Meter data identification method and device, computer equipment and storage medium

Abstract

The invention relates to a meter data identification method and device, computer equipment and a storage medium. The method comprises the following steps: acquiring an original instrument image; inputting the original instrument image into a target detection model to obtain a target instrument image and digital scale information; according to the digital scale information, performing image segmentation on the target meter image to obtain a pointer line mask and a scale line mask; and reading according to the pointer line mask and the scale line mask to obtain a reading result. By adopting the method, the meter data identification accuracy in a complex scene can be improved.

Description

Technical Field

[0001] The present application relates to the field of intelligent manufacturing technology, and in particular to a meter data recognition method, device, computer equipment and storage medium. Background Art

[0002] Pointer meters, as traditional physical measurement devices, are widely used in industrial applications such as power, chemical engineering, and rail transit. They monitor system operating status in real time and ensure equipment safety and stable operation. These meters typically feature a mechanical design and are intuitive, stable, and highly resistant to interference. They are widely used in complex industrial environments, where electronic systems are unstable, or where digital sensors cannot be deployed.

[0003] However, these instruments generally lack digital output interfaces and have long relied on manual inspections to obtain readings, resulting in low data collection efficiency and poor real-time performance. They are also easily affected by factors such as the operator's subjective judgment and observation angle, leading to reading deviations or recording errors.

[0004] With the development of the Industrial Internet and intelligent manufacturing, traditional manual meter reading can no longer meet the intelligent, automated, and structured data collection requirements of modern production systems. Related technologies use image recognition methods to automatically identify pointer-type meters, but most of them can only achieve simple detection or reading extraction, and the recognition accuracy of meters in complex scenarios cannot meet the requirements. Summary of the Invention

[0005] Based on this, it is necessary to provide a meter data recognition method, device, computer equipment and storage medium that can improve the accuracy of meter recognition in complex scenarios to address the above technical problems.

[0006] In a first aspect, the present application provides a meter data identification method, comprising:

[0007] Get the original instrument image;

[0008] Inputting the original instrument image into the target detection model to obtain the target meter image and digital scale information;

[0009] Performing image segmentation on the target meter image according to the digital scale information to obtain a pointer line mask and a scale line mask;

[0010] Readings are performed according to the pointer line mask and the scale line mask to obtain reading results.

[0011] In one embodiment, the target detection model includes a first detection sub-model, a second detection sub-model, and a third detection sub-model; and the training method of the target detection model includes:

[0012] Replacing the convolution module of the original object detection model with a deformable convolution module to obtain a model to be trained; wherein the deformable convolution module includes an offset;

[0013] Acquire a meter data set, and use the meter data set to train the to-be-trained model to obtain a first detection sub-model; wherein the first detection sub-model is used to identify the meter area;

[0014] Acquire a defect data set, and use the defect data set to train the model to be trained to obtain a second detection sub-model;

[0015] A digital data set is obtained, and the model to be trained is trained using the digital data set to obtain a third detection sub-model.

[0016] In one embodiment, inputting the original meter image into a target detection model to obtain a target meter image and digital scale information includes:

[0017] Inputting the original meter image into the first detection sub-model to obtain a meter area image;

[0018] Inputting the meter area image into the second detection sub-model to obtain image defect information;

[0019] performing filtering processing on the meter area image according to the image defect information to obtain a target meter image;

[0020] The target meter image is input into the third detection sub-model to obtain digital scale information.

[0021] In one embodiment, filtering the meter area image according to the image defect information to obtain the target meter image includes:

[0022] When the image defect information indicates that the corresponding meter area image is valid, median filtering is performed on the meter area image to obtain a target meter image.

[0023] In one embodiment, the digital scale information includes digital text position and dial range information; and performing image segmentation on the target meter image based on the digital scale information to obtain a pointer line mask and a scale line mask includes:

[0024] Inputting the digital text position and the dial range information into an image segmentation model to obtain a pointer area mask and a scale mark mask;

[0025] An iterative edge point deletion process is performed on the pointer area mask until the edge points of the pointer area meet a preset condition, thereby obtaining a pointer line mask.

[0026] In one embodiment, the reading according to the pointer line mask and the scale line mask to obtain a reading result includes:

[0027] Determine the straight line where the pointer is located according to the pointer line mask to obtain the pointer straight line;

[0028] Determine scale data and any scale line according to the scale line mask;

[0029] Determine the center of the instrument according to the pointer straight line and the scale straight line;

[0030] Readings are taken according to the pointer straight line, the scale data and the center of the instrument to obtain a reading result.

[0031] In a second aspect, the present application further provides a meter data recognition device, comprising:

[0032] Image acquisition module, used to acquire original instrument images;

[0033] An image recognition module is used to input the original instrument image into a target detection model to obtain a target meter image and digital scale information;

[0034] An image segmentation module, configured to segment the target meter image according to the digital scale information to obtain a pointer line mask and a scale line mask;

[0035] The data processing module is used to perform readings according to the pointer line mask and the scale line mask to obtain reading results.

[0036] In a third aspect, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0037] Get the original instrument image;

[0038] Inputting the original instrument image into the target detection model to obtain the target meter image and digital scale information;

[0039] Performing image segmentation on the target meter image according to the digital scale information to obtain a pointer line mask and a scale line mask;

[0040] Readings are performed according to the pointer line mask and the scale line mask to obtain reading results.

