CN113609894A - Intelligent meter reading algorithm based on neural network - Google Patents

Intelligent meter reading algorithm based on neural network Download PDF

Info

Publication number
CN113609894A
CN113609894A CN202110686435.6A CN202110686435A CN113609894A CN 113609894 A CN113609894 A CN 113609894A CN 202110686435 A CN202110686435 A CN 202110686435A CN 113609894 A CN113609894 A CN 113609894A
Authority
CN
China
Prior art keywords
neural network
camera
meter reading
intelligent meter
algorithm based
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202110686435.6A
Other languages
Chinese (zh)
Inventor
何湫雨
徐晨鑫
朱恩东
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing Beixin Intelligent Technology Co ltd
Original Assignee
Nanjing Beixin Intelligent Technology Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nanjing Beixin Intelligent Technology Co ltd filed Critical Nanjing Beixin Intelligent Technology Co ltd
Priority to CN202110686435.6A priority Critical patent/CN113609894A/en
Publication of CN113609894A publication Critical patent/CN113609894A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses an intelligent meter reading algorithm based on a neural network, and a using method of the intelligent meter reading algorithm comprises the following steps: A. reading a video frame through a camera; and A1, cleaning the lens of the camera. Aiming at the current industrial severe standard required for shooting instruments, the invention utilizes the neural network to calculate the distribution matrix transferred from one probability distribution to another probability distribution, thereby realizing the rapid matching of the instruments and further accurate reading, improving the real-time property and accuracy compared with the traditional SIFT mode, being better used for instruments under different scenes, realizing the purposes of high processing speed, high precision and robustness, and solving the problem that the current mainstream method in the market is to firstly extract feature points by using the traditional algorithm and then read the different instruments after template matching, the method has high requirements for shooting, and the traditional algorithm can cause matching failure due to environmental illumination, deflection angle and the reflection degree of the instrument shell.

