Water meter reading identification method based on deep learning and prior correction

Through deep learning and prior correction methods, combined with projection method identification and correction, the problem of water meter reading identification in the prior art is solved, achieving higher accuracy and efficiency.

CN118968485APending Publication Date: 2024-11-15BOHAI UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410925006.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-11
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

The existing deep learning-based water meter reading recognition method is easy to cause errors in recognition due to inaccurate final results due to inaccurate final results due to gear occlusion clearance, structural error and shooting angle.

Method used

Using deep learning and prior correction methods, the water meter word wheel number and pointer area detection model is trained, combined with projection method to identify pointer readings, and the recognition results are corrected using two rounding functions, making full use of word wheel number and pointer readings The correlation between them is corrected for the final result.

Benefits of technology

It effectively improves the accuracy of water meter reading recognition, overcomes recognition errors caused by gear occlusion gaps, and reduces the calculation amount and improves the recognition speed.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118968485A_ABST
    Figure CN118968485A_ABST
Patent Text Reader

Abstract

The invention provides a water meter reading recognition method based on deep learning and prior correction, and belongs to the technical field of image recognition and deep learning, and the method comprises the steps: collecting and marking a large number of images, and training a water meter character wheel number region detection model, a water meter character wheel number recognition model and a water meter pointer region detection model; detecting, identifying and correcting a number area of a water meter character wheel; detecting, identifying and correcting a water meter pointer area; and combining and correcting a water meter character wheel number identification result and a pointer identification result. According to the method, a character wheel number recognition initial result, a pointer area recognition initial result and a combination result of numbers and pointers are corrected, recognition errors caused by water meter gear meshing gaps and the like can be effectively overcome, and the accuracy of a water meter reading recognition result is improved; and when the pointer reading is identified by the projection method, only the projection from the center point to 20 points is calculated, so that the operand can be greatly reduced, and the pointer reading identification speed is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of image recognition and deep learning, and specifically relates to a water meter reading recognition method based on deep learning and priori correction. Background Art

[0002] With the rapid development of information and intelligent technology, the traditional manual meter reading method has gradually revealed its drawbacks. This method not only requires a lot of manpower and material resources, but is also inefficient. It is also prone to various errors and mistakes due to human factors, which seriously restricts the modernization and intelligentization of water management. The renovation of water meters in old communities is difficult and costly, which has given rise to the emergence of camera meter reading systems.

[0003] At present, there are some applications of water meter reading recognition methods based on deep learning, but these methods generally recognize each digit or pointer of the water meter in isolation, and then simply connect the recognition results of each digit as the final result. However, due to the meshing gap between the counting gears of the water meter, the structural errors of the water meter itself, and the shooting angle, the word wheel may display incomplete numbers, the word wheel may turn between two words, and the pointer may indicate deviation. At this time, if the reading of each digit or pointer is recognized in isolation and simply connected, the final result will be wrong. Summary of the invention

[0004] Based on the above technical problems, the present invention provides a water meter reading recognition method based on deep learning and prior correction. The method makes full use of the prior knowledge of the correlation between the digits on the water meter wheel and the pointer readings to correct the recognition results, which can effectively improve the accuracy of water meter reading recognition.

[0005] The specific technical solution is:

[0006] A water meter reading recognition method based on deep learning and prior correction includes the following steps:

[0007] S1: Collect and annotate a large number of images, and train the water meter wheel digit area detection model, water meter wheel digit recognition model, and water meter pointer area detection model;

[0008] S2: water meter wheel digital area detection, recognition and correction;

[0009] S3: water meter pointer area detection, identification and correction;

[0010] S4: Combining and correcting the water meter wheel digital recognition result and pointer recognition result.

[0011] In addition, a water meter reading recognition method based on deep learning and prior correction in the above technical solution provided by the present invention may also have the following additional technical features:

[0012] In the above technical solution, if the reading in step S1 is a complete or substantially complete number 0, 1, 2, ..., 9, it is directly marked as the corresponding number 0, 1, 2, ..., 9; and if the reading of the character wheel number is between two numbers, that is, between 0-1, between 1-2, between 2-3, between 3-4, between 4-5, between 5-6, between 6-7, between 7-8, between 8-9 and between 9-0, it is marked as A, B, C, D, E, F, G, H, I and J respectively.