[0041] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the following steps are implemented:

[0042] Get the original instrument image;

[0043] Inputting the original instrument image into the target detection model to obtain the target meter image and digital scale information;

[0044] Performing image segmentation on the target meter image according to the digital scale information to obtain a pointer line mask and a scale line mask;

[0045] Readings are performed according to the pointer line mask and the scale line mask to obtain reading results.

[0046] In a fifth aspect, the present application further provides a computer program product, comprising a computer program, which, when executed by a processor, implements the following steps:

[0047] Get the original instrument image;

[0048] Inputting the original instrument image into the target detection model to obtain the target meter image and digital scale information;

[0049] Performing image segmentation on the target meter image according to the digital scale information to obtain a pointer line mask and a scale line mask;

[0050] Readings are performed according to the pointer line mask and the scale line mask to obtain reading results.

[0051] The above-mentioned meter data recognition method, device, computer equipment, storage medium and computer program product,

[0052] By acquiring the original meter image and inputting it into the target detection model, the meter area is automatically located from the entire image to obtain the target meter image, and the digital scale information related to the reading is extracted. This avoids positioning deviations caused by rough manual cropping or meter misidentification, significantly improving the image input quality of the subsequent recognition process. The target meter image is then segmented according to the digital scale information, and the pointer line mask and scale line mask are extracted respectively, thereby achieving precise separation of the reading elements in the dial structure. The slender pointer structure and dense scale distribution in the meter can still be accurately extracted under complex background interference. Finally, readings are performed based on the above mask information, improving the accuracy of the reading results in complex scenarios such as differences in meter appearance, changes in viewing angle, and partial occlusion. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0054] Figure 1 A diagram showing an application environment of a meter data recognition method in one embodiment;

[0055] Figure 2 1 is a flow chart of a meter data identification method according to an embodiment;

[0056] Figure 3 A schematic diagram of using an angle method to perform readings in a meter data recognition method according to one embodiment;

[0057] Figure 4 1 is a flow chart of a method for training an object detection model in one embodiment;

[0058] Figure 5 A comparison diagram of conventional convolution and deformable convolution in one embodiment;

[0059] Figure 6 It is a structural block diagram of a meter data identification device in one embodiment;

[0060] Figure 7 is a diagram of the internal structure of a computer device in one embodiment;

[0061] Figure 8 FIG. 4 is a diagram showing the internal structure of a computer device in another embodiment. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0063] The meter data recognition method provided in the embodiment of the present application can be applied to Figure 1In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The terminal 102 can be used to obtain the original instrument image and send the original instrument image to the server 104 for processing. The terminal 102 can also be used to receive the recognition result sent by the server 104 and display the recognition result. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other network servers. The data storage system can be used to store the original instrument image and the training data of the target detection model, and can also be used to periodically store the recognition results of the meter data. Among them, the terminal 102 can be various photo-taking or video-recording devices, and can also be but not limited to various personal computers, laptops, smart phones, tablets, Internet of Things devices and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart car-mounted devices, etc. Portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The server 104 can be implemented as an independent server or a server cluster consisting of multiple servers.

[0064] In an exemplary embodiment, Figure 2 As shown, a meter data recognition method is provided, which is applied to Figure 1 The server 104 in the example is used as an example to illustrate the method, which includes the following steps S202 to S208.

[0065] Step S202: Acquire the original instrument image.

[0066] A raw instrument image refers to an image containing at least one complete pointer instrument, captured or recorded directly from an industrial field environment using an image acquisition device. Depending on the data acquisition method, raw instrument images can be categorized as either static or dynamic images. A static image is an image frame captured using a fixed-point camera, such as an industrial camera or high-resolution camera capturing an instrument panel at a specific moment. A dynamic image is a video stream obtained through continuous recording, such as video clips captured during the operation of an inspection robot or drone.

[0067] For example, the server 104 may pre-configure the shooting parameters of the acquisition device to ensure that the acquired image has sufficient resolution and suitable lighting conditions.

[0068] When using the static camera mode, server 104 can directly receive image data sent by the image acquisition terminal. The file format can be .jpg or .png. The image can fully include one or more pointer instrument panels, ensuring that each meter is clearly visible, the edges are complete, and the pointer and scale are not significantly obscured. For example, in a substation monitoring system, a camera can be set to capture multiple meters on site every 30 minutes and transmit the images via the network to server 104 for processing.

[0069] If video capture is used, server 104 can receive video stream files, such as those in .mp4 or .avi formats, and perform video framing according to the frame rate parameters set during capture. Server 104 can flexibly adjust the frame rate parameters based on specific scenario needs. For example, in a scenario where the temperature of a power grid transformer changes slowly, server 104 can set a low frame rate mode of extracting one frame per second to reduce the amount of unnecessary data. In a scenario where the speedometer is monitored in real time during train operation, a high frame rate mode of extracting five or ten frames per second can be set to ensure that the meter status at every critical moment is captured. After video frame extraction is completed, server 104 can pass each static image frame as a separate input sample to the subsequent processing module, and enter the recognition process together with the image obtained from the static photo.

[0070] Through the above steps, server 104 ensures that original instrument images with stable quality and complete information can be obtained in both periodic inspection and continuous operation monitoring scenarios, and effectively adapts to various meter deployment methods and acquisition terminal devices, making the above method more widely applicable. For example, it can be applied to water pump fault monitoring, power grid transformer temperature monitoring, and speed monitoring during train operation.

[0071] Step S204: input the original meter image into the target detection model to obtain the target meter image and digital scale information.

