Substation pointer type meter reading method and system based on computer vision

Abstract

The invention discloses a transformer substation pointer type meter reading method and system based on computer vision. The method comprises the steps that a transformer substation pointer type meter image is acquired; performing basic information manual calibration and configuration on meters in the transformer substation pointer type meter image to obtain transformer substation pointer type meter configuration information; the substation pointer type meter configuration information comprises meter types, N groups of scale rays, M pointers needing to be read and F groups of reading demand information; and according to the substation pointer type meter configuration information, optimizing a target detection model based on a batch gradient descent algorithm, and performing intelligent detection and angle mapping on the substation pointer type meter image to obtain a reading value. According to the invention, the universality and precision of reading are improved, and the requirement on image quality is reduced.

Description

Technical Field

[0001] The present invention relates to the field of digital image processing, and in particular to a method and system for reading pointer-type meters in a transformer substation based on computer vision. Background Art

[0002] In the fields of digital image processing, machine learning, deep learning, and computer vision, automatic reading of pointer-type meters is a long-standing challenge. These challenges mainly come from the uncontrollability of image quality (such as blurred range, out-of-focus), the difficulty in accurately correcting the tilt of the shooting angle, the diversity of meter types (a single dial contains one or more ranges), and the complexity of actual application scenarios. Existing technical solutions are mainly divided into two categories:

[0003] Using digital image processing algorithms, such as image binarization and polar coordinate transformation, the meter pointer, scale and other information in the meter image are extracted. This information is used to build an angle-scale model for the dial, and the reading is obtained by mapping the pointer angle to the corresponding scale. This type of solution can be applied to meter reading scenarios with controllable image quality, a single meter type, and little interference information, but its versatility is poor.

[0004] The dial is located through deep learning detection technology, and the scale and pointer are located through key point detection technology. The above information is then filtered and classified according to the specific meter type, and finally the reading is inferred by judging the relationship between the pointer key points and the scale key points. This type of solution also has high requirements for image quality and the number of images used to train the model. The key points are not accurately located, the fault tolerance rate is low, and it cannot handle meter scenes with poor image quality such as tilt, blur and occlusion. The algorithm training cost is high and the versatility is greatly limited.

[0005] For example, the prior art with application number "CN202310591944" and name "Lightweight pointer meter reading method based on multi-task learning" predicts readings through a two-level detection model and direct regression. However, it is difficult to implement and requires a large amount of samples. Moreover, for meters with scale information, key point information is unreliable. The reason is that the key point requires that the position of the inspected point itself must be specific and independent, such as corner point features, and there are a large number of repeated areas in the meter, and the optimization process of the key point model is prone to self-contradictions.

[0006] Therefore, the reading method of the existing pointer type meter has the following defects:

[0007] (1) Poor versatility: Existing technologies are often limited to algorithm design for meters of specific types or conditions, and are difficult to fully adapt to a variety of different types of meters in substations, such as oil temperature meters, pressure gauges, leakage current meters, and lightning arrester counters. These meters have their own characteristics. For example, some single dials contain one or more scales, some scales intersect with each other, and there are even complex concentric circle scale layouts. These scales may cover a range of less than 180 degrees, between 180 degrees and 360 degrees, or a full 360 degrees. These diverse design features pose a severe challenge to existing technologies, making it difficult for the technology to respond flexibly and achieve full compatibility and accurate identification.

[0008] (2) High requirements for image quality: Existing technologies generally have high requirements for image quality, usually requiring that the meter must be photographed directly, or relying on the forward azimuth information provided by the ball camera for meter correction. However, in the actual operating environment of the substation, image quality often faces many challenges. These challenges include but are not limited to: image tilt and defocusing that may occur when the ball camera is shooting, and obstruction caused by obstacles; in addition, there are overexposure or dark conditions caused by improper light control. To make it more complicated, substations have a long history of construction and the meters have been in use for a long time, which makes it easy for dirt and oil to accumulate on the surface of the dial, scratches to appear, and the dial range scale to wear and tear. These practical problems often exceed the coverage capabilities of existing technologies, making it difficult to guarantee image quality.