[0013] In the above technical solution, the correction rule for the initial result of water meter wheel digit recognition in step S2 is:

[0014] Let Z' N ,Z' N-1 ,...,Z'1 is the initial result Z of the character wheel number recognition N ,Z N-1 ,...,Z1 correction value, according to the following rules to correct the first bit (least significant bit) to the Nth bit (most significant bit) in sequence:

[0015] For Z1, if Z1 is a number, then Z'1 = Z1; otherwise, Z'1 = f u (Z1);

[0016] For Z i (i=2,3,...,N), if Z i is a number, then Z' i =Z i ; If Z i Is a letter and Z' i-1 If it is not 0, then Z' i =f u (Z i ); if Z i Is a letter and Z' i-1 is 0, then Z' i =f d (Z i );

[0017] Among them, f u (Z) and f d (Z) are two integer functions, and the domain of the two functions is: {A, B, C, ..., J}. In particular: f u (A) = 0, f u (B) = 1, f u (C) = 2, f u 9D)=3,f u9E)=4,f u 9F)=5,f u (G) = 6, f u (H)=7,f u (I) = 8, f u (J) = 9; f d (A) = 1, f d (B) = 2, f d (C) = 3, f d (D) = 4, f d (E) = 5, f d (F) = 6, f d (G) = 7, f d (H)=8,f d (I) = 9, f d (J)=0.

[0018] In the above technical solution, the method for identifying the water meter pointer reading in step S3 adopts the projection method, which is specifically:

[0019] At the center point P of the pointer area image c On a circle with the short side of the pointer area detection frame as the diameter and the center as the circle, take 20 points P at intervals of 18 degrees in the clockwise direction. i (i=0,1,...,19), where P0 is located directly above the center of the circle. Calculate the P c To P i The projection of (i=0,1,...,19) (i.e., P c To P i The cumulative sum of all pixels on the point connection line) S i (i=0,1,...,19); find the subscript i corresponding to the minimum value of the projection m , the corresponding initial result of pointer reading recognition is i m / 2; the value range of the pointer reading recognition initial result is {0,0.5,1,1.5,...,9.5}; the initial results of the pointer reading recognition are recorded from high to low as: Y M ,Y M-1 ,...,Y1, where M is the number of digits of the water meter pointer reading.

[0020] In the above technical solution, the method for correcting the water meter pointer reading in step S3 is specifically as follows:

[0021] Let Y' M ,Y' M-1 ,...,Y'1 are the initial results of the character wheel number recognition Y M ,Y M-1,...,Y1 correction value; calibrate the 1st bit (least significant bit) to the Mth bit (highest significant bit) in sequence as follows:

[0022] For the first position, in Indicates rounding down;

[0023] For the jth position (j=2,3,...,M), if Y j is an integer and Y′ j-1 ≥6, then otherwise

[0024] In the above technical solution, step S4 includes:

[0025] Assume that the lowest digit of the digital area reading of the wheel is the unit digit, then combine the result of the digital area of ​​the wheel and the result of the pointer area of ​​the water meter into the final complete reading X of the water meter as follows:

[0026]

[0027] If Z1 is numeric and Y' M ≥6, then X=X-1; further, if the lowest digit of the digital area reading of the character wheel is the tenth digit, then X=X / 10.

[0028] Compared with the prior art, the water meter reading recognition method based on deep learning and prior correction of the present invention has the following beneficial effects:

[0029] 1. Correcting the initial results of the wheel digit recognition, the pointer area recognition, and the combination of digits and pointers can effectively overcome the recognition errors caused by the meshing clearance of the water meter gears and improve the accuracy of the water meter reading recognition results.

