CN113610085B - Character wheel image identification method based on attention mechanism - Google Patents
Character wheel image identification method based on attention mechanism Download PDFInfo
- Publication number
- CN113610085B CN113610085B CN202111178572.5A CN202111178572A CN113610085B CN 113610085 B CN113610085 B CN 113610085B CN 202111178572 A CN202111178572 A CN 202111178572A CN 113610085 B CN113610085 B CN 113610085B
- Authority
- CN
- China
- Prior art keywords
- pixels
- inter
- attention
- character wheel
- neural network
- 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.)
- Active
Links
Images
Classifications
-
- 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
- G06N3/045—Combinations of networks
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a character wheel image identification method based on an attention mechanism, which comprises the following steps: acquiring a character wheel image acquired by a camera terminal, and traversing each line of pixels in the character wheel image to calculate the inter-class variance of each line of pixels; inputting the character wheel image acquired by the camera terminal into a neural network based on attention, and obtaining a characteristic image after sampling; acquiring the maximum value of the inter-class variance of each line of pixels, and inputting the maximum value of the inter-class variance of all the lines of pixels into an attention-based neural network as an attention parameter; and (4) acting the attention parameters on the characteristic image, and obtaining characters which are required to be acquired finally by the character wheel image through calculation. The method comprises the steps of firstly calculating the maximum value of the inter-class variance of each row of pixels in a character wheel image, injecting the maximum value of the inter-class variance of all rows into a neural network as an attention parameter, and enabling the neural network to more accurately segment and identify characters in the character wheel image by using the neural network.
Description
Technical Field
The invention relates to the technical field of image recognition, in particular to a character wheel image recognition method based on an attention mechanism.
Background
The camera shooting gas meter is mainly applied to the transformation of an ordinary gas meter, and specifically means that a buckle with a camera and a communication module is additionally arranged on a basic meter of an old ordinary gas meter to sense the number wheel reading of the gas meter. In such a scenario, since the manufacturers of the basic tables are different, the character wheel images (also referred to as images of character wheels, images of characters, and images of characters) on the gas meter captured by the installed cameras are also different.
Typical differences are: (1) only the image of the character to be recognized exists; (2) the character wheel image to be identified is superposed with the shadow; (3) character wheel images with interference rows at the upper ends; (4) character wheel images of interference rows exist at the upper end and the lower end. In general, the character wheel images of different basic table manufacturers have the following problems compared with the standard images: (1) the interference rows are mainly concentrated at the upper end and the lower end; (2) the disturbing lines are mainly embodied as background, i.e. pure white or pure black, and the pixel standard deviation of each line is low.
At present, although there is a document (publication number CN 107610144B) that discloses an infrared image segmentation method by the maximum inter-class variance method, in which a region where a contrast between a target and a sky background is large is segmented, a sky background image is not necessarily present at the time of acquiring a character wheel image of a photographed gas meter, and a difference between a contrast between an interference line and a character in the character wheel image is not large, so that the method of the document is not completely accurate for segmenting a character in the character wheel image.
Disclosure of Invention
The invention aims to inject an attention mechanism into a convolutional neural network, and provides a character wheel image recognition method based on the attention mechanism, which can more accurately recognize characters in a character wheel image.
In order to achieve the above object, the embodiments of the present invention provide the following technical solutions:
the character wheel image identification method based on the attention mechanism comprises the following steps of:
step S1: acquiring a character wheel image acquired by a camera terminal, and traversing each line of pixels in the character wheel image to calculate the inter-class variance of each line of pixels;
step S2: inputting the character wheel image acquired by the camera terminal into a neural network based on attention, and obtaining a characteristic image after sampling;
step S3: acquiring the maximum value of the inter-class variance of each line of pixels, and inputting the maximum value of the inter-class variance of all the lines of pixels into an attention-based neural network as an attention parameter; and (4) acting the attention parameters on the characteristic image, and obtaining characters which are required to be acquired finally by the character wheel image through calculation.