[0009] (3) High maintenance cost: Existing technologies often need to be compatible with different substation sites (whether new or old facilities), and require frequent updates of algorithm logic and adjustment of models to adapt to different meters and monitoring scenarios. This continuous customization has pushed up the cost of algorithm development and maintenance, and significantly extended the system deployment and optimization cycle. For example, due to aging equipment, uneven wear of dial scales, and fluctuations in image quality caused by environmental factors, technicians need to frequently adjust image preprocessing algorithms and recognition models to ensure stable operation of the system. This has increased the burden on the operation and maintenance team and increased the overall operating costs, becoming a pain point in the intelligent transformation and operation and maintenance management of substations. Summary of the invention

[0010] The purpose of the present invention is to provide a method and system for reading pointer-type meters in a substation based on computer vision, which improves the versatility and accuracy of readings and reduces the requirements for image quality by combining manual configuration and deep learning models. The present invention solves the problems of fuzzy image quality, poor versatility, and low reading accuracy in the existing substation meter recognition.

[0011] The present invention is achieved through the following technical solutions:

[0012] In a first aspect, the present invention provides a method for reading pointer-type meters in a substation based on computer vision, the method comprising:

[0013] Obtain the image of pointer type meter of substation;

[0014] Manually calibrate and configure the basic information of the meters in the substation pointer-type meter image to obtain the substation pointer-type meter configuration information; the substation pointer-type meter configuration information includes the meter type, N groups of scale rays, M pointers to be read, and F groups of reading requirement information;

[0015] According to the configuration information of pointer-type meters in substations, the target detection model is optimized based on the batch gradient descent algorithm, and the images of pointer-type meters in substations are intelligently detected and angle mapped to obtain the reading values.

[0016] Furthermore, the basic information of the meters in the substation pointer type meter image is manually calibrated and configured to obtain the substation pointer type meter configuration information, including:

[0017] According to the substation pointer meter image, record the meter type; meter types include oil temperature meter, pressure gauge, leakage current meter and lightning protection counter;

[0018] Manual calibration is performed on various types of dials according to the meter type; manual calibration includes calibrating the dial type, calibrating the start and end rays, calibrating the uneven scale rays and recording the scale, and recording the number of pointers and characteristic numbers;

[0019] After the manual calibration is completed, the meter dial with F group reading requirements is uniformly configured, and each group is configured separately from left to right and from top to bottom according to the preset rules.

[0020] Furthermore, various types of dials are manually calibrated according to the meter type, including:

[0021] Determine the dial type, and calibrate the start ray and the end ray according to the dial type; the dial types include less than 180 degrees, greater than 180 degrees, less than 360 degrees and 360 degrees;

[0022] For each dial type, calibrate N-2 groups of uneven scale rays and record the scales corresponding to the N groups of rays;

[0023] Determine the reverse direction of the x-axis of the image coordinate system as the 0 degree angle reference, and record the angles in a clockwise increasing direction and a counterclockwise decreasing direction;

[0024] Record the number M of pointers of the current meter and the characteristic number of the pointer to be read, and assign pointer characteristic numbers 0, 2, 4, and 6 to the longest characteristic, shortest characteristic, maximum width characteristic, and minimum reading characteristic of the pointer, respectively.

[0025] Furthermore, for each dial type, N-2 groups of uneven scale rays are calibrated, and the scales corresponding to the N groups of rays are recorded, including:

[0026] For each dial type, determine the number of scale ray groups N that need to be calibrated;

[0027] Between the starting ray and the ending ray, calibrate N-2 groups of uneven scale rays;

[0028] And record the scales corresponding to N groups of rays.

[0029] Furthermore, according to the configuration information of the pointer type meter of the substation, the target detection model is optimized based on the batch gradient descent algorithm, and the substation pointer type meter image is intelligently detected and angle mapped to obtain the reading value, including:

[0030] Meter detection: According to the configuration information of the pointer-type meters in the substation, the target detection model is optimized based on the batch gradient descent algorithm to detect the needle of the pointer-type meter image in the substation;

[0031] Pointer and circle center detection: According to the configuration information of the pointer type meter of the substation, the 360-degree random sampling method is used to add data of different rotation angles to the training set, and the pointer and circle center of the substation pointer type meter image are detected;

[0032] Calculate the range interval: Combine the configuration information of the pointer meter of the substation, sort the pointer targets according to the confidence level, obtain the required number of pointers from high to low, map the angle of each pointer to [-360,360], and determine its range interval;

[0033] Calculate reading: Map the pointer angle to the reading based on the calculated scale value of the range interval.