[0030] 2. When using the projection method to identify pointer readings, only the projection from the center point to 20 points is calculated, which can greatly reduce the amount of calculation and increase the speed of pointer reading recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 A flow chart of a water meter reading recognition method based on deep learning and prior correction according to the present invention;

[0032] Figure 2 This is a schematic diagram of pointer reading recognition using the projection method of the present invention;

[0033] Figure 3 This is a schematic diagram of specific implementation steps of Example 2 of the present invention;

[0034] Figure 4 This is a schematic diagram of specific implementation steps of Example 3 of the present invention;

[0035] Figure 5 It is a schematic diagram of specific implementation steps of Example 4 of the present invention. DETAILED DESCRIPTION

[0036] The following is a combination of specific implementation cases and attached Figure 1-5 The present invention is further described below, but the present invention is not limited to these embodiments.

[0037] Embodiment 1:

[0038] A water meter reading recognition method based on deep learning and prior correction includes the following steps:

[0039] S1: Collect and annotate a large number of images, and train the water meter wheel digit area detection model, water meter wheel digit recognition model, and water meter pointer area detection model;

[0040] S2: water meter wheel digital area detection, recognition and correction;

[0041] S3: water meter pointer area detection, identification and correction;

[0042] S4: Combining and correcting the water meter wheel digital recognition result and pointer recognition result.

[0043] The specific implementation method of step S1 includes:

[0044] 1.1 Collect a large number of large images: Collect a large amount of water meter image data. For each type of water meter to be identified, collect more than 500 images with different readings.

[0045] 1.2 Image annotation: Rectangular frames are used to mark the positions of the character wheel digital area and the pointer area respectively. The annotation content of the digital recognition result of the character wheel digital area is the reading in the area. The specific reading annotation rules are as follows: if the reading is a complete or basically complete number 0, 1, 2, ..., 9, it is directly marked as the corresponding number 0, 1, 2, ..., 9; and if the character wheel digital reading is between two numbers, that is, between 0-1, between 1-2, between 2-3, between 3-4, between 4-5, between 5-6, between 6-7, between 7-8, between 8-9 and between 9-0, it is marked as A, B, C, D, E, F, G, H, I and J respectively.

[0046] 1.3 Model training: Use the labeled data to train the water meter wheel digital area detection model, the water meter wheel digital recognition model, and the water meter pointer area detection model respectively.

[0047] The specific implementation method of step S2 includes:

[0048] 2.1 Water meter wheel digit area detection: Use the trained water meter wheel digit area detection model to detect the position of the wheel digit area in the input water meter original image, and obtain the four vertex coordinates of the wheel digit area rectangular detection box, which are recorded as: (x n1 ,y n1 ), (x n2 ,y n2 ), (x n3 ,y n3 ), x n4 ,y n4 ).

[0049] 2.2 Segmentation of wheel digit area: According to the four vertex coordinates of the wheel digit area rectangular detection frame detected in 2.1, the wheel digit area image is cut out from the original image.

[0050] 2.3 Water meter wheel digital recognition: Use the trained water meter wheel digital recognition model to perform digital recognition on the intercepted wheel digital area image to obtain the initial recognition result, which is recorded from high to low as: Z N ,Z N-1 ,...,Z1, where N is the number of digits in the water meter wheel area, N is generally 5 or 6, and the value range of Z is {0,1,2,...,9,A,B,C,...,J}.

[0051] 2.4 rounds of digital recognition result correction: Let Z' N ,Z' N-1 ,...,Z'1 is the initial result Z of the character wheel number recognition N ,Z N-1 ,...,Z1 correction value, according to the following rules to correct the first bit (least significant bit) to the Nth bit (most significant bit) in sequence:

[0052] For Z1, if Z1 is a number, then Z'1 = Z1; otherwise, Z'1 = f u (Z1).

[0053] For Z i (i=2,3,...,N), if Z i is a number, then Z' i =Z i ; If Z i Is a letter and Z' i-1 If it is not 0, then Z' i =f u (Z i ); if Z i Is a letter and Z' i-1 is 0, then Z' i =f d (Z i ).

[0054] Among them, f u (Z) and f d (Z) are two integer functions, and the domain of the two functions is: {A, B, C, ..., J}. In particular: f u (A) = 0, f u (B) = 1, f u (C) = 2, f u (D) = 3, f u (E) = 4, f u (F) = 5, f u (G) = 6, f u (H)=7,f u (I) = 8, f u (J) = 9; f d (A) = 1, f d (B) = 2, f d (C) = 3, f d (D) = 4, f d (E) = 5, f d (F) = 6, f d (G) = 7, f d (H)=8,f d (I) = 9, f d (J)=0.