Description

Intelligent meter reading algorithm based on neural network
Technical Field
The invention relates to the technical field of intelligent meter reading, in particular to an intelligent meter reading algorithm based on a neural network.
Background
The invention adopts the technology of automatically reading the meter by a neural network, saves the labor cost and greatly improves the meter reading efficiency.
Disclosure of Invention
Aiming at the problems in the prior art, the invention aims to provide an intelligent meter reading algorithm based on a neural network, which has the advantages of high processing speed, high precision and strong robustness, and solves the problems that the traditional algorithm is used for extracting feature points and reading after template matching is carried out on different meters, the requirement of the method on shooting is high, and the traditional algorithm can cause matching failure due to ambient illumination, deflection angle and the light reflection degree of a meter shell.
In order to achieve the purpose, the invention provides the following technical scheme: an intelligent meter reading algorithm based on a neural network, wherein the using method comprises the following steps:
A. reading a video frame through a camera;
a1, cleaning the lens of the camera;
a2, adjusting the irradiation angle of the camera and irradiating the instrument panel;
B. acquiring feature points by using a neural network;
C. whether an instrument exists in the video frame or not;
c1, two cases are distinguished: absence and presence;
D. adjusting the camera shooting angle of the camera in the step A under the condition that the camera does not exist in the step C1;
E. the cases existing in step C1 are divided into two cases:
e1, reading the digital table by direct template matching;
and E2, carrying out Hough line detection on the pointer instrument to judge the position of the pointer and read.
Preferably, the camera in the step a is a 1080P high-definition camera, and cleanliness of a lens of the camera is to be ensured.
Preferably, the instrument panel in the step a cannot reflect light in a large area.
Preferably, an encoder is used in the step E to integrate the positions and descriptions of the previously extracted points, and then the characteristic points in and among the graphs are connected in graph form, and the relevance is constructed by using an attention aggregation mechanism.
Preferably, the step E locates feature points appearing in the video frame based on a neural network.
Preferably, said step E uses a neural network to calculate a distribution matrix for the transition from one probability distribution to another.
Compared with the prior art, the invention has the following beneficial effects:
1. aiming at the current industrial severe standard required for shooting instruments, the invention utilizes the neural network to calculate the distribution matrix transferred from one probability distribution to another probability distribution, thereby realizing fast instrument matching and accurate reading, improving the real-time property and accuracy compared with the traditional SIFT mode, being better used for instruments under different scenes, realizing the purposes of fast processing speed, high precision and robustness, and solving the problem that the current mainstream method in the market is to extract feature points by using the traditional algorithm firstly and then read after template matching aiming at different instruments, the method has high requirements for shooting, and the traditional algorithm can cause matching failure due to environmental illumination, deflection angle and the reflection degree of an instrument shell.
Drawings
Fig. 1 is a schematic diagram of the working principle of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In the description of the present invention, it should be noted that the terms "upper", "lower", "inner", "outer", "top/bottom", and the like indicate orientations or positional relationships based on those shown in the drawings, and are only for convenience of description and simplification of description, but do not indicate or imply that the referred device or element must have a specific orientation, be constructed in a specific orientation, and be operated, and thus, should not be construed as limiting the present invention. Furthermore, the terms "first" and "second" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance.
In the description of the present invention, it should be noted that, unless otherwise specifically stated or limited, the terms "mounted," "disposed," "sleeved/connected," "connected," and the like are to be construed broadly, e.g., "connected," which may be fixedly connected, detachably connected, or integrally connected; can be mechanically or electrically connected; they may be connected directly or indirectly through intervening media, or they may be interconnected between two elements. The specific meanings of the above terms in the present invention can be understood in specific cases to those skilled in the art.
Referring to fig. 1, an intelligent meter reading algorithm based on a neural network is used, and the method includes the following steps:
A. reading a video frame through a camera;
a1, cleaning the lens of the camera;
a2, adjusting the irradiation angle of the camera and irradiating the instrument panel;
B. acquiring feature points by using a neural network;
C. whether an instrument exists in the video frame or not;
c1, two cases are distinguished: absence and presence;
D. adjusting the camera shooting angle of the camera in the step A under the condition that the camera does not exist in the step C1;
E. the cases existing in step C1 are divided into two cases:
f1, reading the digital table by direct template matching;
e2, for the pointer instrument, hough line detection is performed to determine the pointer position for reading, and the basic working principle can be seen from the flowchart, wherein the innovation point is that the instrument is matched in the environment, which is essentially the processing of extracting characteristic values: firstly, an encoder is used for integrating the positions and descriptions of the points extracted from the previous steps, and then characteristic points in the graph and between the graphs are connected in a graph form, and the relevance is constructed by using an attention aggregation mechanism: aggregating and calculating m epsilon → i of the message, then performing self-attention and cross-attention operations, wherein self-attention is based on self-edges, cross-attention is based on cross-edges, similar to database retrieval, and the representation of i is based on the attribute values v of some elementsjThe message is calculated as a weighted average of these values:
Figure RE-GDA0003290103570000051
in which α isijKey-value pair similarity for Softmax:
Figure RE-GDA0003290103570000052
calculating key, query and value, then taking the key, query and value as linear projection of the neural network depth feature, considering that the query key point i is in the image Q, all the source key points are in the image S, and (Q, S) belongs to { A, B }2Then, it can be:
Figure RE-GDA0003290103570000053
Figure RE-GDA0003290103570000054
in the neural network, each layer l has its own projection parameters, and learns and shares all the key points of the two images. This supports us to find similar, salient key points in the vicinity, based on the idea that we can perform geometric projection, and the final matched linear projection is:
Figure RE-GDA0003290103570000055
the result is matching the approximate points in the graph B. The optimization problem is that Sinkhorn algorithm is used, so that neural network sum is the most
Figure RE-GDA0003290103570000061
And (3) minimizing:
Figure RE-GDA0003290103570000062
after the characteristic points are successfully predicted, the table can be read by a classical algorithm, the values of the characteristic points are retrieved and calculated by a neural network, and the distribution matrix of one probability distribution is transferred to another probability distribution, so that the instrument can be quickly matched, the reading can be accurately performed, the real-time performance and the accuracy are improved compared with the traditional SIFT mode, the method can be better used for instruments in different scenes, the purposes of high processing speed, high precision and strong robustness are realized, and the main stream method in the market at present is to firstly extract the characteristic points by the traditional algorithm and then extract the characteristic points for different scenesThe instrument performs reading after template matching, the method has high requirements on shooting, and the traditional algorithm can cause the problem of matching failure due to ambient light, deflection angle and reflection degree of the instrument shell.
Specifically, the camera in the step a is a 1080P high-definition camera, and cleanliness of a lens of the camera is to be ensured.
Specifically, the instrument panel in step a cannot reflect light in a large area.
Specifically, an encoder is used to integrate the positions and descriptions of the points extracted previously in step E, and then the characteristic points in and between the graphs are connected in graph form, and the attention aggregation mechanism is used for constructing the correlation.
Specifically, step E locates feature points appearing in the video frame based on the neural network.
In particular, step E utilizes a neural network to compute a distribution matrix that transitions from one probability distribution to another.
During the use, at first read the video frame through the camera, when shining the panel board, clean the camera lens of camera, adjust the camera and shine the panel board after shining the angle, later utilize neural network to obtain the characteristic point, then whether have the instrument in the detection video frame, when can not detect the panel board, adjust the angle of shining of camera, when can detecting the panel board, divide into two kinds of situations: firstly, direct template matching is carried out on a digital meter for reading, secondly, Hough line detection is carried out on a pointer meter for judging the position of the pointer for reading, a neural network is utilized for calculating a distribution matrix transferred from one probability distribution to another probability distribution, so that the meter can be quickly matched, and then the reading can be accurately carried out, the aims of high processing speed, high precision and robustness can be fulfilled, and the problems that the conventional algorithm fails to match due to environmental illumination, deflection angle and reflection degree of a meter shell because the conventional method firstly uses the conventional algorithm for extracting characteristic points and then carries out template matching on different meters and then reads are solved.
The standard parts used in the present application document can be purchased from the market, and can be customized according to the description of the specification and the accompanying drawings, the specific connection mode of each part adopts the conventional means of mature bolt, rivet, welding and the like in the prior art, the machines, parts and equipment adopt the conventional models in the prior art, the control mode is automatically controlled by a controller, the control circuit of the controller can be realized by simple programming of the technicians in the field, the control circuit belongs to the common knowledge in the field, and the present application document is mainly used for protecting the mechanical device, so the control mode and the circuit connection are not explained in detail in the present application.
It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.