[0034] Furthermore, the needle detection of the pointer type meter image of the substation is performed, which is divided into a first training stage and a first reasoning stage; wherein:

[0035] The first training stage: collect the pointer-type meter images of the substation, and use the external rectangular frame to mark the meter location and category; according to the characteristics of the substation inspection system, optimize the target detection model based on the batch gradient descent algorithm, set the specific parameters of the cosine annealing strategy, and conduct model training;

[0036] The first reasoning stage: input the inspection image and output the location and category of the bounding rectangle of the meter target.

[0037] Furthermore, the pointer and center detection of the pointer type meter image of the substation is performed, which is divided into a second training stage and a second reasoning stage; wherein:

[0038] The second training stage: collect substation pointer meter images, and use a rotating circumscribed rectangular frame to mark the pointer and center position and category; according to the characteristics of the substation meter points, use the 360-degree random sampling method to add data with different rotation angles to the training set, with the angle as a continuous value and each degree as a category, for a total of 180 categories;

[0039] The second reasoning stage: input the meter area image, output the position and category of the rotated circumscribed rectangular box of the pointer target, and the rotated circumscribed rectangular box is represented by an ellipse equation, namely the center, angle, length and width.

[0040] Furthermore, the angle classification in the second training stage generates labels based on the Gaussian function, making the generation of labels smoother. In the circular smooth label CSL, the angle range is set to [0°, 179°], and the expression is:

[0041]

[0042] Where: μ is the mean value in the Gaussian function, which represents the position of the target center angle; r represents the radius of the Gaussian window, which is the window radius of the annular smooth label.

[0043] In a second aspect, the present invention further provides a substation pointer meter reading system based on computer vision, the system comprising:

[0044] An acquisition unit, used for acquiring an image of a pointer-type meter in a substation;

[0045] The calibration and configuration unit is used to manually calibrate and configure the basic information of the meter in the substation pointer meter image to obtain the configuration information of the substation pointer meter; the substation pointer meter configuration information includes the meter type, N groups of scale rays, M pointers to be read, and F groups of reading requirement information;

[0046] The detection and angle mapping unit is used to optimize the target detection model based on the batch gradient descent algorithm according to the configuration information of the pointer-type meter of the substation, perform intelligent detection and angle mapping on the image of the pointer-type meter of the substation, and obtain the reading value.

[0047] Furthermore, the calibration and configuration unit includes:

[0048] The meter type recording subunit is used to record the meter type according to the substation pointer meter image; the meter types include oil temperature meter, pressure gauge, leakage current meter and lightning protection counter;

[0049] The calibration subunit is used to manually calibrate various types of dials according to the type of meter; the manual calibration includes calibrating the dial type, calibrating the start and end rays, calibrating the uneven scale rays and recording the scale, and recording the number of pointers and characteristic numbers;

[0050] The configuration subunit is used to uniformly configure the situations where there are F groups of reading requirements on the dial of the meter after manual calibration is completed. Each group is configured separately from left to right and from top to bottom according to preset rules.

[0051] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0052] 1. The present invention provides a method and system for reading pointer-type meters in substations based on computer vision. By combining manual configuration and deep learning models, the versatility and accuracy of readings are improved, and the requirements for image quality are reduced. The present invention solves the problems of fuzzy image quality, poor versatility, and low reading accuracy in existing substation meter recognition.

[0053] 2. The present invention provides a method and system for reading pointer-type meters in a substation based on computer vision, which has good versatility: the present invention can be applied to all pointer-type meters, regardless of the specific model, appearance, pointer, dial type, etc. of the meter. As long as the pointer rotates around the center, this solution can cover it.