[0055] The specific implementation method of step S3 includes:

[0056] 3.1 Rotation correction: Calculate the angle θ between the long side of the detection frame and the horizontal direction according to the coordinates of the four vertices of the rectangular detection frame of the wheel digit area, and rotate the original image with a rotation angle of θ or -θ, so that the long side of the rectangular detection frame of the wheel digit area in the rotated image is in the horizontal direction.

[0057] 3.2 Water meter pointer area detection: Use the trained water meter pointer area detection model to detect the water meter pointer area and obtain the detection frame coordinates of each water meter pointer area.

[0058] 3.3 Pointer area segmentation: According to the coordinates of the water meter pointer area detection frame, each pointer area image is cut out from the rotation-corrected water meter image.

[0059] 3.4 Projection method to identify pointer readings: For each pointer area image, first convert it into a grayscale image, and then use the projection method to identify the pointer readings. Figure 2 As shown, the specific method is:

[0060] At the center point P of the pointer area image c On a circle with the short side of the pointer area detection frame as the diameter and the center as the circle, take 20 points P at intervals of 18 degrees in the clockwise direction.i (i=0,1,...,19), where P0 is located directly above the center of the circle. Calculate the P c To P i The projection of (i=0,1,...,19) (i.e., P c To P i The cumulative sum of all pixels on the point connection line) S i (i=0,1,...,19). Find the subscript i corresponding to the minimum value of the projection m , the corresponding initial result of pointer reading recognition is i m / 2. The value range of the pointer reading recognition initial result is {0, 0.5, 1, 1.5, ..., 9.5}. The initial results of the pointer reading recognition are recorded from high to low as: Y M ,Y M-1 ,...,Y1, where M is the number of digits of the water meter pointer reading.

[0061] 3.5 Pointer reading correction

[0062] Let Y' M ,Y' M-1 ,...,Y'1 are the initial results of the character wheel number recognition Y M ,Y M-1 ,...,Y1 correction value. Correct the 1st bit (least significant bit) to the Mth bit (most significant bit) in sequence according to the following rules:

[0063] For the first position, in Indicates rounding down;

[0064] For the jth position (j=2,3,...,M), if Y j is an integer and Y j ' -1 ≥6, then otherwise

[0065] The specific implementation method in step S4 includes:

[0066] 4.1 Combination: Assume that the lowest digit of the digital area reading of the wheel is the unit digit, then combine the result of the digital area of ​​the wheel and the result of the pointer area of ​​the water meter into the final complete reading X of the water meter according to the following formula:

[0067]

[0068] 4.2 Correction: If Z1 is a number and Y' M ≥6, then X=X-1; in addition, if the lowest digit of the digital area reading of the character wheel is the tenth digit, then X=X / 10.

[0069] Embodiment 2:

[0070] Take a water meter produced by Ningbo Jinquan Instrument Co., Ltd. as an example, which includes a 5-digit wheel number (the lowest digit is the ones digit) and a 4-digit pointer (style as Figure 3 The specific steps are as follows:

[0071] S1: related model training;

[0072] S2: water meter wheel digital area detection, recognition and correction;

[0073] S3: water meter pointer area detection, identification and correction;

[0074] S4: Combining and correcting the water meter wheel digital recognition result and pointer recognition result.

[0075] The specific implementation method of step S1 includes:

[0076] Collect a large amount of water meter image data. For each type of water meter to be identified, collect more than 500 images of different readings. In this example, collect 1,000 water meter images, of which 700 are used as training sets and 300 are used as test sets.

[0077] For the wheel number area and the pointer area, use rectangular frames to mark the positions respectively. For the number recognition results of the wheel number area, the marking content is the reading in the area. The specific reading marking rules are as follows: if the reading is a complete or substantially complete number 0, 1, 2, ..., 9, it is directly marked as the corresponding number 0, 1, 2, ..., 9; and for the wheel number reading between two numbers, that is, between 0-1, between 1-2, between 2-3, between 3-4, between 4-5, between 5-6, between 6-7, between 7-8, between 8-9 and between 9-0, it is marked as A, B, C, D, E, F, G, H, I and J respectively. In this example, use the PPOCRLabel tool for marking. For the wheel number area, use the rectangular frame to mark the position and the reading at the same time; for the pointer area, only the position needs to be marked.