In the scheme, the maximum value of the inter-class variance of each row of pixels in the character wheel image is calculated firstly, the maximum value of the inter-class variance of all rows is used as an attention parameter to be injected into the neural network, and the neural network based on attention is formed by the common neural network, so that the segmentation and identification of characters in the character wheel image by the neural network are more accurate.
The step of traversing each row of pixels in the word-wheel image to calculate the inter-class variance of each row of pixels includes:
step S11: the method comprises the following steps of (1) totally obtaining n rows of pixels, and calculating the minimum gray value qmin and the maximum gray value qmax of each row of pixels;
step 12: traversing each gray value between the minimum gray value qmin and the maximum gray value qmax of each row of pixels, and calculating the inter-class variance G under each gray value, wherein G = { G =i}，giRepresenting the inter-class variance under the gray value of the ith traversal;
step S13: taking the inter-class variance g in each row of pixelsiAs the inter-class variance of the pixels of the row, Ln = max (g).
The step of calculating the inter-class variance g under each gray value comprises the following steps:
step S121: the gray value of each traversal is recorded as ti，tiRepresenting the gray value of the ith traversal;
step S122: with tiTo separate the thresholds, the pixels in the row are smaller than or equal to tiThe pixel point of (2) is used as a background pixel and is larger than tiThe pixel point of (2) is taken as a foreground pixel;
step S123: calculating the mean gray value mu of the foreground pixels0And the ratio omega of the number of background pixels to the total number of pixels of the row0(ii) a Calculating the mean gray value mu of the foreground pixels1And the ratio omega of the number of background pixels to the total number of pixels of the row1；
Step S124: calculating the variance g = omega between current classes0*ω1*(μ0-μ1)2。
The attention-based neural network consists of 1 convolutional neural network and 1 Bp neural network; the convolutional neural network consists of 1 input layer, 2 convolutional layers, 2 pooling layers and 1 full-connection layer; the Bp neural network is composed of 1 input layer, 1 hidden layer and 1 output layer.
The character wheel image acquired by the camera terminal is input into a neural network based on attention, and a characteristic image is obtained after sampling, and the method comprises the following steps:
step S21: the pixel of the character wheel image collected by the camera terminal is 28 × 28, the character wheel image is input into a first convolution layer of a convolution neural network, and a feature vector diagram with 6 pixels being 24 × 24 is generated after 6 convolution kernels of the first convolution layer are calculated; inputting the generated feature vector diagram with 6 pixels of 24 into a first pooling layer, performing 2 x 2 maximum downsampling calculation on the feature vector diagram with the pixels of 24 into a feature vector diagram with the pixels of 12 x 12, and generating a feature vector diagram with the 6 pixels of 12 x 12;
step S22: inputting the feature vector diagram with 6 pixels of 12 × 12 generated by the first pooling layer into a second pooling layer, and generating a feature vector diagram with 12 pixels of 8 × 8 after 12 × 5 × 6 convolution kernels of the second pooling layer are calculated; and inputting the generated feature vector diagram with 12 pixels of 8 × 8 into the second pooling layer, performing 2 × 2 maximum downsampling calculation on the feature vector diagram with the pixels of 8 × 8, and converting the feature vector diagram with the pixels of 4 × 4 into the feature vector diagram with the pixels of 4 × 4, thereby generating the feature vector diagram with 12 pixels of 4 × 4.
The maximum value of the inter-class variance of each line of pixels is obtained, and the maximum value of the inter-class variance of all the lines of pixels is used as an attention parameter and is input into an attention-based neural network; the method comprises the following steps of applying attention parameters to a characteristic image, and obtaining characters which are finally required to be obtained by a character wheel image through calculation, wherein the steps comprise:
step S31: acquiring the maximum value of the inter-class variance of the 28 rows of pixels, inputting the maximum value as 28 attention parameters into an input layer of a Bp neural network, and outputting 4 neurons by an output layer after the number of nodes of a hidden layer is estimated;
step S32: and respectively acting 4 neurons output by the output layer on the feature vector diagram with each pixel being 4 x 4 generated by the second pooling layer, obtaining the feature diagram after weighting processing, and acquiring the required characters from the feature diagram.