[0054] 3. The present invention provides a method and system for reading pointer-type meters in a substation based on computer vision, with high reading accuracy: the regional scale configuration method of the present invention can solve the problem of large reading errors caused by uneven scales of all types. The factors that cause uneven scales include the meter's own design, tilted shooting, etc. As long as the scale is uniform within the minimum configured interval; and the second stage of the rotating target detection model of the present invention, using the relationship between the center of the circle and the pointer, can accurately predict the quadrant and angle information of the pointer in the current range, so that the prediction error of the angle is 0. The above two points ensure that the angle and the scale correspond accurately, thereby improving the reading accuracy.

[0055] 4. The present invention provides a method and system for reading pointer-type meters in a substation based on computer vision, which has low requirements for image quality: since the scale-related information comes from manual configuration, only the pointer needs to be clearly visible, and the reading is not disturbed. For common image quality problems, such as blur, occlusion, too dark, overexposure, etc., even if the range is unclear or even blocked, as long as it does not affect the pointer area, the reading will not be affected.

[0056] 5. The present invention provides a method and system for reading pointer-type meters in a substation based on computer vision, which has low maintenance cost: the present invention obtains fixed information in the inspection point by manual configuration, and obtains information about changes in the point by detection model (meter detection, pointer detection). According to the characteristic that the inspection point is basically unchanged relative to the target viewing angle of the shooting, this scheme does not involve the use of any image absolute coordinate information, so it is compatible with image changes such as translation and scaling. As long as no rotation occurs, the manually configured information can be maintenance-free. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0058] Figure 1 This is a flow chart of a method for reading pointer-type meters in a substation based on computer vision according to the present invention;

[0059] Figure 2 It is a schematic diagram of the pointer rotation center, measuring range scale range, and measuring range angle range of the recording meter of the present invention;

[0060] Figure 3 It is a schematic diagram of multiple groups of readings in the disk of the present invention;

[0061] Figure 4 The original image and the enhanced image of the present invention, the upper part is the original image, and the lower part is the various enhanced images;

[0062] Figure 5 The present invention is a structural block diagram of a pointer-type meter reading system for a substation based on computer vision. DETAILED DESCRIPTION

[0063] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0064] The present invention aims to solve the problems existing in the existing substation meter identification, such as fuzzy quality of ball camera collected images, poor versatility, low reading accuracy, etc. The method of the present invention adopts a deep learning model that is more suitable for meter and pointer detection. There is no ambiguity in the model optimization process, so ideal accuracy is achieved. Specifically, step 1, the basic information of the meter in the substation pointer meter image screen is manually calibrated and configured to obtain the type of meter, N groups of scale rays, M pointers to be read, and F groups of reading requirements; according to the configuration information, the substation pointer meter image is intelligently detected and angle mapped to obtain accurate reading values.

[0065] The present invention is applicable to all pointer-type meters, regardless of the specific model, appearance, pointer, dial type, etc. of the meter. As long as the mode of the pointer rotating around the center is met, the present solution can cover it.

[0066] Example 1

[0067] like Figure 1 As shown, the present invention provides a method for reading pointer-type meters in a substation based on computer vision, the method comprising:

[0068] Step 1, obtaining a pointer meter image of a substation;

[0069] Step 2, manually calibrate and configure the basic information of the meter in the substation pointer meter image to obtain the substation pointer meter configuration information; the substation pointer meter configuration information includes the meter type, N groups of scale rays, M pointers to be read, and F groups of reading requirement information;

[0070] Step 2 specifically includes:

[0071] Step 2.1: Record the meter types according to the substation pointer meter images; meter types include oil temperature meter, pressure gauge, leakage current meter and lightning arrester counter;

[0072] Step 2.2: Manually calibrate various types of dials according to the meter type; manual calibration includes calibrating the dial type, calibrating the start and end rays, calibrating the uneven scale rays and recording the scale, and recording the number of pointers and characteristic numbers;

[0073] In this embodiment, step 2.2 specifically includes the following sub-steps:

[0074] Step 2.2.1: Determine the dial type, and calibrate the start ray and the end ray according to the dial type; the dial types include less than 180 degrees, greater than 180 degrees, less than 360 degrees, and 360 degrees;

[0075] Step 2.2.2: For each dial type, calibrate N-2 groups of uneven scale rays and record the scales corresponding to the N groups of rays;

[0076] Step 2.2.3: Determine the opposite direction of the x-axis of the image coordinate system as the 0 degree angle reference, and record the angles according to the rule of increasing angles in the clockwise direction and decreasing angles in the counterclockwise direction;

[0077] Step 2.2.4: Record the number of pointers M of the current meter and the characteristic number of the pointer to be read, and assign pointer characteristic numbers 0, 2, 4, and 6 to the longest characteristic, shortest characteristic, maximum width characteristic, and minimum reading characteristic of the pointer, respectively.

