to obtain the reading information of the pointer instrument panel include: Acquiring a plurality of scale digital center points based on the key information of the instrument; carrying out ellipse fitting based on a plurality of scale digital center points and a preset ellipse formula to obtain fitting ellipse parameters; Acquiring an ellipse center point based on the fitting ellipse parameters, and determining a dial center point based on the ellipse center point; Acquiring an elliptical dial plate image based on the elliptical central point, fitting elliptical parameters and the dial plate area image; pointer identification is carried out based on the oval dial plate image, and pointer line data are obtained; And performing reading calculation based on the instrument key information, the pointer line data, the number center points of the scales and the dial center points to obtain pointer type instrument panel reading information.
6. The intelligent reading method of the pointer instrument based on the deep learning as set forth in claim 5, wherein the pointer identification based on the oval dial image, the obtaining of pointer line data, comprises: performing binarization extraction processing based on the oval dial plate image to obtain a plurality of pointer identification foreground points; Constructing a polar coordinate parameter space, and acquiring pointer polar coordinate data based on the polar coordinate parameter space and a plurality of pointer identification foreground points; And carrying out standard linear equation conversion based on the pointer polar coordinate data to obtain pointer line data.
7. The intelligent reading method of pointer instrument based on deep learning as claimed in claim 6, wherein the binarization extraction processing is performed based on the elliptical dial plate image to obtain a plurality of pointer identification foreground points; performing pixel binarization processing based on the elliptical dial plate image to obtain a binarized dial plate image; performing morphological optimization operation based on the binarized dial image to obtain an optimized image; and carrying out binarization extraction processing based on the optimized processing image to obtain a plurality of pointer identification foreground points.
8. The intelligent reading method of the pointer instrument based on the deep learning as set forth in claim 5, wherein the calculating the readings based on the key information of the instrument, the pointer line data, the number center points of the scales and the dial center points to obtain the reading information of the pointer instrument panel comprises: acquiring a relative position relation based on the pointer line data and a dial center point; acquiring a pointer direction based on the relative position relation; acquiring adjacent scale center point pairs based on a plurality of scale digital center points and pointer line data; Acquiring a first relative angle, a second relative angle and a pointer angle based on the adjacent scale center point pairs, dial center points and a preset angle function; Acquiring scale data corresponding to the adjacent scale center point pairs based on the instrument key information; And acquiring pointer instrument panel reading information based on the first relative angle, the second relative angle, the pointer angle and the scale data.
9. The method for intelligent reading of a pointer instrument based on deep learning of claim 8, wherein said obtaining pointer instrument panel reading information based on said first relative angle, said second relative angle, said pointer angle and said scale data comprises: Performing difference calculation based on the first relative angle, the second relative angle and the pointer angle to obtain a pointer angle duty ratio relation; And performing reading calculation based on the pointer angle duty ratio relation and the scale data to obtain pointer instrument panel reading information.
10. A deep learning-based pointer meter intelligent reading system for implementing a deep learning-based pointer meter intelligent reading method as claimed in any one of claims 1 to 9, the system comprising: The data set construction module is used for acquiring a plurality of pointer type instrument images, carrying out dial frame selection and content annotation based on the pointer type instrument images, and acquiring an image verification set, an instrument positioning data set and an instrument reading data set; the model training module is used for carrying out target positioning training on a preset target detection model based on the image verification set and the instrument positioning data set to obtain an instrument positioning model, acquiring text information based on the instrument positioning data set and a preset semantic recognition model, carrying out visual coding based on the instrument positioning data set to obtain a feature map, carrying out feature fusion based on the text information, the feature map and the preset multi-mode model to obtain text classification data, carrying out semantic recognition training on the preset semantic recognition model based on the text classification data and the instrument reading data set to obtain a dial understanding model; And the identification reading module is used for acquiring an original instrument photo, carrying out instrument identification reading based on the original instrument photo, the instrument positioning model and the dial understanding model, and acquiring pointer instrument panel reading information.

Description

Pointer instrument intelligent reading method and system based on deep learning Technical Field The invention relates to the field of intelligent monitoring of power equipment, in particular to an intelligent reading method and system of a pointer instrument based on deep learning. Background In the power inspection process, the running state of equipment needs to be judged according to the information of the instrument in time so as to prevent equipment problems from affecting the safe running of a power system. Although the current newly-built system starts to apply the internet of things instrument so as to facilitate timely detection and response, because a large number of pointer instruments used by existing equipment cannot be networked, manual inspection by inspection personnel is still needed, so that the inspection efficiency is greatly reduced, the response and processing cannot be timely performed, meanwhile, the possibility of mistaken inspection exists in the manual inspection, and the stable operation of the power distribution system is not facilitated. On this basis, since the power distribution system has popular monitoring equipment at present, the monitoring image is utilized to realize the automation of meter reading, which greatly helps the smooth running of the power system and helps to reduce the potential economic loss caused by power failure. However, the current pointer instrument method is limited by a shooting angle method, larger errors can be generated when the vision is not perpendicular to the disk surface, and the partial method adopts segmentation and ellipse fitting to generate an affine transformation matrix to restore the angle, but the method is slower in speed and does not meet the real-time requirement. Therefore, there is a need for a meter reading algorithm that can both correct image lifting accuracy and ensure speed of operation. Disclosure of Invention The invention provides an intelligent reading method and an intelligent reading system for a pointer instrument based on deep learning, which can solve the technical problems that the pointer instrument reading is limited by a shooting angle method, has larger error and slower speed in the prior art, and effectively improves the efficiency and the accuracy of the pointer reading in the pointer instrument. The invention provides an intelligent reading method of a pointer instrument based on deep learning, which comprises the following steps: Acquiring a plurality of pointer instrument images; Dial frame selection and content labeling are carried out based on a plurality of pointer instrument images, and an image verification set, an instrument positioning data set and an instrument reading data set are obtained; Performing target positioning training on a preset target detection model based on the image verification set and the instrument positioning data set to obtain an instrument positioning model; acquiring text information based on the instrument positioning data set and a preset semantic recognition model; Performing visual coding based on the instrument positioning data set to obtain a feature map; Performing feature fusion based on the text information, the feature map and a preset multi-mode model to obtain text classification data; carrying out semantic recognition training on a preset semantic recognition model based on the text classification data and the meter reading data set to obtain a dial understanding model; And acquiring an original instrument photo, and performing instrument identification reading based on the original instrument photo, an instrument positioning model and a dial understanding model to acquire pointer instrument panel reading information. The pointer type instrument intelligent reading method based on deep learning comprises the steps of firstly constructing an image verification set, an instrument positioning data set and an instrument reading data set through acquiring pointer type instrument images, providing a high-quality data basis for subsequent model training, enabling the models to better face complex recognition application conditions, then training the instrument positioning model based on the image verification set and the instrument positioning data set, realizing accurate extraction of a dial area from an original image based on the instrument positioning model, reducing background interference, solving the problem of fuzzy dial positioning in the traditional method, firstly reducing the processing capacity of subsequent steps through pre-extracting the dial area and eliminating interference of irrelevant areas in the process of model construction and subsequent corresponding reading recognition, further acquiring character information and a visual coding feature map through a preset semantic recognition model according to the instrument positioning data set, combining with multi-mode models to perform feature fusion to obtain character classification data, effective