In the scheme, the maximum value of the inter-pixel-class variance of all rows of the character wheel image is used as an attention parameter to be injected into the convolutional neural network, which is equivalent to that an attention mechanism is added in the ordinary convolutional neural network to form a newly-added Bp neural network, and the convolution neural network has a better basis in the character recognition process of the character wheel image by combining the inter-pixel-class variance of the character wheel image, so that the finally-recognized character is more accurate.
Compared with the prior art, the invention has the beneficial effects that:
the method comprises the steps of firstly calculating the maximum value of the inter-class variance of each row of pixels in a character wheel image, injecting the maximum value of the inter-class variance of all rows into a neural network as an attention parameter, and enabling the neural network to more accurately segment and identify characters in the character wheel image by using the neural network.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present invention and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained according to the drawings without inventive efforts.
FIG. 1 is a flow chart of a print wheel image recognition method of the present invention;
FIG. 2 is a diagram illustrating an example of an interfering row at the top and bottom ends of a character;
FIG. 3 is a diagram illustrating a maximum value curve of variance between pixel classes according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of an attention-based neural network according to an embodiment of the present invention;
FIG. 5 is a schematic diagram of a Bp neural network structure according to an embodiment of the present invention;
FIG. 6 is a diagram illustrating the calculation of the feature vector graph with attention parameters according to the embodiment 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. The components of embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of the embodiments of the present invention, presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments of the present invention without making any creative effort, shall fall within the protection scope of the present invention.
It should be noted that: like reference numbers and letters refer to like items in the following figures, and thus, once an item is defined in one figure, it need not be further defined and explained in subsequent figures. Also, in the description of the present invention, the terms "first", "second", and the like are used for distinguishing between descriptions and not necessarily for describing a relative importance or implying any actual relationship or order between such entities or operations.
Example (b):
the invention is realized by the following technical scheme, as shown in figure 1, the character wheel image identification method based on the attention mechanism comprises the following steps:
step S1: and acquiring a character wheel image acquired by the camera terminal, and traversing each row of pixels in the character wheel image to calculate the inter-class variance of each row of pixels.
In this embodiment, a character wheel image with 28 × 28 pixels is described, and the character wheel image has 28 rows of pixels, and the minimum grayscale value qmin and the maximum grayscale value qmax of each row of pixels are calculated. For example, the minimum gray value of the first row of pixels is qmin1Maximum gray value qmax1The minimum gray value in the first row is qmin1And maximum gray value qmax1If there are 10 gray values in between, then calculate the inter-class variance G under each gray value in turn, and there is a set G = { G = { (G) }1,g2,...g10In which g is1Represents the inter-class variance at the gray value of the 1 st traversal, i.e., the minimum gray value is qmin1And maximum gray value qmax1Between 1 st gray value; same principle g10Representing the gray value of the 10 th traversal.
When traversing the gray value, the gray value traversed each time is ti，tiRepresenting the gray value of the i-th traversal, e.g. t1Then it indicates thatGray value of 1 traversal, t10The gray value for the 10 th traversal is indicated. Because the gray values need to be traversed sequentially, when the 1 st gray value is traversed, the t1 is used as a separation threshold, the pixel points which are less than or equal to t1 in the 1 st line of pixels are used as background pixels, and the pixel points which are greater than t1 are used as foreground pixels.
Suppose that 15 pixels in 27 pixels in the 1 st row (excluding the pixel at t1 at this time) are less than or equal to t1, and 12 pixels are greater than t1, then the 15 pixels less than or equal to t1 are used as background pixels, and the 12 pixels greater than t1 are used as foreground pixels.