[0078] In the above technical solution, step 2.2 calibrates two groups of basic rays, namely the starting ray and the ending ray, for various dial types (such as less than 180 degrees, greater than 180 degrees and less than 360 degrees, 360 degrees, etc.). Calibrate (N-2) groups of uneven scale rays (such as Figure 2 ), and record the scales corresponding to N groups of rays. In this process, we define the opposite direction of the x-axis of the image coordinate system as the 0-degree angle reference, follow the principle of increasing angles in the clockwise direction and decreasing angles in the counterclockwise direction, and the starting ray angle must be smaller than the ending ray angle, so the angle range is between [-360, 360] degrees (such as Figure 3, with picture "SF2"), and record the number M of the pointers of the current meter and the characteristic number of the pointer to be read (the longest needle code is 0, the shortest needle code is 2, the maximum width pointer code is 4, and the minimum reading pointer code is 6).

[0079] Specifically, step 2.2.2 is as follows:

[0080] Step 2.2.2.1, for each dial type, determine the number of scale ray groups N that need to be calibrated;

[0081] Step 2.2.2.2, calibrate N-2 groups of uneven scale rays between the starting ray and the ending ray;

[0082] Step 2.2.2.3, and record the scales corresponding to N groups of rays.

[0083] Step 2.3: After the manual calibration is completed, the situations where there are F groups of reading requirements in the dial of the meter are uniformly configured, that is, if there are F groups of reading requirements in the unified dial, each group is configured separately from left to right and from top to bottom according to the preset rules.

[0084] The present invention only needs to draw M rays on the current meter and match the corresponding range readings and meter types. The configuration is simple. The traditional configuration needs to match various ranges and various readings, and requires the numbers on the dial to be as clear and visible as possible without tilting. The present invention only requires that the pointer in the current dial is visible and the dial is not too tilted (such as tilting more than 80 degrees, which can also be reflected in the diagrams in the subsequent configuration).

[0085] Step 3: According to the configuration information of the pointer-type meter of the substation, the target detection model is optimized based on the batch gradient descent algorithm, and the pointer-type meter image of the substation is intelligently detected and angle mapped to obtain accurate reading values.

[0086] Step 3 specifically includes:

[0087] Step 3.1: Meter detection: According to the configuration information of the pointer-type meter in the substation, the target detection model is optimized based on the batch gradient descent algorithm to detect the needle of the pointer-type meter image in the substation;

[0088] In this embodiment, the needle detection of the pointer type meter image of the substation is performed, which is divided into a first training stage and a first reasoning stage; wherein:

[0089] The first training stage: collect the images of pointer-type meters in the substation, and use an external rectangular frame to mark the meter position and category; according to the characteristics of the substation inspection system, most of the meter points are fixed-point inspections, but different substation meters are rich in types and have variable ranges. Accurately locating the meter position and meter type is an important capability that the current model needs to possess. The present invention optimizes the target detection model through the batch gradient descent algorithm (Batch Gradient Descent, BGD), sets the specific parameters of the cosine annealing strategy (such as the initial learning rate, minimum learning rate, maximum number of iterations, etc.) according to the data set, and more effectively controls the learning step size during the training process to improve the training effect and convergence speed of the model, and improves the generalization ability of the model, and also ensures that it can accurately and efficiently identify and locate various types of meters when facing complex scenarios of different substations, so that the model performance meets expectations;

[0090] In the first training stage of the present invention, the original image and the enhanced image are as follows: Figure 4 shown.

[0091] The first reasoning stage: input the inspection image and output the location and category of the bounding rectangle of the meter target.