[0072] For example, server 104 can use a trained object detection model to automatically identify the meter area from a raw meter image set against a complex background and extract scale digital information closely related to the meter reading. The target meter image refers to an image cropped from the original image by server 104 that contains a single meter area; the digital scale information refers to the numerical characters printed on the dial representing the range and their corresponding positions, such as "0," "50," and "100." For example, server 104 can use the YOLOv8 model to detect the input raw meter image. The YOLOv8 algorithm is a deep learning object detection algorithm that includes a BackBone (backbone network), a Neck (neck module), and a Head (head network). The BackBone (backbone network) serves as the core of the model's feature extraction; the Neck (neck module) serves as the intermediate layer connecting the BackBone and Head, and is used for multi-scale feature fusion and enhancement; and the Head is used to complete the detection task based on the features extracted from the BackBone and Neck.

[0073] Furthermore, the target detection model used by server 104 may not be a single structure, but may be composed of multiple sub-models working together to form an efficient detection system with task decoupling and parallel execution. For example, the target detection model may include a first detection sub-model, a second detection sub-model, and a third detection sub-model, each of which undertakes different sub-tasks and is trained with a dedicated data set to improve recognition accuracy and model robustness. The first detection sub-model is used to identify the meter area, that is, to locate the position and boundary of each dial in the image; the second detection sub-model is used to identify image defect information, such as stains, cracks, obstructions, and other interference factors that may affect recognition accuracy; and the third detection sub-model is used to identify meter digital information, such as scales such as "0", "100", and "500", for subsequent angle mapping.

[0074] Exemplarily, the server 104 may input the original meter image into the first detection sub-model to obtain a meter area image; input the meter area image into the second detection sub-model to obtain image defect information; filter the meter area image according to the image defect information to obtain a target meter image; and input the target meter image into the third detection sub-model to obtain digital scale information.

[0075] The first, second, and third detection sub-models can all be built based on the YOLOv8 network. Server 104 can input the original meter image into the first detection sub-model. The first detection sub-model features high precision, lightweight design, and real-time response. It can accurately identify the border position of each complete meter despite various background interferences, removing unnecessary complex backgrounds and reducing the difficulty of automated reading. The border information output by the model can be used for image cropping, cropping the meter area from the original captured image to obtain a dashboard image. Based on this, server 104 can extract multiple independent meter area images from the original meter image. A meter area image refers to an image containing only a single dial area.

[0076] Subsequently, server 104 inputs the meter area image into the second detection sub-model. This second detection sub-model is trained for image defects and can be used to identify common interferences such as glass cracks, dust, water stains, blurred areas, or partial occlusions on the pointer dial. It not only outputs the presence and type of defects but also provides information about the spatial location of the defects. Based on this defect information, server 104 can perform filtering on the meter area image. This can include masking the defect area, performing local enhancement, or removing frames with severe interference, thereby obtaining a higher-quality target meter image. For example, if the model identifies an obstructing sticker in the lower left corner of the dial, server 104 can perform weighted noise reduction on this area to prevent it from affecting subsequent scale recognition or pointer segmentation.

[0077] Next, server 104 can input the processed target meter image into the third detection sub-model to identify the digital scale information on the dial. The third detection sub-model can adapt to digital characters with different fonts, angles, scales, and printing quality, and can extract the text and position coordinates of key scales such as "0," "10," "50," and "250." The digital scale information can be used to determine the range and the mapping relationship between angles and values ​​during the reading process.

[0078] Through the above three-level sub-model division of labor and cooperation processing mechanism, server 104 can not only efficiently identify the meter location and digital information, but also actively identify and correct interference factors in the image that may affect the recognition accuracy, thereby obtaining a target meter image and digital scale information with complete structure, clear vision, and clear semantics, which significantly improves the overall anti-interference ability and adaptability of the system.

[0079] Furthermore, when the image defect information indicates that the corresponding meter area image is valid, median filtering is performed on the meter area image to obtain a target meter image.

[0080] For example, server 104 can perform further image quality enhancement processing on the meter area image to ensure that subsequent image segmentation operations can obtain clear and accurate structural information. At this point, server 104 can first determine whether the image defect information output by the second detection sub-model indicates that the current meter area image is in a valid state. A valid state means that the image has no recognizable defects, or that although there are minor defects such as dust on the edges or slight blur, the pointer, scale lines, and digital information are still discernible and not severely obscured or lost, and the overall image has the clarity and integrity required for identification and reading. Conversely, if the defect information indicates that the dial is covered by large stains, cracks, or stickers, making it impossible to extract the key area, server 104 can determine that the image is invalid and will no longer participate in subsequent calculations or transfer it to the manual review queue.

[0081] If the meter area image is determined to be valid, server 104 can perform a median filter on it to remove random noise and local interference. Median filtering is a classic image noise reduction method. Taking each pixel as the center, it takes the grayscale values ​​of several neighboring pixels and calculates the median, then uses this median value to replace the original value of the current pixel. When performing this operation, server 104 can traverse the entire image in 3×3 or 5×5 pixel neighborhoods. This effectively filters out abnormal bright and dark spots caused by sensor noise, compression errors, or dust reflections on site without destroying image edges.

[0082] For example, if the edge of a pointer in a dial image is affected by slight reflections, it may appear as broken or partially white pixels without filtering, which will affect the subsequent generation of the pointer mask. After median filtering, the pixel values ​​in this area will be smoother and more continuous, which helps the segmentation network accurately identify the complete pointer structure.