The bottleneck gray value mu of the foreground pixel is then calculated0And the ratio omega of the number of background pixels to the total number of pixels of the row0(ii) a Calculating the mean gray value mu of the foreground pixels1And the ratio omega of the number of background pixels to the total number of pixels of the row1. After obtaining these parameters, the inter-class variance g of the 1 st traversal gray value can be calculated1=ω0*ω1*(μ0-μ1)2。
According to the method, by analogy, the inter-class variance of the 10-time traversal gray-scale values in the lower 1 row of pixels can be calculated, that is, the set G = { G = { G }1,g2,...g10And acquiring a maximum value Ln = max (G) in the set G as the inter-class variance of the pixels in the first row. Then after the inter-class variance calculation for the 28 rows of pixels, the maximum inter-class variance for the 28 pixels can be obtained.
After the maximum value of the inter-class variance of the pixels is calculated in step S1, the interval between the upper end and the lower end (the dashed-line box A, B in fig. 2) in the character "0" shown in fig. 2 can be distinguished, please refer to the maximum value curve diagram of the inter-class variance of the pixels shown in fig. 3, where the abscissa is the serial number of the pixel row, 28 rows are labeled in total, the ordinate is the maximum value of the inter-class variance corresponding to each row of pixels, and it can be seen that the maximum value of the inter-class variance is very small because there is no character in the pixels of the first row and the pixels of the last row. Therefore, after the maximum value of the variance among the pixel classes is calculated, the lines with characters and the lines without characters in the character wheel image can be distinguished.
Step S2: and inputting the character wheel image acquired by the camera terminal into a neural network based on attention, and obtaining a characteristic image after sampling.
Referring to fig. 4, the attention-based neural network is composed of 1 convolutional neural network and 1 Bp neural network. The convolutional neural network consists of 1 input layer, 2 convolutional layers (a first convolutional layer and a second convolutional layer respectively), 2 pooling layers (a first pooling layer and a second pooling layer respectively) and 1 full-connection layer; the Bp neural network is composed of 1 input layer, 1 hidden layer and 1 output layer. It should be noted that the dashed boxes in fig. 4 only represent the structure of the convolutional neural network and the structure of the Bp neural network, and the dashed boxes have no other specific meaning.
Inputting a character wheel image with 28 × 28 pixels acquired by a camera terminal into a first convolution layer of a convolution neural network, and generating a feature vector diagram with 24 × 24 pixels after 6 convolution kernels of the first convolution layer are calculated by 5 × 5 convolution kernels; the generated feature vector diagram with 6 pixels of 24 is input into the first pooling layer, and after 2 × 2 maximum downsampling calculation of the first pooling layer, the feature vector diagram with 24 pixels of 24 is converted into a feature vector diagram with 12 pixels of 12, so that the feature vector diagram with 6 pixels of 12 is generated.
Then inputting the feature vector diagram with 6 pixels of 12 x 12 generated by the first pooling layer into a second pooling layer, and generating a feature vector diagram with 12 pixels of 8 x 8 after 12 convolution kernels of 5 x 6 of the second pooling layer are calculated; and inputting the generated feature vector diagram with 12 pixels of 8 × 8 into the second pooling layer, performing 2 × 2 maximum downsampling calculation on the feature vector diagram with the pixels of 8 × 8, and converting the feature vector diagram with the pixels of 4 × 4 into the feature vector diagram with the pixels of 4 × 4, thereby generating the feature vector diagram with 12 pixels of 4 × 4.
Step S3: acquiring the maximum value of the inter-class variance of each line of pixels, and inputting the maximum value of the inter-class variance of all the lines of pixels into an attention-based neural network as an attention parameter; and (4) acting the attention parameters on the characteristic image, and obtaining characters which are required to be acquired finally by the character wheel image through calculation.
Referring to fig. 4, the present solution is equivalent to adding an attention-based network, i.e., a Bp neural network, to the convolutional neural network, and using the maximum value of the inter-class variance of 28 pixels obtained through inter-class variance calculation as an input layer of 28 attention parameters input to the Bp neural network, referring to fig. 5, after the number of 10 nodes of the hidden layer is estimated, 4 neurons are output by an output layer. It should be noted that the dashed boxes in fig. 5 only represent the input layer, the hidden layer, and the output layer, and the dashed boxes have no other specific meaning.