[0092] Step 3.2: Pointer and circle center detection: According to the configuration information of the pointer type meter of the substation, the 360-degree random sampling method is used to add data of different rotation angles to the training set, and the pointer and circle center of the substation pointer type meter image are detected;

[0093] In this embodiment, the pointer and center detection of the pointer type meter image of the substation is performed, which is divided into a second training stage and a second reasoning stage; wherein:

[0094] The second training stage: collect the images of pointer-type meters in the substation, and use a rotating circumscribed rectangular frame to mark the position and category of the pointer and the center of the circle; in view of the fact that most meters in the substation meter points are in a fixed range for a long time and in a rotating state, the present invention proposes an innovative optimization strategy. In order to prevent the pointer detection from over-relying on the background information of the dial and causing overfitting, we abandon the traditional rectangular frame positioning method and effectively decouple the pointer information and the dial range information. This enables the rotating target detection model to focus more on the identification of the pointer itself and its angle information. During the training process, the present invention uses a 360-degree random sampling method to randomly increase data of different rotation angles in the training set. In terms of the model, we use the deep learning model YOLO, which performs well in natural image or object detection tasks, as the starting point of our model training, and convert the output angle θ of the network from a regression problem to a classification problem of 0-180 degrees, effectively avoiding the boundary problem in the regression task. We have effectively enhanced the generalization ability of the model from sample addition to model modification. Even when we face an unseen dial type, the fully trained model can still accurately locate the position of the pointer and predict the angle. The network output angle is converted from the regression problem to the classification problem of 0-180 degrees. According to the circular smooth label

[0095] Application of CircularSmoothLabel (CSL), in deep learning multi-classification problems, traditionally one-hot encoding is often used to label data. However, in angle multi-classification problems, one-hot encoding can lead to data sparsity problems. To solve this problem, we generate labels based on Gaussian functions to make label generation smoother. In CSL, the angle range is set to [0°, 179°], and the specific definition is as follows:

[0096]

[0097] Where: μ is the mean value in the Gaussian function, which represents the position of the target center angle; r represents the radius of the Gaussian window, which is the window radius of the annular smooth label.

[0098] The second reasoning stage: input the meter area image, output the position and category of the rotated circumscribed rectangular box of the pointer target, and the rotated circumscribed rectangular box is represented by an ellipse equation, namely the center, angle, length and width.

[0099] Step 3.3: Calculate the range interval: Combined with the configuration information of the pointer meter of the substation, sort the pointer targets according to the confidence level, obtain the required number of pointers from high to low, map the angle of each pointer to [-360, 360], and determine its range interval;

[0100] Step 3.4: Calculate the reading: Map the pointer angle to the reading based on the calculated scale value of the range interval.

[0101] The detailed process of step 3.4 is as follows:

[0102] Assume that the number of configured rays is n, and the angle of the ray from the start to the end is a i ,i∈[0,n), scale value is v i ,i∈[0,n).

[0103] The angle value output by the algorithm is a∈[0,360]

[0104] If the meter scale is not 360 degrees, the calculation steps are:

[0105] Step 1: According to a 0 Re-determine the sign of a,

[0106] If a 0 >=0:

[0107] a=a

[0108] If a 0 <0:

[0109] First determine which end of the range the pointer is closer to;

[0110] The difference between the pointer and the end angle is dist n =|a n -a|;

[0111] The difference between the pointer and the starting angle dist 0 =|(a-360)-a 0 |;

[0112] Determine the sign of a:

[0113]

[0114] Step 2: Determine which two ray angle intervals a is within and calculate the readings;

[0115] If a∈[a 0 , a n ), and a is located in ∈[a j , a j+1 ], and its corresponding interval of scale value is [v j , v j+1 ]

[0116]

[0117] if It means the pointer points out of the range, then:

[0118]

[0119] If the meter scale is 360 degrees, the calculation steps are:

[0120] Step 1: Determine the sign of a:

[0121] a 0 ′=360+a 0

[0122]

[0123] Step 2: Determine which two ray angle intervals a is in and calculate the reading. a is located in ∈[a j , a j+1 ], and its corresponding interval of scale value is [v j , v j+1 ];

[0124]