[0083] Furthermore, the server 104 can continue to perform adaptive histogram equalization on the image. The adaptive method can divide the image into multiple small grid areas, perform contrast equalization in each area, and then smoothly splice the areas, thereby solving the problem of some areas being too dark and some areas being overexposed in the instrument image. For example, in an environment with sunlight in the morning and evening or uneven indoor lighting, the pointer may be in a bright area and the scale is in a dark area, resulting in the inability of traditional processing methods to enhance the contrast of the two at the same time. Adaptive histogram equalization can automatically enhance the local contrast according to the brightness distribution of each area, so that the overall image presents a more balanced and clearer visual effect. Especially for instrument images with thin scale lines and pointer colors close to the background, this method can significantly enhance the contrast between them and the background, thereby improving the sensitivity of subsequent models to edge and line structures.

[0084] Through a combination of median filtering and adaptive equalization, server 104 generates a target meter image with excellent visual quality and structural clarity. This image not only has low noise, but also boasts balanced local contrast and clear details. It fully adapts to industrial environmental uncertainties during image preprocessing, improving the recognition algorithm's input quality control capabilities.

[0085] Step S206 : performing image segmentation on the target meter image according to the digital scale information to obtain a pointer line mask and a scale line mask.

[0086] Image segmentation involves server 104 semantically classifying each pixel in the image, determining whether the pixel belongs to the background, a pointer, or a scale line, ultimately generating a pixel mask with logical semantic distinctions. The pointer line mask and the scale line mask represent the sets of all pixels in the image that belong to the pointer structure and the scale line structure, respectively. Digital scale information can include the location of digital text and dial range information.

[0087] For example, server 104 can utilize the digital scale information output by the third detection sub-model. This information can include the digital text on the dial and its geometric position coordinates in the image, as well as the actual numerical value represented by the number, such as the dial's scale range (e.g., "0-500"). Server 104 can input the digital text position and scale information into a trained image segmentation model. This model can utilize a U-Net semantic segmentation architecture, which offers pixel-level accuracy and strong representation of local structural details.

[0088] Among them, UNet is a convolutional neural network architecture. Its unique U-shaped structure and jump connection enable it to perform well in the field of image segmentation, especially when the amount of data is limited. The structure of UNet consists of an encoder (downsampling path) and a decoder (upsampling path), and is U-shaped as a whole. The function of the encoder part is to extract image features, gradually reduce the resolution, and reduce the feature map size. The decoder restores the image resolution, gradually restores the feature map size, and generates the segmentation result. After the model receives the input image, it can first extract the multi-level features of the image through the encoder path, and then restore the image spatial structure layer by layer through the decoder path, and fuse high-resolution detail features through jump connections. Finally, the model can predict the category label for each pixel in the image and output a segmentation mask map. Based on the classification results of each pixel in the mask map, the server 104 composes the pixels belonging to the pointer area into a pointer area mask, the pixels belonging to the scale area into a scale line mask, and the rest is the background area.

[0089] Exemplarily, the server 104 can input the digital text position and dial range information into the image segmentation model to obtain the pointer area mask and the scale line mask; perform iterative edge point deletion on the pointer area mask until the edge points of the pointer area meet the preset conditions to obtain the pointer line mask. After completing the preliminary segmentation, the server 104 can also perform further geometric refinement on the pointer area mask to improve the accuracy of subsequent angle calculations. Since the original pointer area mask may have problems such as blurred edges, inconsistent thickness, and bifurcation, the server 104 can use a skeleton extraction algorithm based on iterative edge point deletion to perform morphological shrinkage on the mask. Exemplarily, the server 104 can identify all edge pixels in the mask and determine whether they meet the deletable conditions based on the pixel structure within their 8-neighborhood, that is, whether the skeleton will not be broken after deletion. Then, the server 104 can delete the edge points that meet the conditions in rounds, and re-judge and update the edge set after each round of operation. This iterative process can continue until all edge points do not meet the preset deletion conditions. At this time, the pointer area has been converged into a continuous line segment with a width of one pixel, which is the final pointer line mask.

[0090] For example, in a target meter image, the pointer region mask may initially have a width of 4 pixels, with irregularly fluctuating edges. During the thinning process, server 104 can gradually remove each layer of edges, retaining only the central skeleton, so that the pointer line ultimately appears as a clear line segment radiating from the center of the dial. This thinning process ensures the uniqueness of the pointer direction and the stability of angle calculation.

[0091] Through the above steps, the server 104 not only accurately extracts the scale and pointer areas from the image, but also compresses the complex image data into high-quality structural information that can be used for geometric modeling.

[0092] Step S208: Reading is performed according to the pointer line mask and the scale line mask to obtain a reading result.

[0093] Exemplarily, the server 104 can determine the straight line where the pointer is located based on the pointer line mask to obtain the pointer straight line; determine the scale data and any scale straight line based on the scale line mask; determine the center of the instrument based on the pointer straight line and the scale straight line; read the reading based on the pointer straight line, the scale data and the instrument center to obtain the reading result.