And respectively acting 4 neurons output by the output layer on the feature vector diagram with 4 × 4 pixels generated by the second pooling layer. Referring to fig. 6, it is assumed that parameters of the 1 st feature vector diagram in the feature vector diagram with 12 pixels of 4 × 4 are shown as S1, and after 4 neurons P1 act, the 1 st row of S1 is multiplied by the 1 st row of P1 to obtain the 1 st row of N1, and so on to obtain the feature vector diagram N1. Then 12 feature vector diagrams after the action can be obtained, a final feature diagram is obtained after weighting processing, and the required characters can be obtained from the feature diagram.
The above description is only for the specific embodiments of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art can easily conceive of the changes or substitutions within the technical scope of the present invention, and all the changes or substitutions should be covered within the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (3)
1. The character wheel image identification method based on the attention mechanism is characterized in that: the method comprises the following steps:
step S1: acquiring a character wheel image acquired by a camera terminal, and traversing each line of pixels in the character wheel image to calculate the inter-class variance of each line of pixels;
step S2: inputting the character wheel image acquired by the camera terminal into a neural network based on attention, and obtaining a characteristic image after sampling;
the attention-based neural network consists of 1 convolutional neural network and 1 Bp neural network; the convolutional neural network consists of 1 input layer, 2 convolutional layers, 2 pooling layers and 1 full-connection layer; the Bp neural network consists of 1 input layer, 1 hidden layer and 1 output layer;
the character wheel image acquired by the camera terminal is input into a neural network based on attention, and a characteristic image is obtained after sampling, and the method comprises the following steps:
step S21: the pixel of the character wheel image collected by the camera terminal is 28 × 28, the character wheel image is input into a first convolution layer of a convolution neural network, and a feature vector diagram with 6 pixels being 24 × 24 is generated after 6 convolution kernels of the first convolution layer are calculated; inputting the generated feature vector diagram with 6 pixels of 24 into a first pooling layer, performing 2 x 2 maximum downsampling calculation on the feature vector diagram with the pixels of 24 into a feature vector diagram with the pixels of 12 x 12, and generating a feature vector diagram with the 6 pixels of 12 x 12;
step S22: inputting the feature vector diagram with 6 pixels of 12 × 12 generated by the first pooling layer into a second pooling layer, and generating a feature vector diagram with 12 pixels of 8 × 8 after 12 × 5 × 6 convolution kernels of the second pooling layer are calculated; inputting the generated feature vector diagram with 12 pixels being 8 × 8 into a second pooling layer, performing 2 × 2 maximum downsampling calculation on the second pooling layer, and converting the feature vector diagram with the pixels being 8 × 8 into a feature vector diagram with the pixels being 4 × 4, thereby generating a feature vector diagram with 12 pixels being 4 × 4;
step S3: acquiring the maximum value of the inter-class variance of each line of pixels, and inputting the maximum value of the inter-class variance of all the lines of pixels into an attention-based neural network as an attention parameter; the attention parameters are acted on the characteristic image, and characters which are required to be acquired finally by the character wheel image are obtained through calculation;
the maximum value of the inter-class variance of each line of pixels is obtained, and the maximum value of the inter-class variance of all the lines of pixels is used as an attention parameter and is input into an attention-based neural network; the method comprises the following steps of applying attention parameters to a characteristic image, and obtaining characters which are finally required to be obtained by a character wheel image through calculation, wherein the steps comprise:
step S31: acquiring the maximum value of the inter-class variance of the 28 rows of pixels, inputting the maximum value as 28 attention parameters into an input layer of a Bp neural network, and outputting 4 neurons by an output layer after the number of nodes of a hidden layer is estimated;
step S32: and respectively acting 4 neurons output by the output layer on the feature vector diagram with each pixel being 4 x 4 generated by the second pooling layer, obtaining the feature diagram after weighting processing, and acquiring the required characters from the feature diagram.