[0125] Example 2

[0126] like Figure 5 As shown, the difference between this embodiment and embodiment 1 is that this embodiment provides a substation pointer meter reading system based on computer vision, and the system corresponds to the substation pointer meter reading method based on computer vision in embodiment 1 in a one-to-one manner; the system includes:

[0127] An acquisition unit, used for acquiring an image of a pointer-type meter in a substation;

[0128] The calibration and configuration unit is used to manually calibrate and configure the basic information of the meter in the substation pointer meter image to obtain the configuration information of the substation pointer meter; the substation pointer meter configuration information includes the meter type, N groups of scale rays, M pointers to be read, and F groups of reading requirement information;

[0129] The detection and angle mapping unit is used to optimize the target detection model based on the batch gradient descent algorithm according to the configuration information of the pointer-type meter of the substation, perform intelligent detection and angle mapping on the image of the pointer-type meter of the substation, and obtain accurate reading values.

[0130] As a further implementation, the calibration and configuration unit includes:

[0131] The meter type recording subunit is used to record the meter type according to the substation pointer meter image; the meter types include oil temperature meter, pressure gauge, leakage current meter and lightning protection counter;

[0132] The calibration subunit is used to manually calibrate various types of dials according to the type of meter; the manual calibration includes calibrating the dial type, calibrating the start and end rays, calibrating the uneven scale rays and recording the scale, and recording the number of pointers and characteristic numbers;

[0133] The configuration subunit is used to uniformly configure the situations where there are F groups of reading requirements on the dial of the meter after manual calibration is completed. Each group is configured separately from left to right and from top to bottom according to preset rules.

[0134] Among them, the execution process of each unit can be executed according to the process steps of a method for reading pointer-type meters in a substation based on computer vision in Example 1, and will not be described one by one in this embodiment.

[0135] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0136] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0137] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0138] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0139] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for reading pointer-type meters in a substation based on computer vision, characterized in that: The method includes: Obtain the image of pointer type meter of substation; Manually calibrate and configure basic information of the meter in the substation pointer meter image to obtain substation pointer meter configuration information; the substation pointer meter configuration information includes meter type, N groups of scale rays, M pointers to be read, and F groups of reading requirement information; According to the configuration information of the pointer-type meter of the substation, the target detection model is optimized based on the batch gradient descent algorithm, and the image of the pointer-type meter of the substation is intelligently detected and angle mapped to obtain a reading value.

2. A method for reading pointer type meters in a substation based on computer vision according to claim 1, characterized in that: Manually calibrate and configure basic information of the meters in the substation pointer-type meter image to obtain configuration information of the substation pointer-type meter, including: According to the substation pointer meter image, record the meter type; the meter type includes an oil temperature meter, a pressure gauge, a leakage current meter and a lightning protection counter; Manually calibrate various types of dials according to the meter type; the manual calibration includes calibrating the dial type, calibrating the start and end rays, calibrating the uneven scale rays and recording the scale, and recording the number of pointers and characteristic numbers; After the manual calibration is completed, the meter dial with F group reading requirements is uniformly configured, and each group is configured separately from left to right and from top to bottom according to the preset rules.

3. A method for reading pointer type meters in a substation based on computer vision according to claim 2, characterized in that: Manual calibration of various types of dials is performed, depending on the meter type, including: Determine the dial type, and calibrate the starting ray and the ending ray according to the dial type; the dial type includes less than 180 degrees, greater than 180 degrees, less than 360 degrees and 360 degrees; For each dial type, calibrate N-2 groups of uneven scale rays and record the scales corresponding to the N groups of rays; Determine the reverse direction of the x-axis of the image coordinate system as the 0 degree angle reference, and record the angles in a clockwise increasing direction and a counterclockwise decreasing direction; Record the number M of pointers of the current meter and the characteristic number of the pointer to be read, and assign pointer characteristic numbers 0, 2, 4, and 6 to the longest characteristic, shortest characteristic, maximum width characteristic, and minimum reading characteristic of the pointer, respectively.

4. A method for reading pointer type meters in a substation based on computer vision according to claim 3, characterized in that: For each dial type, calibrate N-2 groups of uneven scale rays and record the scales corresponding to the N groups of rays, including: For each dial type, determine the number of scale ray groups N that need to be calibrated; Between the starting ray and the ending ray, calibrate N-2 groups of uneven scale rays; And record the scales corresponding to N groups of rays.