[0094] For example, the server 104 can determine the straight line structure where the pointer is located based on the pointer line mask. The server 104 can traverse all the pixel points marked as pointers in the mask and perform weighted least squares straight line fitting using the coordinate values ​​of these points as input. . Let the equation of the line be The corresponding least weighted squares error function is as follows:

[0095]

[0096] Solve the following equation to get the extreme point:

[0097]

[0098] in, is the slope, is the intercept, is the weight of the pixel. The coordinate value of the pixel is determined by taking the upper left corner of the image as the origin. According to the characteristics of the extreme point, there is a unique and The error function is minimized to obtain the straight line on which the pointer is located. During the fitting process, server 104 may also assign weights based on the distance between the pixel points and the image center. For example, points closer to the center may be given higher weights to improve the stability of the fitting and the accuracy of restoring the actual direction of the pointer. After the fitting is complete, server 104 can obtain a mathematically expressed straight line on the pointer.

[0099] Subsequently, the server 104 can extract the scale information from the scale mark mask. The server 104 can cluster the pixels marked as scale in the mask, identify all the scale lines in the image, and select any clear and complete scale line as the angle reference. This scale line can select a scale line close to the 0 scale value or the starting point of the scale to construct the starting direction of the angle. In addition, the server 104 can also establish a mapping relationship between each scale line in the image and the physical value it represents based on the position and content of the digital text identified by the third detection sub-model, forming a structured scale data table. For example, the server 104 can infer the scale range and scale distribution density of the entire dial based on the position of the numbers marked as "0", "100", and "200" in the image.

[0100] Next, if Figure 3 As shown, the server 104 can determine the center position of the instrument based on the spatial relationship between the pointer straight line and the selected scale straight line. The server 104 can use the intersection of these two straight lines as a reference to backtrack and solve the intersection of their extension lines, which is defined as the geometric center of the dial. This center is not only the starting point for angle calculation, but also the center point for constructing the polar coordinate system. All subsequent angle measurements are carried out with this point as the origin. After the center position is determined, the server 104 can use the pointer straight line, scale data and center coordinates together to calculate the value pointed by the current pointer. The server 104 constructs a polar coordinate system with the scale straight line as the baseline, measures the angle θ between the pointer line and the baseline, and obtains the deflection angle of the current pointer. The server 104 further searches for the maximum angle of the range previously obtained based on digital recognition For example, if the maximum angle of the dial is 270° and the corresponding maximum value is 500, the current angle can be mapped to the reading value according to the linear proportional relationship:

[0101]

[0102] This process can be automatically converted through a preset function, or it can be corrected using nonlinear interpolation according to the non-uniform distribution of the actual dial scale.

[0103] For example, server 104 determines that the angle between the current pointer and the "0" scale reference line is 135°, the maximum range is 500°, and the maximum angle is 270°. Therefore, the current reading is (135 ÷ 270) × 500 = 250. If the dial has an asymmetrical scale, for example, "0-100" is distributed within 180°, server 104 can also perform interval positioning and interpolation based on the scale data table to ensure that the numerical correspondence is reasonable.

[0104] Through the above steps, the server 104 not only achieves accurate mapping from image structure information to numerical values, but also adapts linear and nonlinear angle reading models according to different dial layouts, significantly improving the adaptability and accuracy of meter recognition.

[0105] In the above-mentioned meter data recognition method, by obtaining the original meter image and inputting it into the target detection model, the meter area is automatically located from the entire image to obtain the target meter image, and the digital scale information related to the reading is extracted, thereby avoiding the positioning deviation caused by manual rough cropping or meter misidentification, and significantly improving the image input quality of the subsequent recognition process. The target meter image is then segmented according to the digital scale information, and the pointer line mask and scale line mask are extracted respectively, thereby achieving accurate separation of the reading elements in the dial structure, so that the slender pointer structure and dense scale distribution in the meter can still be accurately extracted under complex background interference. Finally, reading is performed based on the above-mentioned mask information, which improves the accuracy of the reading results in complex scenarios such as instrument appearance differences, perspective changes and partial occlusion.

[0106] In an exemplary embodiment, Figure 4 As shown, the target detection model includes a first detection sub-model, a second detection sub-model and a third detection sub-model; the training method of the target detection model may include steps S302 to S308.

[0107] Step S302: Replace the convolution module of the original object detection model with a deformable convolution module to obtain a model to be trained.

[0108] Among them, the deformable convolution module contains an offset.

[0109] For example, in the initial training phase, the server 104 may first build an original model structure based on the initial YOLOv8 object detection architecture and replace all standard convolution modules with deformable convolution modules. The core feature of this module is the introduction of an offset learning mechanism, that is, based on the fixed sampling points of the traditional convolution kernel, a learnable offset is added to each sampling position. , so that the sampling of the convolution kernel on the image feature map is no longer limited to a regular grid, but can be adaptively adjusted according to the geometric shape of the target. Its formula is:

[0110]

[0111] in The input feature map is The coordinates of the point, is the value of the input feature map at the specified coordinate. is the total number of sampling points, It is the value of the output feature map at the specified coordinate. Is a predefined fixed offset Is the kth weight parameter. The comparison between deformable convolution and conventional convolution DCNv1 is shown in the figure Figure 5 As shown, the left picture is conventional convolution and the right picture is deformable convolution.

[0112] Furthermore, server 104 can use EIoU to split the aspect ratio loss term in CIoU into the difference between the predicted width and height and the minimum bounding box width and height, directly optimizing the difference between the width and height of the box. This can more accurately reflect the difference between the width and height of the predicted box and the ground truth box, resulting in faster convergence. The formula is:

[0113]

[0114] in, is the intersection-over-union ratio of the predicted box and the true box; The center point of the prediction box The center point of the real frame ; It is the square of the diagonal length of the minimum enclosing rectangle covering the predicted box and the true box. and is the width of the predicted box and the true box; and are the heights of the predicted box and the true box respectively. is the width of the minimum enclosing rectangle (the difference between the maximum right coordinate and the minimum left coordinate of the two boxes). The height of the minimum enclosing rectangle (the difference between the maximum lower coordinate and the minimum upper coordinate of the two boxes).