[0078] Use the labeled data to train the water meter wheel digit area detection model, water meter wheel digit recognition model and water meter pointer area detection model respectively.

[0079] In the embodiment of the present invention, the water meter wheel digit area detection adopts the ch_PP-OCRv4_det training model, the water meter wheel digit recognition adopts the en_PP-OCRv3_rec training model, and the water meter pointer area detection adopts the ch_PP-OCRv3_det training model. The three models are trained respectively using the training tools provided in the PaddleOCR package to obtain the water meter wheel digit area detection model, the water meter wheel digit recognition model, and the water meter pointer area detection model after training.

[0080] The specific implementation method of step S2 includes:

[0081] Water meter wheel digit area detection: Use the trained water meter wheel digit area detection model to detect the position of the wheel digit area in the input water meter original image, and obtain the coordinates of the four vertices of the wheel digit area rectangular detection box, which are recorded as: (x n1 ,y n1 ), (x n2 ,y n2 ), (x n3 ,y n3 ), (x n4 ,y n4 ).

[0082] Wheel digit area segmentation: According to the four vertex coordinates of the detected wheel digit area rectangular detection frame, the wheel digit area image is cut out from the original image.

[0083] Water meter wheel digit recognition: Use the trained water meter wheel digit recognition model to perform digital recognition on the intercepted wheel digit area image to obtain the initial recognition result, which is recorded from high to low as: Z N ,Z N-1 ,...,Z1. In this example, N is 5, and the specific results are: 0, 0, 2, 0, D.

[0084] Correction of character wheel digit recognition result: Set Z' N ,Z' N-1 ,...,Z'1 is the initial result Z of the character wheel number recognition N ,Z N-1 ,...,Z1 correction value, according to the following rules to correct the first bit (least significant bit) to the Nth bit (most significant bit) in sequence:

[0085] For Z1, if Z1 is a number, then Z'1 = Z1; otherwise, Z'1 = f u (Z1).

[0086] For Z i (i=2,3,...,N), if Z i is a number, then Z' i=Z i ; If Z i Is a letter and Z' i-1 If it is not 0, then Z' i =f u (Z i ); if Z i Is a letter and Z' i-1 is 0, then Z' i =f d (Z i ).

[0087] Among them, f u (Z) and f d (Z) are two integer functions, and the domain of the two functions is: {A, B, C, ..., J}. In particular: f u (A) = 0, f u (B) = 1, f u (C) = 2, f u (D) = 3, f u (E) = 4, f u (F) = 5, f u (G) = 6, f u (H)=7,f u (I) = 8, f u (J) = 9; f d (A) = 1, f d (B) = 2, f d (C) = 3, f d (D) = 4, f d (E) = 5, f d (F) = 6, f d (G) = 7, f d (H) = 8, f d (I) = 9, f d (J)=0.

[0088] In the embodiment of the present invention, the character wheel digital recognition correction result is: 0, 0, 2, 0, 4.

[0089] The specific implementation method of step S3 includes:

[0090] Rotation correction: Calculate the angle θ between the long side of the detection frame and the horizontal direction based on the coordinates of the four vertices of the rectangular detection frame of the wheel digit area, and rotate the original image with a rotation angle of θ or -θ, so that the long side of the rectangular detection frame of the wheel digit area in the rotated image is in the horizontal direction.

[0091] Water meter pointer area detection: Use the trained water meter pointer area detection model to detect the water meter pointer area and obtain the detection frame coordinates of each water meter pointer area.

[0092] Pointer area segmentation: According to the coordinates of the water meter pointer area detection frame, each pointer area image is cut out from the rotation-corrected water meter image.