2. The attention-based character wheel image recognition method of claim 1, wherein: the step of traversing each row of pixels in the word-wheel image to calculate the inter-class variance of each row of pixels includes:
step S11: the method comprises the following steps of (1) totally obtaining n rows of pixels, and calculating the minimum gray value qmin and the maximum gray value qmax of each row of pixels;
step 12: traversing each gray value between the minimum gray value qmin and the maximum gray value qmax of each row of pixels, and calculating the inter-class variance G under each gray value, wherein G = { G =i}，giRepresenting the inter-class variance under the gray value of the ith traversal;
step S13: taking the inter-class variance g in each row of pixelsiAs the inter-class variance of the pixels of the row, Ln = max (g).
3. The attention-based character wheel image recognition method of claim 2, wherein: the step of calculating the inter-class variance g under each gray value comprises the following steps:
step S121: the gray value of each traversal is recorded as ti，tiRepresenting the gray value of the ith traversal;
step S122: with tiTo separate the thresholds, the pixels in the row are smaller than or equal to tiThe pixel point of (2) is used as a background pixel and is larger than tiThe pixel point of (2) is taken as a foreground pixel;
step S123: calculating the mean gray value mu of the foreground pixels0And the ratio omega of the number of background pixels to the total number of pixels of the row0(ii) a Calculating average gray of foreground pixelsValue of mu1And the ratio omega of the number of background pixels to the total number of pixels of the row1；
Step S124: calculating the variance g = omega between current classes0*ω1*(μ0-μ1)2。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111178572.5A CN113610085B (en) | 2021-10-10 | 2021-10-10 | Character wheel image identification method based on attention mechanism |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111178572.5A CN113610085B (en) | 2021-10-10 | 2021-10-10 | Character wheel image identification method based on attention mechanism |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113610085A CN113610085A (en) | 2021-11-05 |
| CN113610085B true CN113610085B (en) | 2021-12-07 |
Family
ID=78343443
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111178572.5A Active CN113610085B (en) | 2021-10-10 | 2021-10-10 | Character wheel image identification method based on attention mechanism |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113610085B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114866297B (en) * | 2022-04-20 | 2023-11-24 | 中国科学院信息工程研究所 | Network data detection method and device, electronic equipment and storage medium |
Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2010128590A (en) * | 2008-11-25 | 2010-06-10 | Chugoku Electric Power Co Inc:The | Meter-reading terminal and meter reading system using the same |
| CN107610144A (en) * | 2017-07-21 | 2018-01-19 | 哈尔滨工程大学 | A kind of improved IR image segmentation method based on maximum variance between clusters |
| CN109255344A (en) * | 2018-08-15 | 2019-01-22 | 华中科技大学 | A kind of digital display instrument positioning and Recognition of Reading method based on machine vision |
| US10425687B1 (en) * | 2017-10-10 | 2019-09-24 | Facebook, Inc. | Systems and methods for determining television consumption behavior |
| CN110414377A (en) * | 2019-07-09 | 2019-11-05 | 武汉科技大学 | A kind of remote sensing images scene classification method based on scale attention network |
| CN110458797A (en) * | 2019-06-18 | 2019-11-15 | 南开大学 | A kind of conspicuousness object detecting method based on depth map filter |
| CN110728697A (en) * | 2019-09-30 | 2020-01-24 | 华中光电技术研究所（中国船舶重工集团有限公司第七一七研究所） | Infrared dim target detection tracking method based on convolutional neural network |
| CN111031222A (en) * | 2019-12-27 | 2020-04-17 | 山东厚德测控技术股份有限公司 | Real-time recognition device and method for character wheel of camera type gas meter |
| CN111275714A (en) * | 2020-01-13 | 2020-06-12 | 武汉大学 | Prostate MR image segmentation method based on attention mechanism 3D convolutional neural network |