[0115] In this way, server 104 can construct a training model that maintains the high efficiency of the YOLOv8 detection structure while gaining stronger detection capabilities for target geometric deformation. For example, when processing tilted dials, curved pointers, or uneven scales, deformable convolution can dynamically capture key feature points, significantly improving the accuracy of detection results.

[0116] Step S304: Acquire a meter data set, and use the meter data set to train the to-be-trained model to obtain a first detection sub-model.

[0117] The first detection sub-model is used to identify the meter area.

[0118] Exemplarily, the server 104 can acquire and construct a meter dataset, which contains a large number of industrial site images containing pointer-type instruments, and provides accurate meter border annotation information for each image. Image types include samples of multiple angles, multiple lighting, and multiple instrument styles to enhance the versatility of the model. The server 104 can input the dataset into the model to be trained and use supervised learning for training, setting a suitable loss function (such as EIoU loss) for regressing the position and size of the prediction box, and using the classification branch to identify the target category. During the training process, the server 104 can also use data enhancement strategies, such as random cropping, rotation, brightness perturbation, etc., to improve the adaptability of the model to different on-site working conditions. Finally, the server 104 can output the trained first detection sub-model, which has the ability to accurately identify the overall border area of ​​the instrument in a complex background.

[0119] Step S306: Obtain a defect data set, and use the defect data set to train the to-be-trained model to obtain a second detection sub-model.

[0120] For example, server 104 can obtain a defect dataset that can contain a large number of instrument image samples with real defect annotations, where each defect area is accompanied by a location box and defect type label (such as cracks, stains, occlusion, blur, etc.). Server 104 can use this dataset to train the same set of models to be trained, and perform multi-class classification with the defect category as the target category label, while accurately locating the defect area through the bounding box regression task. Because defects are usually small in size, have irregular edges, and are close in color to the background, server 104 can rely on the perception capabilities of deformable convolution in this model to enable the model to better capture irregular structures and introduce appropriate image enhancement methods, such as blur simulation and transparent occlusion synthesis, to simulate real interference in industrial sites. After training is completed, the second detection sub-model that server 104 can obtain can stably identify and locate key defects in images that have potential impacts on readings in high-noise scenarios.

[0121] Step S308: Acquire a digital data set, and use the digital data set to train the to-be-trained model to obtain a third detection sub-model.

[0122] For example, server 104 can prepare a special digital data set, in which each image can contain a clear dial area and its corresponding digital area with precise annotations, and the digital text includes both character content (such as "0", "50", "100") and its position box in the image. During the training process, server 104 can improve the model's robustness to similar fonts and rotation angle changes by improving the resolution input, adding an attention mechanism, or introducing additional character discrimination loss terms. In addition, server 104 can also expand samples for different fonts and different manufacturers' dial styles to ensure that the model is universal for diverse character styles in actual deployment.

[0123] Through the aforementioned training process combining structural optimization and task decoupling, server 104 can construct three detection sub-models with distinct functional positioning. This model not only fully inherits the efficient detection capabilities of YOLOv8 but also integrates a deformable perception mechanism at the structural level, making the overall model performance more robust and adaptable in complex industrial image recognition. This training method improves recognition accuracy by combining structural replacement with data-based task training.

[0124] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0125] Based on the same inventive concept, embodiments of the present application further provide a meter data recognition device for implementing the aforementioned meter data recognition method. The solution provided by this device is similar to the solution described in the aforementioned method. Therefore, the specific limitations in one or more embodiments of the meter data recognition device provided below can be found in the limitations of the meter data recognition method described above and will not be further elaborated here.

[0126] In an exemplary embodiment, Figure 6As shown, a meter data recognition device is provided, comprising: an image acquisition module 602, an image recognition module 604, an image segmentation module 606 and a data processing module 608, wherein:

[0127] Image acquisition module 602, used to acquire original instrument images;

[0128] Image recognition module 604, used to input the original meter image into the target detection model to obtain the target meter image and digital scale information;

[0129] An image segmentation module 606 is used to segment the target meter image according to the digital scale information to obtain a pointer line mask and a scale line mask;

[0130] The data processing module 608 is used to perform readings according to the pointer line mask and the scale line mask to obtain a reading result.

[0131] In one embodiment, the target detection model includes a first detection sub-model, a second detection sub-model and a third detection sub-model; the device also includes: a model training module, used to replace the convolution module of the original target detection model with a deformable convolution module to obtain a model to be trained; wherein the deformable convolution module includes an offset; obtaining a meter data set, and using the meter data set to train the model to be trained to obtain a first detection sub-model; wherein the first detection sub-model is used to identify the meter area; obtaining a defect data set, and using the defect data set to train the model to be trained to obtain a second detection sub-model; obtaining a digital data set, and using the digital data set to train the model to be trained to obtain a third detection sub-model.

[0132] In one embodiment, the image recognition module 604 is specifically used to: input the original meter image into the first detection sub-model to obtain a meter area image; input the meter area image into the second detection sub-model to obtain image defect information; filter the meter area image based on the image defect information to obtain a target meter image; and input the target meter image into the third detection sub-model to obtain digital scale information.