Claims (6)

1. An intelligent meter reading algorithm based on a neural network is characterized in that the using method comprises the following steps:
A. reading a video frame through a camera;
a1, cleaning the lens of the camera;
a2, adjusting the irradiation angle of the camera and irradiating the instrument panel;
B. acquiring feature points by using a neural network;
C. whether an instrument exists in the video frame or not;
c1, two cases are distinguished: absence and presence;
D. adjusting the camera shooting angle of the camera in the step A under the condition that the camera does not exist in the step C1;
E. the cases existing in step C1 are divided into two cases:
e1, reading the digital table by direct template matching;
and E2, carrying out Hough line detection on the pointer instrument to judge the position of the pointer and read.
2. The intelligent meter reading algorithm based on the neural network as claimed in claim 1, wherein: the camera in the step A is a 1080P high-definition camera, and the cleanliness of a lens of the camera is guaranteed.
3. The intelligent meter reading algorithm based on the neural network as claimed in claim 1, wherein: and B, the instrument panel cannot reflect light in a large area in the step A.
4. The intelligent meter reading algorithm based on the neural network as claimed in claim 1, wherein: in the step E, an encoder is firstly used for integrating the positions and the descriptions of the points extracted previously, and then characteristic points in the graph and between the graphs are connected in a graph form, and an attention aggregation mechanism is used for constructing the correlation.
5. The intelligent meter reading algorithm based on the neural network as claimed in claim 1, wherein: said step E locates feature points occurring in the video frame based on the neural network.
6. The intelligent meter reading algorithm based on the neural network as claimed in claim 1, wherein: said step E uses a neural network to compute a distribution matrix for the transition from one probability distribution to another.
CN202110686435.6A 2021-06-21 2021-06-21 Intelligent meter reading algorithm based on neural network Pending CN113609894A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110686435.6A CN113609894A (en) 2021-06-21 2021-06-21 Intelligent meter reading algorithm based on neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110686435.6A CN113609894A (en) 2021-06-21 2021-06-21 Intelligent meter reading algorithm based on neural network