5. A method for reading pointer type meters in a substation based on computer vision according to claim 1, characterized in that: According to the configuration information of the pointer type meter of the substation, the target detection model is optimized based on the batch gradient descent algorithm, and the image of the pointer type meter of the substation is intelligently detected and angle mapped to obtain the reading value, including: Meter detection: According to the configuration information of the pointer-type meter of the substation, the target detection model is optimized based on the batch gradient descent algorithm, and the needle of the pointer-type meter image of the substation is detected; Pointer and circle center detection: According to the configuration information of the pointer type meter of the substation, a 360-degree random sampling method is used to add data of different rotation angles to the training set, and the pointer and circle center of the pointer type meter image of the substation are detected; Calculate the range interval: Combine the configuration information of the pointer type meter of the substation, sort the pointer targets according to the confidence level, obtain the required number of pointers from high to low, map the angle of each pointer to [-360, 360], and determine the range interval in which it is located; Calculate reading: Map the pointer angle to the reading based on the calculated scale value of the range interval.

6. A method for reading pointer type meters in a substation based on computer vision according to claim 5, characterized in that: The needle detection of the pointer type meter image of the substation is performed, which is divided into a first training stage and a first reasoning stage; wherein: The first training stage: collect the pointer-type meter images of the substation, and use the external rectangular frame to mark the meter location and category; according to the characteristics of the substation inspection system, optimize the target detection model based on the batch gradient descent algorithm, set the specific parameters of the cosine annealing strategy, and conduct model training; The first reasoning stage: input the inspection image and output the location and category of the bounding rectangle of the meter target.

7. A method for reading pointer type meters in a substation based on computer vision according to claim 5, characterized in that: The pointer and center detection of the pointer type meter image of the substation is performed, which is divided into a second training stage and a second reasoning stage; wherein: The second training stage: collect substation pointer meter images, and use a rotating circumscribed rectangular frame to mark the pointer and center position and category; according to the characteristics of the substation meter points, use the 360-degree random sampling method to add data with different rotation angles to the training set, with the angle as a continuous value and each degree as a category, for a total of 180 categories; The second reasoning stage: input the meter area image, output the position and category of the rotated circumscribed rectangular box of the pointer target, and the rotated circumscribed rectangular box is represented by an ellipse equation, namely the center, angle, length and width.

8. A method for reading pointer type meters in a substation based on computer vision according to claim 7, characterized in that: The angle classification in the second training stage generates labels based on the Gaussian function. In the circular smooth label CSL, the angle range is set to [0°, 179°], and the expression is: Where: μ is the mean value in the Gaussian function, which represents the position of the target center angle; r represents the radius of the Gaussian window, which is the window radius of the annular smooth label.

9. A computer vision-based substation pointer meter reading system, characterized in that: The system includes: An acquisition unit, used for acquiring an image of a pointer-type meter in a substation; A calibration and configuration unit, used to manually calibrate and configure the basic information of the meter in the substation pointer meter image to obtain the substation pointer meter configuration information; the substation pointer meter configuration information includes the meter type, N groups of scale rays, M pointers to be read, and F groups of reading requirement information; The detection and angle mapping unit is used to optimize the target detection model based on the batch gradient descent algorithm according to the configuration information of the substation pointer meter, perform intelligent detection and angle mapping on the substation pointer meter image, and obtain a reading value.

10. A computer vision-based substation pointer meter reading system according to claim 9, characterized in that: The calibration and configuration unit comprises: A meter type recording subunit, used to record the meter type according to the substation pointer meter image; the meter type includes an oil temperature meter, a pressure gauge, a leakage current meter and a lightning protection counter; A calibration subunit, used to manually calibrate various types of dials according to the meter type; the manual calibration includes calibrating the dial type, calibrating the start and end rays, calibrating the uneven scale rays and recording the scale, and recording the number of pointers and characteristic numbers; The configuration subunit is used to uniformly configure the situations where there are F groups of reading requirements on the dial of the meter after manual calibration is completed. Each group is configured separately from left to right and from top to bottom according to preset rules.

Images