[0093] Projection method to identify pointer readings: For each pointer area image, first convert it into a grayscale image, and then use the projection method to identify the pointer readings. Specific methods (such as Figure 2 shown) is:

[0094] At the center point P of the pointer area image c On a circle with the short side of the pointer area detection frame as the diameter and the center as the circle, take 20 points P at intervals of 18 degrees in the clockwise direction. i (i=0,1,...,19), where P0 is located directly above the center of the circle. Calculate the P c To P i The projection of (i=0,1,...,19) (i.e., P c To P i The cumulative sum of all pixels on the point connection line) S i (i=0,1,...,19). Find the subscript i corresponding to the minimum value of the projection m , the corresponding initial result of pointer reading recognition is i m / 2. The value range of the pointer reading recognition initial result is {0, 0.5, 1, 1.5, ..., 9.5}. The initial results of the pointer reading recognition are recorded from high to low as: Y M ,Y M-1 ,...,Y1, where M is the number of digits of the water meter pointer reading. In this example, M = 4, and the initial results of pointer recognition are: 2, 7, 0, 7.5.

[0095] Pointer reading correction: Set Y' M ,Y' M-1 ,...,Y'1 are the initial results of the character wheel number recognition Y M ,Y M-1 ,...,Y1 correction value. Correct the 1st bit (least significant bit) to the Mth bit (most significant bit) in sequence according to the following rules:

[0096] For the first position, in Indicates rounding down;

[0097] For the jth position (j=2,3,...,M), if Y j is an integer and Y j ' -1 ≥6, then otherwise

[0098] In the embodiment of the present invention, the pointer identification and correction results are: 1, 6, 9, 7 in sequence.

[0099] The specific implementation method of step S4 includes:

[0100] Combination: Assume that the lowest digit of the digital area reading of the wheel is the unit digit, then combine the result of the digital area of ​​the wheel and the result of the pointer area of ​​the water meter into the final complete reading X of the water meter as follows:

[0101]

[0102] In the embodiment of the present invention, the combined result is: 204.1697.

[0103] Correction: If Z1 is numeric and Y' M ≥6, then X=X-1; in addition, if the lowest digit of the digital area of ​​the character wheel is the tenth digit, then X=X / 10. In this example, Y' M =1 and the lowest digit of the digital area reading of the character wheel is the ones digit, so no correction is required. The final result is: 204.1697.

[0104] Embodiment 3:

[0105] The purpose of this embodiment is to further illustrate the implementation method and importance of step 2.4 character wheel digit recognition result correction. Figure 4 As shown, the image of the digital area of ​​the water meter wheel shows that the ones (least significant digit) and tens have been transferred to the position of 0, while the hundreds is currently between 2 and 3 due to the gear meshing clearance. At this time, the initial result of the wheel digital recognition is: 0, 0, C, 0, 0. According to the method described in step 2.4 of the manual, the wheel digital recognition correction result can be obtained as: 0, 0, 3, 0, 0.

[0106] Embodiment 4:

[0107] The purpose of this embodiment is to further illustrate the implementation method of combining and correcting the digital and pointer recognition results in step S4. Figure 5 As shown, the water meter wheel digital recognition correction result is: 6, 2, 0, 0, 3; the pointer recognition correction result is: 9, 5, 1. According to the method of step S4, first assume that the lowest digit of the wheel digital area reading is the unit digit, and combine the water meter wheel digital area result and the pointer area result into the water meter complete reading X, in this example X = 62003.951.

[0108] Since the highest bit of the pointer identification result is 9 (≥6), X=X-1, and X=62002.951;

[0109] Since the lowest digit of the digital area reading of the character wheel in this example is the tenth digit, X=X / 10, and X=6200.2951.

[0110] In the description of the present invention, the term "plurality" refers to two or more than two. Unless otherwise clearly defined, the orientation or positional relationship indicated by the terms "upper" and "lower" is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present invention. The terms "connection", "installation", "fixation", etc. should be understood in a broad sense. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be directly connected or indirectly connected through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0111] In the description of the present invention, the description of the terms "one embodiment", "some embodiments", "specific embodiments", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In the present invention, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0112] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A water meter reading recognition method based on deep learning and prior correction, characterized in that: The steps include: S1: Collect and annotate a large number of images, and train the water meter wheel digital area detection model, water meter wheel digital recognition model, and water meter pointer area detection model; S2: water meter wheel digital area detection, recognition and correction; S3: water meter pointer area detection, identification and correction; S4: Combining and correcting the water meter wheel digital recognition result and pointer recognition result.