[0133] In one embodiment, the image recognition module 604 is further configured to perform median filtering on the meter area image to obtain a target meter image when the image defect information indicates that the corresponding meter area image is valid.

[0134] In one embodiment, the digital scale information includes the digital text position and the dial range information; the image segmentation module 606 is specifically used to: input the digital text position and the dial range information into the image segmentation model to obtain the pointer area mask and the scale line mask; perform iterative edge point deletion processing on the pointer area mask until the edge points of the pointer area meet the preset conditions to obtain the pointer line mask.

[0135] In one embodiment, the data processing module 608 is specifically used to: determine the straight line where the pointer is located according to the pointer line mask to obtain the pointer straight line; determine the scale data and any scale straight line according to the scale line mask; determine the center of the instrument according to the pointer straight line and the scale straight line; read the number according to the pointer straight line, the scale data and the center of the instrument to obtain the reading result.

[0136] Each module in the aforementioned meter data recognition device may be implemented in whole or in part through software, hardware, or a combination thereof. Each module may be embedded in or independent of a processor in a computer device in the form of hardware, or may be stored in a memory in the computer device in the form of software, so that the processor can call and execute the corresponding operations of each module.

[0137] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Figure 7 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. The processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store original instrument images and training data of the target detection model, and can also be used to periodically store the recognition results of meter data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a meter data recognition method is implemented.

[0138] In an exemplary embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as shown in FIG. Figure 8As shown. The computer device includes a processor, memory, an input / output interface, a communication interface, a display unit, and an input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals via wired or wireless means, and the wireless means can be achieved via Wi-Fi, mobile cellular networks, NFC (near-field communication), or other technologies. When executed by the processor, the computer program implements a meter data recognition method. The display unit of the computer device is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device casing, or an external keyboard, touchpad or mouse.

[0139] Those skilled in the art will understand that Figure 8 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0140] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented: obtaining an original meter image; inputting the original meter image into a target detection model to obtain a target meter image and digital scale information; performing image segmentation on the target meter image based on the digital scale information to obtain a pointer line mask and a scale line mask; and reading the meter based on the pointer line mask and the scale line mask to obtain a reading result.

[0141] In one embodiment, when the processor executes the computer program, the following steps are further implemented: replacing the convolution module of the original target detection model with a deformable convolution module to obtain a model to be trained; wherein the deformable convolution module includes an offset; obtaining a meter data set, and using the meter data set to train the model to be trained to obtain a first detection sub-model; wherein the first detection sub-model is used to identify the meter area; obtaining a defect data set, and using the defect data set to train the model to be trained to obtain a second detection sub-model; obtaining a digital data set, and using the digital data set to train the model to be trained to obtain a third detection sub-model.

[0142] In one embodiment, when executing the computer program, the processor further implements the following steps: inputting the original meter image into the first detection sub-model to obtain a meter area image; inputting the meter area image into the second detection sub-model to obtain image defect information; filtering the meter area image based on the image defect information to obtain a target meter image; and inputting the target meter image into the third detection sub-model to obtain digital scale information.

[0143] In one embodiment, when the processor executes the computer program, the following steps are further implemented: when the image defect information indicates that the corresponding meter area image is valid, median filtering is performed on the meter area image to obtain a target meter image.

[0144] In one embodiment, when the processor executes the computer program, the following steps are also implemented: the digital text position and dial range information are input into the image segmentation model to obtain a pointer area mask and a scale line mask; the pointer area mask is iteratively edge point deleted until the edge points of the pointer area meet preset conditions to obtain a pointer line mask.

[0145] In one embodiment, when the processor executes the computer program, it also implements the following steps: determining the straight line where the pointer is located according to the pointer line mask to obtain the pointer straight line; determining the scale data and any scale straight line according to the scale line mask; determining the center of the instrument according to the pointer straight line and the scale straight line; reading the instrument according to the pointer straight line, the scale data and the center of the instrument to obtain the reading result.

[0146] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented: obtaining an original meter image; inputting the original meter image into a target detection model to obtain a target meter image and digital scale information; performing image segmentation on the target meter image based on the digital scale information to obtain a pointer line mask and a scale line mask; and taking readings based on the pointer line mask and the scale line mask to obtain a reading result.

[0147] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: replacing the convolution module of the original target detection model with a deformable convolution module to obtain a model to be trained; wherein the deformable convolution module includes an offset; obtaining a meter data set, and using the meter data set to train the model to be trained to obtain a first detection sub-model; wherein the first detection sub-model is used to identify the meter area; obtaining a defect data set, and using the defect data set to train the model to be trained to obtain a second detection sub-model; obtaining a digital data set, and using the digital data set to train the model to be trained to obtain a third detection sub-model.

[0148] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: inputting the original meter image into the first detection sub-model to obtain a meter area image; inputting the meter area image into the second detection sub-model to obtain image defect information; filtering the meter area image based on the image defect information to obtain a target meter image; and inputting the target meter image into the third detection sub-model to obtain digital scale information.

[0149] In one embodiment, when the computer program is executed by the processor, the following steps are further implemented: when the image defect information indicates that the corresponding meter area image is valid, median filtering is performed on the meter area image to obtain a target meter image.

[0150] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: the digital text position and dial range information are input into the image segmentation model to obtain a pointer area mask and a scale line mask; the pointer area mask is iteratively edge point deleted until the edge points of the pointer area meet preset conditions to obtain a pointer line mask.

[0151] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: determining the straight line where the pointer is located according to the pointer line mask to obtain the pointer straight line; determining the scale data and any scale straight line according to the scale line mask; determining the center of the instrument according to the pointer straight line and the scale straight line; reading the number according to the pointer straight line, the scale data and the center of the instrument to obtain the reading result.

[0152] In one embodiment, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the following steps: acquiring an original meter image; inputting the original meter image into a target detection model to obtain a target meter image and digital scale information; performing image segmentation on the target meter image based on the digital scale information to obtain a pointer line mask and a scale line mask; and performing readings based on the pointer line mask and the scale line mask to obtain a reading result.

[0153] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: replacing the convolution module of the original target detection model with a deformable convolution module to obtain a model to be trained; wherein the deformable convolution module includes an offset; obtaining a meter data set, and using the meter data set to train the model to be trained to obtain a first detection sub-model; wherein the first detection sub-model is used to identify the meter area; obtaining a defect data set, and using the defect data set to train the model to be trained to obtain a second detection sub-model; obtaining a digital data set, and using the digital data set to train the model to be trained to obtain a third detection sub-model.

[0154] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented: inputting the original meter image into the first detection sub-model to obtain a meter area image; inputting the meter area image into the second detection sub-model to obtain image defect information; filtering the meter area image based on the image defect information to obtain a target meter image; and inputting the target meter image into the third detection sub-model to obtain digital scale information.

[0155] In one embodiment, when the computer program is executed by the processor, the following steps are further implemented: when the image defect information indicates that the corresponding meter area image is valid, median filtering is performed on the meter area image to obtain a target meter image.

[0156] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: the digital text position and dial range information are input into the image segmentation model to obtain a pointer area mask and a scale line mask; the pointer area mask is iteratively edge point deleted until the edge points of the pointer area meet preset conditions to obtain a pointer line mask.

[0157] In one embodiment, when the computer program is executed by the processor, the following steps are also implemented: determining the straight line where the pointer is located according to the pointer line mask to obtain the pointer straight line; determining the scale data and any scale straight line according to the scale line mask; determining the center of the instrument according to the pointer straight line and the scale straight line; reading the number according to the pointer straight line, the scale data and the center of the instrument to obtain the reading result.

[0158] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. In particular, any reference to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may be, but are not limited to, general-purpose processors, central processing units (CPUs), graphics processing units (GPUs), digital signal processors (DSPs), programmable logic devices (PLDs), data processing logic devices based on quantum computing, and the like.

[0159] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0160] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A meter data recognition method, characterized in that: The method comprises: Get the original instrument image; Inputting the original instrument image into the target detection model to obtain the target meter image and digital scale information; Performing image segmentation on the target meter image according to the digital scale information to obtain a pointer line mask and a scale line mask; Readings are performed according to the pointer line mask and the scale line mask to obtain reading results.

2. The method according to claim 1, characterized in that The target detection model includes a first detection sub-model, a second detection sub-model and a third detection sub-model; and the training method of the target detection model includes: Replacing the convolution module of the original object detection model with a deformable convolution module to obtain a model to be trained; wherein the deformable convolution module includes an offset; Acquire a meter data set, and use the meter data set to train the to-be-trained model to obtain a first detection sub-model; wherein the first detection sub-model is used to identify the meter area; Acquire a defect data set, and use the defect data set to train the model to be trained to obtain a second detection sub-model; A digital data set is obtained, and the model to be trained is trained using the digital data set to obtain a third detection sub-model.

3. The method according to claim 2, characterized in that The step of inputting the original meter image into a target detection model to obtain a target meter image and digital scale information includes: Inputting the original meter image into the first detection sub-model to obtain a meter area image; Inputting the meter area image into the second detection sub-model to obtain image defect information; performing filtering processing on the meter area image according to the image defect information to obtain a target meter image; The target meter image is input into the third detection sub-model to obtain digital scale information.

4. The method according to claim 3, characterized in that The filtering process of the meter area image according to the image defect information to obtain the target meter image includes: When the image defect information indicates that the corresponding meter area image is valid, median filtering is performed on the meter area image to obtain a target meter image.

5. The method according to any one of claims 1 to 4, characterized in that The digital scale information includes digital text position and dial range information; and the target meter image is segmented according to the digital scale information to obtain a pointer line mask and a scale line mask, including: Inputting the digital text position and the dial range information into an image segmentation model to obtain a pointer area mask and a scale mark mask; An iterative edge point deletion process is performed on the pointer area mask until the edge points of the pointer area meet a preset condition, thereby obtaining a pointer line mask.

6. The method according to any one of claims 1 to 4, characterized in that The step of performing readings according to the pointer line mask and the scale line mask to obtain a reading result includes: Determine the straight line where the pointer is located according to the pointer line mask to obtain the pointer straight line; Determine scale data and any scale line according to the scale line mask; Determine the center of the instrument according to the pointer straight line and the scale straight line; Readings are taken according to the pointer straight line, the scale data and the center of the instrument to obtain a reading result.

7. A meter data recognition device, characterized in that: The device comprises: Image acquisition module, used to acquire original instrument images; An image recognition module is used to input the original instrument image into a target detection model to obtain a target meter image and digital scale information; An image segmentation module, configured to segment the target meter image according to the digital scale information to obtain a pointer line mask and a scale line mask; The data processing module is used to perform readings according to the pointer line mask and the scale line mask to obtain reading results.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images