| CN111832924A (en) * | 2020-06-30 | 2020-10-27 | 北方工业大学 | Dynamic risk assessment method and device for community gas system based on graph neural network |
| CN111915613A (en) * | 2020-08-11 | 2020-11-10 | 华侨大学 | Image instance segmentation method, device, equipment and storage medium |
| CN113205574A (en) * | 2021-04-30 | 2021-08-03 | 武汉大学 | Art character style migration system based on attention system |
| CN113283419A (en) * | 2021-04-29 | 2021-08-20 | 国网浙江省电力有限公司湖州供电公司 | Convolutional neural network pointer instrument image reading identification method based on attention |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030171851A1 (en) * | 2002-03-08 | 2003-09-11 | Peter J. Brickfield | Automatic energy management and energy consumption reduction, especially in commercial and multi-building systems |
| WO2020132478A1 (en) * | 2018-12-20 | 2020-06-25 | Dqpn, Llc | Diet quality fingerprinting |
| CN109902751B (en) * | 2019-03-04 | 2022-07-08 | 福州大学 | Dial digital character recognition method integrating convolution neural network and half-word template matching |
| US20200359550A1 (en) * | 2019-05-13 | 2020-11-19 | Bao Tran | Farm ecosystem |
| CA3145371A1 (en) * | 2019-06-25 | 2020-12-30 | Owkin Inc. | Systems and methods for image preprocessing |
| US11423678B2 (en) * | 2019-09-23 | 2022-08-23 | Proscia Inc. | Automated whole-slide image classification using deep learning |
| CN111008639B (en) * | 2019-10-17 | 2024-02-27 | 安徽清新互联信息科技有限公司 | License plate character recognition method based on attention mechanism |
| US12062178B2 (en) * | 2020-03-05 | 2024-08-13 | Case Western Reserve University | Automated segmentation and guided correction of endothelial cell images |
| CN111582241B (en) * | 2020-06-01 | 2022-12-09 | 腾讯科技（深圳）有限公司 | Video subtitle recognition method, device, equipment and storage medium |
-
2021
- 2021-10-10 CN CN202111178572.5A patent/CN113610085B/en active Active
Patent Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2010128590A (en) * | 2008-11-25 | 2010-06-10 | Chugoku Electric Power Co Inc:The | Meter-reading terminal and meter reading system using the same |
| CN107610144A (en) * | 2017-07-21 | 2018-01-19 | 哈尔滨工程大学 | A kind of improved IR image segmentation method based on maximum variance between clusters |
| US10425687B1 (en) * | 2017-10-10 | 2019-09-24 | Facebook, Inc. | Systems and methods for determining television consumption behavior |
| CN109255344A (en) * | 2018-08-15 | 2019-01-22 | 华中科技大学 | A kind of digital display instrument positioning and Recognition of Reading method based on machine vision |
| CN110458797A (en) * | 2019-06-18 | 2019-11-15 | 南开大学 | A kind of conspicuousness object detecting method based on depth map filter |
| CN110414377A (en) * | 2019-07-09 | 2019-11-05 | 武汉科技大学 | A kind of remote sensing images scene classification method based on scale attention network |
| CN110728697A (en) * | 2019-09-30 | 2020-01-24 | 华中光电技术研究所（中国船舶重工集团有限公司第七一七研究所） | Infrared dim target detection tracking method based on convolutional neural network |
| CN111031222A (en) * | 2019-12-27 | 2020-04-17 | 山东厚德测控技术股份有限公司 | Real-time recognition device and method for character wheel of camera type gas meter |
| CN111275714A (en) * | 2020-01-13 | 2020-06-12 | 武汉大学 | Prostate MR image segmentation method based on attention mechanism 3D convolutional neural network |
| CN111832924A (en) * | 2020-06-30 | 2020-10-27 | 北方工业大学 | Dynamic risk assessment method and device for community gas system based on graph neural network |
| CN111915613A (en) * | 2020-08-11 | 2020-11-10 | 华侨大学 | Image instance segmentation method, device, equipment and storage medium |
| CN113283419A (en) * | 2021-04-29 | 2021-08-20 | 国网浙江省电力有限公司湖州供电公司 | Convolutional neural network pointer instrument image reading identification method based on attention |
| CN113205574A (en) * | 2021-04-30 | 2021-08-03 | 武汉大学 | Art character style migration system based on attention system |
Non-Patent Citations (7)
| Title |
|---|
| "Automatic Utility Meter Reading ";Kaicheng Xie;《https://engagedscholarship.csuohio.edu/cgi/viewcontent.cgi?article=1682&context=etdarchive》;20201013;1-83 * |
| "Convolutional neural networks for automatic meter reading";Rayson Laroca等;《Journal of Electronic Imaging 》;20190205;第28卷(第1期) * |
| "Improving the Performance of Convolutional Neural Network for the Segmentation of Optic Disc in Fundus Images Using Attention Gates and Conditional Random Fields";B. J. Bhatkalkar等;《 IEEE 》;20200207;第 8卷;29299-29310 * |
| "基于卷积神经网络的图像识别研究综述";黄志强 等;《汽车工程师 》;20201025(第10期);11-13+31 * |
| "基于图像识别与LoRa的自动远传抄表系统设计与实现";熊诚;《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》;20181215(第（2018）12期);C030-98 * |
| "基于深度学习的工业仪表识别读数算法研究及应用;何配林;《中国优秀硕士学位论文全文数据库 (基础科学辑)》;20200715(第（2020）7期);C030-83 * |
| "燃气表读数的数字图像识别研究";李发传;《中国优秀硕士学位论文全文数据库 信息科技辑》;20150215(第（2016）2期);I138-1883 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113610085A (en) | 2021-11-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109584248B (en) | Infrared target instance segmentation method based on feature fusion and dense connection network | |
| CN108229490B (en) | Key point detection method, neural network training method, device and electronic equipment | |
| US10733477B2 (en) | Image recognition apparatus, image recognition method, and program | |
| CN108108746B (en) | License plate character recognition method based on Caffe deep learning framework | |
| CN111460931A (en) | Face spoofing detection method and system based on color channel difference image characteristics | |
| CN107273870A (en) | The pedestrian position detection method of integrating context information under a kind of monitoring scene | |
| CN112036260B (en) | Expression recognition method and system for multi-scale sub-block aggregation in natural environment | |
| CN113989794B (en) | License plate detection and recognition method | |
| CN113610085B (en) | Character wheel image identification method based on attention mechanism | |
| CN111815528A (en) | Bad weather image classification enhancement method based on convolution model and feature fusion | |
| WO2023011280A1 (en) | Image noise degree estimation method and apparatus, and electronic device and storage medium | |
| CN111428753B (en) | Training set acquisition method, electric power facility detection method and device | |
| CN115841438A (en) | Infrared image and visible light image fusion method based on improved GAN network | |
| CN107274425B (en) | A kind of color image segmentation method and device based on Pulse Coupled Neural Network | |
| Zhang et al. | Single image dehazing based on bright channel prior model and saliency analysis strategy | |
| CN113554568B (en) | Unsupervised cyclic rain-removing network method based on self-supervision constraint and unpaired data | |
| CN108710881B (en) | Neural network model, candidate target area generation method and model training method | |
| CN113807237A (en) | Training of in vivo detection model, in vivo detection method, computer device, and medium | |
| Zhang et al. | Dehazing with improved heterogeneous atmosphere light estimation and a nonlinear color attenuation prior model | |
| CN110781936B (en) | Construction method of threshold learnable local binary network based on texture description and deep learning and remote sensing image classification method | |
| Satrasupalli et al. | End to end system for hazy image classification and reconstruction based on mean channel prior using deep learning network | |
| CN116012299A (en) | Composite insulator hydrophobicity grade detection method based on target identification | |
| CN111223050A (en) | Real-time image edge detection algorithm | |
| Yadav et al. | Image detection in noisy images | |
| JP7279817B2 (en) | Image processing device, image processing method and image processing program |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CP01 | Change in the name or title of a patent holder | ||
| CP01 | Change in the name or title of a patent holder |
Address after: No. 536, Section 1, airport 1st Road, Southwest Airport, Shuangliu District, Chengdu, Sichuan 610211 Patentee after: Chengdu Qianjia Technology Co.,Ltd. Address before: No. 536, Section 1, airport 1st Road, Southwest Airport, Shuangliu District, Chengdu, Sichuan 610211 Patentee before: Chengdu Qianjia Technology Co.,Ltd. |