2. A water meter reading recognition method based on deep learning and prior correction according to claim 1, characterized in that: In the step S1, if the reading is a complete or substantially complete number 0, 1, 2, ..., 9, it is directly marked as the corresponding number 0, 1, 2, ..., 9; and if the reading of the character wheel number is between two numbers, that is, between 0-1, between 1-2, between 2-3, between 3-4, between 4-5, between 5-6, between 6-7, between 7-8, between 8-9 and between 9-0, it is marked as A, B, C, D, E, F, G, H, I and J respectively.

3. A water meter reading recognition method based on deep learning and prior correction according to claim 1, characterized in that: In step S2, the correction rule for the initial result of water meter wheel digit recognition is: Let Z' N ,Z' N-1 ,...,Z'1 is the initial result Z of the character wheel number recognition N ,Z N-1 ,...,Z1 correction value, according to the following rules to correct the first bit (least significant bit) to the Nth bit (most significant bit) in sequence: For Z1, if Z1 is a number, then Z'1 = Z1; otherwise, Z'1 = f u (Z1); For Z i (i=2,3,...,N), if Z i is a number, then Z' i =Z i ; If Z i Is a letter and Z' i-1 If it is not 0, then Z' i =f u (Z i ); if Z i Is a letter and Z' i-1 is 0, then Z' i =f d (Z i ); Among them, f u (Z) and f d (Z) are two integer functions, and the domain of the two functions is: {A, B, C, ..., J}. In particular: f u (A) = 0, f u (B) = 1, f u (C) = 2, f u (D) = 3, f u (E) = 4, f u (F) = 5, f u (G) = 6, f u (H)=7,f u (I) = 8, f u (J) = 9; f d (A) = 1, f d (B) = 2, f d (C) = 3, f d (D) = 4, f d (E) = 5, f d (F) = 6, f d (G) = 7, f d (H) = 8, f d (I) = 9, f d (J)=0.

4. The water meter reading recognition method based on deep learning and prior correction according to claim 1 is characterized in that: In step S3, the method for identifying the water meter pointer reading adopts the projection method, which is specifically: At the center point P of the pointer area image c On a circle with the short side of the pointer area detection frame as the diameter and the center as the circle, take 20 points P at intervals of 18 degrees in the clockwise direction. i (i=0,1,...,19), where P0 is located directly above the center of the circle. Calculate the P c To P i The projection of (i=0,1,...,19) (i.e., P c To P i The cumulative sum of all pixels on the point connection line) S i (i=0,1,...,19); find the subscript i corresponding to the minimum value of the projection m , the corresponding initial result of pointer reading recognition is i m / 2; the value range of the pointer reading recognition initial result is {0,0.5,1,1.5,...,9.5}; the initial results of the pointer reading recognition are recorded from high to low as: Y M ,Y M-1 ,...,Y1, where M is the number of digits of the water meter pointer reading.

5. The water meter reading recognition method based on deep learning and prior correction according to claim 1 is characterized in that: The method for correcting the water meter pointer reading in step S3 is specifically as follows: Let Y' M ,Y' M-1 ,...,Y'1 are the initial results of the character wheel number recognition Y M ,Y M-1 ,...,Y1 correction value; calibrate the 1st bit (least significant bit) to the Mth bit (highest significant bit) in sequence as follows: For the first position, in Indicates rounding down; For the jth position (j=2,3,...,M), if Y j is an integer and Y j ' -1 ≥6, then otherwise 6. A water meter reading recognition method based on deep learning and prior correction according to claim 1, characterized in that: The step S4 comprises: Assume that the lowest digit of the digital area reading of the wheel is the unit digit, then combine the result of the digital area of ​​the wheel and the result of the pointer area of ​​the water meter into the final complete reading X of the water meter as follows: If Z1 is numeric and Y' M ≥6, then X=X-1; further, if the lowest digit of the digital area reading of the character wheel is the tenth digit, then X=X / 10.