Publications (1)

Publication Number Publication Date
CN113609894A true CN113609894A (en) 2021-11-05

Family

ID=78336639

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110686435.6A Pending CN113609894A (en) 2021-06-21 2021-06-21 Intelligent meter reading algorithm based on neural network

Country Status (1)

Country Link
CN (1) CN113609894A (en)

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107153848A (en) * 2017-06-15 2017-09-12 南京工程学院 Instrument image automatic identifying method based on OpenCV

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107153848A (en) * 2017-06-15 2017-09-12 南京工程学院 Instrument image automatic identifying method based on OpenCV

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
姬玉柱: "图像及视频显著性物体检测方法研究", 《中国博士学位论文全文数据库 信息科技辑》, 15 January 2021 (2021-01-15) *
陈佳; 胡浩博; 何儒汉; 胡新荣: "基于注意力机制的增强特征描述子", 《计算机工程》, 31 May 2021 (2021-05-31), pages 260 - 266 *

Similar Documents

Publication Publication Date Title
WO2022002150A1 (en) Method and device for constructing visual point cloud map
CN111178197B (en) Mass R-CNN and Soft-NMS fusion based group-fed adherent pig example segmentation method
WO2019119391A1 (en) Cloud-based meter reading method and device
Cong et al. Does thermal really always matter for RGB-T salient object detection?
CN105894036A (en) Image feature template matching method being applied to detection of mobile phone screen defects
CN111445459B (en) Image defect detection method and system based on depth twin network
CN1506903A (en) Automatic fingerprint distinguishing system and method based on template learning
CN111027456B (en) Mechanical water meter reading identification method based on image identification
CN113723157B (en) Crop disease identification method and device, electronic equipment and storage medium
Xuesong et al. Aphid identification and counting based on smartphone and machine vision
CN117274257B (en) Automatic classification system for book looks based on machine vision
CN110399828A (en) A kind of vehicle recognition methods again of the depth convolutional neural networks based on multi-angle
CN113705564A (en) Pointer type instrument identification reading method
CN113609894A (en) Intelligent meter reading algorithm based on neural network
Chen et al. Bi-deformation-UNet: recombination of differential channels for printed surface defect detection
CN114996466B (en) Method and system for establishing medical standard mapping model and using method
CN110451035A (en) A kind of electronic cigarette cigarette detection device and method based on machine vision technique
CN114283280A (en) Water surface floating garbage identification method based on improved convolutional neural network
CN111078912B (en) Power equipment image data warehouse and power equipment defect detection method
CN114463746A (en) Target recognition model training and cell recognition method and device and electronic equipment
Wang et al. Layout and perspective distortion independent recognition of captured chinese document image
CN112115888A (en) Plant disease diagnosis system based on disease spot correlation
CN113537197B (en) Meter automatic modeling method based on machine vision
CN110956130A (en) Method and system for four-level face detection and key point regression
CN117237367B (en) Spiral blade thickness abrasion detection method and system based on machine vision

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination