An image recognition-based water meter parameter monitoring method and system, and a storage medium

Abstract

The application provides a kind of water meter parameter monitoring method and system based on image recognition and storage medium, it is related to water meter parameter monitoring technical field, comprising: obtaining water meter original image and carrying out multidimensional random enhancement, identifying water meter model and matching reading frame structure template, calculating the fitting degree coefficient of template and image actual reading frame;Based on the fitting degree coefficient dynamic adjustment mark precision threshold value, and according to the standard length-width ratio of template and diagonal line mark rule, mark the reading frame area, determine its complete contour;Adaptive scaling is carried out on the marked image using the mechanism of fixed-point scaling and center-point scaling, to obtain the standard image matched with the input size of identification model, input the initial identification result into water meter reading identification model, calculate the trend matching coefficient combined with historical meter reading data, and calculate the reliability index of each result by integrating the fitting degree coefficient, review and exclude suspicious results, and output the final water meter parameter monitoring result.

Description

Technical Field

[0001] This invention relates to the field of water meter parameter monitoring technology, specifically to a water meter parameter monitoring method, system, and storage medium based on image recognition. Background Technology

[0002] With the continuous improvement of water management intelligence, automatic meter reading technology based on image recognition has become an important means to replace manual meter reading and improve meter reading efficiency. At present, there are various methods for water meter digital recognition using deep learning. These methods usually achieve the detection of reading areas and character recognition by preprocessing, enhancing and training models on water meter images. However, due to the complex installation environment and diverse shooting conditions of actual water meters, existing methods still have considerable room for improvement in terms of recognition accuracy, robustness and reliability assessment of reading results.

[0003] In the existing technology, a deep learning-based remote water meter digit recognition method (publication number CN115984862A) achieves automatic recognition of water meter digits through steps such as receiving water meter information, parsing data, image enhancement, dataset creation, model training, and real-time recognition. Although this method can improve the recognition accuracy of full and half characters and adapt to different installation environments, it still has the following shortcomings: First, the method still relies on the traditional four-point positioning method when annotating the reading area, which has low annotation efficiency and is prone to human error; Second, it does not perform adaptive processing for different shooting distances and target sizes during the image scaling stage, affecting the recognition effect of small targets or water meters photographed from a distance; Third, it lacks a mechanism for quantitatively evaluating the credibility of the recognition results, and cannot effectively combine historical data to verify and filter abnormal recognition results, thus limiting its reliability and applicability in actual deployment.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a water meter parameter monitoring method, system, and storage medium based on image recognition, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A method for monitoring water meter parameters based on image recognition, comprising the following steps:

[0008] S1: Obtain the original image data of the water meter to be identified, and perform multi-dimensional random enhancement processing on the original image data to obtain enhanced image data;

[0009] S2: Based on the preset water meter type judgment rules, identify the water meter model corresponding to the enhanced image data, and match the corresponding reading frame structure template according to the water meter model. Simultaneously calculate the adaptation coefficient used to characterize the matching degree between the reading frame structure template and the actual reading frame in the enhanced image.

[0010] S3: Based on the adaptation coefficient, dynamically adjust the annotation accuracy threshold, and according to the standard length-width ratio and diagonal annotation rules determined by the matched reading frame structure template, annotate the water meter reading frame area in the enhanced image data to determine the complete outline of the reading frame.

[0011] S4: For the enhanced image data with a complete outline reading frame marked, a scaling mechanism combining fixed-point scaling and center-point scaling is used to obtain standard image data that matches the preset input size of the water meter reading recognition model.

[0012] S5: Input the standard image data into the preset water meter reading recognition model, output the initial recognition result, and determine the trend matching coefficient to characterize the consistency between the initial recognition result and the historical data trend based on the time series characteristics of the user's historical meter reading data.

[0013] S6: Combine the fit coefficient and trend matching coefficient to calculate the result reliability index of each initial identification result. Review the initial identification results with result reliability index lower than the preset threshold, exclude abnormal data, and select the initial identification results that meet the requirements as the final water meter parameter monitoring results.

[0014] Furthermore, the specific execution process of performing multi-dimensional random enhancement processing on the original image data is as follows:

[0015] The process involves acquiring the original image data of the water meter to be identified, randomly selecting a cropping area and adjusting the cropping ratio while maintaining the integrity of the reading frame, applying Gaussian blur to the cropped image, and simulating different degrees of blurring using a randomly generated Gaussian kernel standard deviation. Next, random noise conforming to a normal distribution is superimposed on the image pixel values. Then, a mask layer with random shape and transparency is generated to simulate the effect of natural light occlusion. Different lighting environments are simulated by adjusting the image brightness, contrast, and color temperature. Random rotation and small-angle oscillation operations are then performed to rotate the image within a preset angle range. Finally, the image's hue, saturation, and brightness parameters are randomly fine-tuned to simulate the color performance differences of different shooting devices.

[0016] The execution order and intensity parameters of each enhancement operation are randomly generated in each round of processing, and the final output is enhanced image data with diverse shooting conditions. The intensity parameters include cropping ratio, Gaussian kernel standard deviation, noise intensity, mask transparency, brightness adjustment range, contrast adjustment range, color temperature offset, rotation angle range, hue offset, saturation adjustment range, and brightness adjustment range.

[0017] Furthermore, based on the enhanced image data, the dial shape, digital area distribution, scale line style, and brand logo of the water meter are extracted. These features are then compared with a database of known water meter models using a pre-defined classification model to identify the corresponding water meter model.

[0018] After identifying the water meter model, the system matches the corresponding model's structure template from the preset reading frame structure template library based on the model index. This template includes the layout information of the length-width ratio, number of digits, and decimal point position of the water meter reading frame for that model.

[0019] The adaptation coefficient, which characterizes the degree of matching between the reading frame structure template and the actual reading frame in the enhanced image, is calculated synchronously using the following formula:

[0020]

[0021] In the formula, This represents the fitness coefficient, and its value range is... ; Structural similarity is represented by extracting the Canny edge feature maps of the reading frame structural template and the actual reading frame in the image, and calculating the cosine similarity between the two. This represents the normalized positional deviation distance, calculated as the ratio of the Euclidean distance between the geometric center point of the reading frame template and the actual center point of the image reading frame, divided by the length of the image diagonal. Its value range is... ; The directional consistency parameter is calculated by taking the absolute value of the angle between the principal direction of the template reading frame and the actual image reading frame, and then normalizing it by dividing by 90 degrees. Its value range is... ; , , Assign a preset weight to the corresponding indicator, and satisfy the following conditions: .

[0022] Furthermore, a labeling accuracy threshold is dynamically set based on the fit coefficient, wherein the labeling accuracy threshold... The calculation formula is as follows:

[0023]

[0024] in, Based on the basic annotation accuracy threshold, This is the adjustment coefficient;

[0025] Subsequently, the standard length and width ratio was determined based on the matching reading frame structure template. , For template width, The template height is used, and a diagonal annotation method is employed to automatically detect and annotate the upper left corner of the water meter reading frame area in the enhanced image data. and the bottom right corner ;

[0026] Finally, based on the two corner points mentioned in the annotation... and According to the aforementioned standard aspect ratio The coordinates of the other two corner points of the reading frame are calculated using the following geometric relationships:

[0027]

[0028]

[0029] Thus determine by , , , The complete outline of the reading frame formed by the four corner points.

[0030] Furthermore, the specific execution process of S4 is as follows:

[0031] First, obtain the standard input dimensions of the preset water meter reading recognition model. Then, calculate the actual pixel percentage of the reading frame region in the enhanced image data with the complete outline of the reading frame. If this actual pixel percentage is less than a preset percentage threshold... If the actual pixel ratio is greater than or equal to the preset ratio threshold, a fixed-point scaling mechanism is adopted. Using the geometric center point of the reading frame as a fixed reference, the size is scaled proportionally according to the ratio coefficient between the standard input size and the actual reading frame size, ensuring that the numbers and scales within the reading frame are completely preserved without distortion. The center point scaling mechanism is adopted to keep the geometric center point of the reading frame unchanged and adjust the overall scaling ratio of the image according to the standard input size so that the reading frame area is adapted to the model input requirements. During the scaling process, bilinear interpolation algorithm is used to supplement pixels at the image edges to avoid jagged distortion. Finally, the output size is a standard image data that is completely matched with the model input size.

[0032] Furthermore, the standard image data is input into the pre-trained water meter reading recognition model, and the initial water meter parameter recognition result is output based on the layout information of the number of digits and the position of the decimal point contained in the reading frame structure template. The initial recognition result includes a digital reading with decimal places. The water meter reading recognition model is obtained by supervised learning training through a convolutional neural network structure based on a large amount of manually annotated water meter reading image data.

[0033] At the same time, the user's historical meter reading data for the past six months was retrieved, and its time series characteristics, including average daily water consumption, were extracted. Slope of water volume change trend and the upward period to The slope of the water volume change trend is obtained by linear fitting of historical meter reading data;

[0034] Based on the time-series characteristics of the historical meter reading data and the initial identification results, a trend matching coefficient is calculated using the following formula:

[0035]

[0036] In the formula, This represents the trend matching coefficient, with a value range of [value range missing]. ; These are the readings from the initial identification results; This represents the average daily water volume. To determine the slope based on historical water volume change trends With the upward period to The calculated current forecast reading is as follows: ,in This is the last valid meter reading; The maximum daily water volume threshold is set based on historical data; This is the tolerance parameter; , , For the preset weights, and satisfying ; For indicator function: when the initial recognition result Less than or equal to When the value is 1, the indicator function takes the value 1; when Greater than When the value is 0, the indicator function takes the value 0.

[0037] Furthermore, based on the fitness coefficient Matching coefficient with trend The reliability index of each initial identification result is calculated using the following formula:

[0038]

[0039] in, This represents the reliability index of the results, with a value range of [value missing]. ; This is the fit coefficient; , To preset weights, And satisfy ;

[0040] The reliability index of the result is lower than the preset reliability index threshold. The initial identification results were marked as suspicious results;

[0041] The initial identification results marked as suspicious are reviewed, specifically: the meter reading data of the same user for several historical meter reading cycles before the meter reading time point corresponding to the suspicious result is extracted, the historical average daily water consumption within these historical meter reading cycles is calculated, and it is compared with the current daily water consumption calculated based on the suspicious result; if the current daily water consumption does not exceed the preset multiple threshold of the historical average daily water consumption, the suspicious result is corrected to a reliable result, otherwise it is excluded.

[0042] The reliability index shall not be lower than All initial identification results, as well as the initial identification results corrected to reliable results, are output as the final water meter parameter monitoring results.

[0043] The present invention also provides a water meter parameter monitoring system based on image recognition, wherein the water meter parameter monitoring system based on image recognition is used to execute the above-mentioned water meter parameter monitoring method based on image recognition, comprising:

[0044] The image enhancement processing module is used to acquire the original image data of the water meter to be identified, and to perform multi-dimensional random enhancement processing on the original image data to obtain enhanced image data.

[0045] The matching module is used to identify the water meter model corresponding to the enhanced image data based on the preset water meter type judgment rules, and match the corresponding reading frame structure template according to the water meter model, and simultaneously calculate the fit coefficient used to characterize the degree of matching between the structure template and the actual reading frame in the enhanced image.

[0046] The reading frame annotation module is used to dynamically adjust the annotation accuracy threshold based on the matched reading frame structure template, diagonal annotation rules and adaptation coefficient, to annotate the water meter reading frame area in the enhanced image data and determine the complete outline of the reading frame.

[0047] The adaptive scaling adjustment module is used to scale the enhanced image data with a complete outline reading frame by combining fixed-point scaling and center-point scaling to obtain standard image data that matches the preset input size of the water meter reading recognition model.

[0048] The identification and trend matching module is used to input the standard image data into a preset water meter reading identification model, output the initial identification result, and determine the trend matching coefficient to characterize the consistency between the initial identification result and the historical data trend based on the time series characteristics of the user's historical meter reading data.

[0049] The comprehensive verification module is used to combine the fit coefficient and the trend matching coefficient to calculate the result reliability index of each initial identification result, and to review the initial identification results with a result reliability index lower than a preset threshold, exclude abnormal data, and select the initial identification results that meet the requirements as the final water meter parameter monitoring results.

[0050] The present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for monitoring water meter parameters based on image recognition.

[0051] Compared with the prior art, the beneficial effects of the present invention are:

[0052] First, the present invention adopts a diagonal marking rule based on the reading frame structure template, which improves the efficiency and accuracy of marking the water meter reading area. Traditional methods usually require manual marking of four points to determine the rectangular frame, while the present invention only needs to locate the upper left and lower right corner points and automatically calculate the complete rectangular outline according to the standard length and width ratio of the template. This not only significantly improves the theoretical marking speed, but also avoids the shape deviation introduced by manual marking, ensuring more accurate positioning of the reading frame.

[0053] Secondly, by introducing an adaptive scaling mechanism and a multi-dimensional credibility evaluation system, this invention significantly enhances the adaptability of the recognition system to different shooting conditions and complex environments. For long-distance, small-target water meter images, it adopts an intelligent processing strategy that combines fixed-point scaling and center-point scaling, effectively improving the recognition rate of small-sized dials. At the same time, by calculating the fit coefficient, trend matching coefficient, and result reliability index, a multi-level quantitative evaluation and verification mechanism for recognition results is constructed. This mechanism can be combined with historical water usage data to review and filter abnormal recognition results, thereby ensuring the accuracy and reliability of monitoring results at both the data source and model output ends.

[0054] Furthermore, this invention achieves full automation and intelligence in the process of water meter model identification, structural template matching, image enhancement, reading frame positioning, character recognition, and result verification. The system can adaptively match reading frame templates according to different water meter models and dynamically adjust annotation parameters based on template adaptability, making the entire method highly scalable and applicable. The final monitoring results have a clear reliability evaluation, facilitating quality control and anomaly early warning in actual deployment, and providing reliable technical support for intelligent and refined water management. Attached Figure Description

[0055] Figure 1 This is a schematic diagram of the overall method flow of the present invention;

[0056] Figure 2 A dual Y-axis image of structural similarity, positional deviation distance, and fit coefficient;

[0057] Figure 3 This is a parallel coordinate image of structural similarity, positional deviation distance, orientation consistency parameters, and fit coefficient;

[0058] Figure 4 This is a schematic diagram of the overall system modules of the present invention. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0060] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0061] Example:

[0062] Please see Figures 1-3 The present invention provides a technical solution:

[0063] A method for monitoring water meter parameters based on image recognition, comprising the following steps:

[0064] S1: Obtain the original image data of the water meter to be identified, and perform multi-dimensional random enhancement processing on the original image data to obtain enhanced image data;

[0065] In this embodiment, the specific execution process of performing multi-dimensional random enhancement processing on the original image data is as follows:

[0066] The process involves acquiring the original image data of the water meter to be identified, randomly selecting a cropping area and adjusting the cropping ratio while maintaining the integrity of the reading frame, applying Gaussian blur to the cropped image, and simulating different degrees of blurring using a randomly generated Gaussian kernel standard deviation. Next, random noise conforming to a normal distribution is superimposed on the image pixel values. Then, a mask layer with random shape and transparency is generated to simulate the effect of natural light occlusion. Different lighting environments are simulated by adjusting the image brightness, contrast, and color temperature. Random rotation and small-angle oscillation operations are then performed to rotate the image within a preset angle range. Finally, the image's hue, saturation, and brightness parameters are randomly fine-tuned to simulate the color performance differences of different shooting devices.

[0067] The execution order and intensity parameters of each enhancement operation are randomly generated in each round of processing, and the final output is enhanced image data with diverse shooting conditions. The intensity parameters include cropping ratio, Gaussian kernel standard deviation, noise intensity, mask transparency, brightness adjustment range, contrast adjustment range, color temperature offset, rotation angle range, hue offset, saturation adjustment range, and brightness adjustment range.

[0068] S2: Based on the preset water meter type judgment rules, identify the water meter model corresponding to the enhanced image data, and match the corresponding reading frame structure template according to the water meter model. Simultaneously calculate the adaptation coefficient used to characterize the matching degree between the reading frame structure template and the actual reading frame in the enhanced image.

[0069] In this embodiment, based on the enhanced image data, the dial shape, digital area distribution, scale line style and brand logo of the water meter are extracted. These features are compared with a database of known water meter models using a preset classification model to identify the corresponding water meter model.

[0070] The process of identifying water meter signals is as follows: Based on the enhanced image data, the shape features of the water meter dial are first extracted, including the aspect ratio, roundness, and number of polygon sides of the dial outline, as well as the distribution features of the digital areas inside the dial. The position and arrangement density of the digital areas are located through image binarization and connected component analysis. At the same time, the scale line style features are extracted, including the length, spacing, direction, and arrangement pattern of the scale lines, and the straight line segments and arc segments are identified by combining edge detection and Hough transform. In addition, the brand logo characters or trademark patterns printed on the dial are identified through OCR technology. These extracted features are input into an image classification model based on a convolutional neural network and compared with the template features in a database of known water meter models. This database stores standardized feature vectors of each model of water meter in terms of dial shape, digital layout, scale style, and brand logo. The model calculates the similarity between the features of the image to be identified and the features of each template, and selects the model with the highest similarity as the identification result. This process realizes the automated mapping from multi-dimensional image features to water meter models, providing a reliable model basis for subsequent matching of reading frame structure templates. After identifying the water meter model, the system matches the corresponding model's structure template from the preset reading frame structure template library based on the model index. This template includes the layout information of the length-width ratio, number of digits, and decimal point position of the water meter reading frame for that model.

[0071] The adaptation coefficient, which characterizes the degree of matching between the reading frame structure template and the actual reading frame in the enhanced image, is calculated synchronously using the following formula:

[0072]

[0073] In the formula, This represents the fitness coefficient, and its value range is... ; Structural similarity is represented by extracting the Canny edge feature maps of the reading frame structural template and the actual reading frame in the image, and calculating the cosine similarity between the two. This represents the normalized positional deviation distance, calculated as the ratio of the Euclidean distance between the geometric center point of the reading frame template and the actual center point of the image reading frame, divided by the length of the image diagonal. Its value range is... ; The directional consistency parameter is calculated by taking the absolute value of the angle between the principal direction of the template reading frame and the actual image reading frame, and then normalizing it by dividing by 90 degrees. Its value range is... ; , , The preset weights for the corresponding indicators, And satisfy The weights are set in this way because, during the matching process of water meter reading frames, structural similarity is considered. This directly determines the consistency between the template and the contour of the reading region in the image, and is the core foundation of matching reliability; therefore, it is given the highest weight. Positional deviation distance This reflects the degree of center offset of the target in the image. Although important, a certain range of translation is permissible in actual shooting, hence the weight... Secondly; directional consistency parameters This describes the angular difference between the two. In normal shooting, the angular change is usually small and has a relatively weak impact on the matching results, therefore it is assigned the lowest weight. .

[0074] In calculating structural similarity When extracting the Canny edge feature maps of the reading frame structure template and the actual reading frame in the image, if there is a mismatch in image size, the edge feature maps are normalized to ensure they have the same size. Specifically: First, based on the contour range of the actual reading frame in the image, the Canny edge map of that region is extracted and its size is adjusted to be the same as the width and height of the template edge feature map. Bilinear interpolation is used to maintain the continuity of the edge structure. Then, the adjusted actual reading frame edge feature map and template edge feature map are converted into two-dimensional vector forms respectively, and the structural similarity between them is calculated using cosine similarity. The calculation formula is as follows:

[0075]

[0076] in, The template edge feature vector, This represents the feature vector of the actual reading frame edge of the image.

[0077] To ensure dimensional alignment of feature vectors, if there are blank areas introduced by scaling in the actual reading frame edge map, they are filled with zero values; if the template edge map size is larger than the actual reading frame area, sliding window clipping or global pooling is used for alignment.

[0078] For the formula used to calculate the fit coefficient, the dependent variable fit coefficient This is used to quantify the overall matching degree between the reading frame structure template and the actual reading frame in the image, and its value range is: , The closer the value is to 1, the higher the degree of matching between the template and the actual reading frame in terms of structure, position, and orientation; the more accurate the reading area positioning, and the higher the reliability of the subsequent recognition results. Conversely, the closer the value is to 1, the lower the degree of matching between the template and the actual reading frame in terms of structure, position, and orientation. The lower the value, the worse the matching quality, which may be caused by image blurring, occlusion, tilting, or improper template selection. In this case, the system needs to increase the annotation threshold or trigger the verification mechanism to ensure the reliability of recognition.

[0079] Structural similarity Used to reflect the degree of fit between the template and the actual reading frame edge contour. The larger the value, the closer the shapes are, and the stronger the match. Follow Increase and improve; positional deviation distance This indicates the relative offset between the center points of the two points. The larger the value, the more significant the deviation of the reading frame in the image, and the worse the consistency of the actual matching position. Follow Increase and decrease; directional consistency parameter Used to demonstrate the consistency of their orientation angles. The higher the value, the smaller the tilt or rotation angle, and the better the directional alignment, which is beneficial for subsequent digit recognition. Follow Increase and improve.

[0080] This formula uses a linear weighted combination to normalize and fuse the features of the three dimensions of structure, position, and orientation into a single fitness index, which is then weighted. , , The adjustment can flexibly adapt to the matching focus of different water meter models or application scenarios; at the same time, in the formula This results in a negative correlation between positional deviation and fit, which aligns with intuitive understanding.

[0081] Table 1: Statistics of Fit Coefficient

[0082]

[0083] Analysis of the data in Table 1 shows that the fit coefficient With structural similarity Consistency of direction It shows a significant positive correlation with the normalized positional bias, but is significantly positively correlated with the normalized positional bias. The correlation is clearly negative, which confirms the effectiveness of this invention in dynamically evaluating the matching quality of the reading frame: it not only focuses on the fit of the structural contour, but also considers the consistency of spatial position and direction, so as to comprehensively reflect the matching degree between the template and the actual reading frame of the image. This multi-dimensional evaluation mechanism provides a quantitative basis for subsequent adaptive annotation and recognition result verification.

[0084] Further analysis of the data trends reveals that when the structural similarity is high, the positional deviation is small, and the orientation is consistent, the fit coefficient approaches the ideal value, indicating that the system has excellent matching performance under ideal shooting conditions. However, when the structural features are weakened or the positional and directional deviations increase, the fit coefficient decreases accordingly, reflecting the system's sensitivity to image quality and shooting angle. This characteristic helps to identify low-quality images and trigger a verification mechanism in actual deployment, thereby improving the overall reliability and environmental adaptability of water meter parameter monitoring.

[0085] S3: Based on the adaptation coefficient, dynamically adjust the annotation accuracy threshold, and according to the standard length-width ratio and diagonal annotation rules determined by the matched reading frame structure template, annotate the water meter reading frame area in the enhanced image data to determine the complete outline of the reading frame.

[0086] In this embodiment, the annotation accuracy threshold is dynamically set based on the fit coefficient. The calculation formula is as follows:

[0087]

[0088] in, Based on the basic annotation accuracy threshold, For adjustment coefficients, ; By statistically analyzing the matching error distribution of a large number of correctly labeled water meter image samples during the initial labeling stage, and combining this with expert experience, adjustment coefficients were set. By collecting different fit coefficients in real-world application scenarios The annotation results are fed back and optimized using gradient descent or grid search methods.

[0089] Labeling accuracy threshold This value controls the strictness of automatic annotation of the water meter reading frame in the image. A higher value indicates a higher requirement for feature matching of the reading frame area, a more rigorous annotation process, and lower fault tolerance; conversely, a lower value indicates a lower requirement. The smaller the value, the more lenient the annotation process, allowing for a certain range of matching deviations; this can be achieved through dynamic adjustment. The system can adapt to different image qualities and matching confidence levels, improving annotation accuracy under high fit and relaxing annotation standards under low fit, thereby ensuring positioning accuracy while avoiding annotation failures caused by excessive strictness.

[0090] Adaptability coefficient This reflects the degree of matching between the reading frame template and the actual image area in terms of structure, position, and orientation. A higher value indicates better matching quality and more reliable positioning of the reading frame. In this case, the system can appropriately lower the annotation accuracy threshold. To improve annotation efficiency while ensuring accurate positioning; conversely, when A low value indicates significant uncertainty in the matching process, which may be caused by image blur, occlusion, or a biased shooting angle. In this case, the value should be increased. The method enhances the rigor of annotation and avoids mis-annotation or omission. It demonstrates the system's adaptive adjustment capability when image quality fluctuates, ensuring robust annotation results under different matching confidence levels.

[0091] The formula uses a linear adjustment form, with a base threshold. The initial annotation standard and adjustment coefficient provided by the system Controlling the impact of fit, through Transforming fitness into a reverse adjustment factor, i.e., the negative correlation logic of lower fitness and higher threshold, aligns with the operational intuition in practical applications that the worse the match, the more cautious the annotation needs to be.

[0092] Subsequently, the standard length and width ratio was determined based on the matching reading frame structure template. , For template width, The template height is used, and a diagonal annotation method is employed to automatically detect and annotate the upper left corner of the water meter reading frame area in the enhanced image data. and the bottom right corner ;

[0093] Finally, based on the two corner points mentioned in the annotation... and According to the aforementioned standard aspect ratio The coordinates of the other two corner points of the reading frame are calculated using the following geometric relationships:

[0094]

[0095]

[0096] Thus determine by , , , The complete outline of the reading frame formed by the four corner points.

[0097] Determine the standard aspect ratio in the matching reading frame structure template. Then, the system determines the accuracy threshold based on the annotation. By setting sensitivity and employing an image processing workflow that combines edge detection and morphological operations, the system automatically identifies and locates candidate regions most relevant to the template structure in the enhanced image data. Then, based on region connectivity and contour fitting strategies, it accurately extracts the diagonal endpoint of the water meter reading frame region, i.e., the upper left corner. and the bottom right corner Subsequently, the system based on the aforementioned standard aspect ratio Through a fixed geometric mapping relationship, by , and proportion The coordinates of the other two corner points are directly calculated, thus uniquely determining the location of the point. , , , The complete outline of the standard rectangular reading frame formed by the four corner points is completed, enabling automated and structured annotation of the reading frame.

[0098] S4: For the enhanced image data with a complete outline reading frame marked, a scaling mechanism combining fixed-point scaling and center-point scaling is used to obtain standard image data that matches the preset input size of the water meter reading recognition model.

[0099] In this embodiment, the specific execution process of S4 is as follows:

[0100] First, obtain the standard input dimensions of the preset water meter reading recognition model. Then, calculate the actual pixel percentage of the reading frame region in the enhanced image data with the complete outline of the reading frame. If this actual pixel percentage is less than a preset percentage threshold... If the actual pixel ratio is greater than or equal to the preset ratio threshold, a fixed-point scaling mechanism is adopted. Using the geometric center point of the reading frame as a fixed reference, the size is scaled proportionally according to the ratio coefficient between the standard input size and the actual reading frame size, ensuring that the numbers and scales within the reading frame are completely preserved without distortion. The center point scaling mechanism is adopted to keep the geometric center point of the reading frame unchanged and adjust the overall scaling ratio of the image according to the standard input size so that the reading frame area is adapted to the model input requirements. During the scaling process, bilinear interpolation algorithm is used to supplement pixels at the image edges to avoid jagged distortion. Finally, the output size is a standard image data that is completely matched with the model input size.

[0101] In the specific execution of S4, the standard input dimensions of the preset water meter reading recognition model are first obtained, and the width is denoted as . Gao Wei Calculate the actual pixel percentage of the reading frame region in the enhanced image with a fully outlined reading frame. The calculation formula is as follows:

[0102]

[0103] In the formula, The actual pixel area of ​​the reading frame is calculated from the coordinates of the four corner points; Represents the total pixel area of ​​the image;

[0104] Set preset percentage threshold ,like Then a fixed-point scaling mechanism is adopted: using the geometric center point of the reading frame as a fixed reference, the actual width and height of the reading frame are calculated, and the width scaling ratio is calculated respectively. and height scaling ,Pick This is used as a proportional magnification factor to ensure the content of the reading frame is complete and undistorted; during magnification, an affine transformation is performed on the image with the geometric center point of the reading frame as the center to maintain the sharp edges of the numbers and scales; if Then, a center-point scaling mechanism is used: the geometric center point of the reading frame remains unchanged, and the overall scaling ratio of the image is calculated. The entire image is scaled proportionally to adapt the reading frame area to the model input size. Under both scaling paths, bilinear interpolation is used to resample the image pixels, and the areas that exceed the boundary after scaling are filled with edge fill. The fill value is the mean of the edge pixels of the image. Finally, standard image data with a size that perfectly matches the model input size is output, ensuring that the image size is uniform and the content structure is intact before inputting into the model.

[0105] in, The value range is set to The rationale for this setting is as follows: if the reading frame area occupies too low a proportion of the total image area, it indicates that the shooting distance is far or the target size is small. In this case, using a fixed-point scaling mechanism centered on the reading frame can effectively magnify the details of the reading area and avoid the loss of clarity of numbers and scales due to global scaling. Conversely, if the reading frame occupies a high proportion, it indicates that the shooting is relatively close or the target is large. Using overall proportional scaling centered on the reading frame can adapt to the model input size while maintaining the integrity of the content structure, taking into account both processing efficiency and recognition accuracy. This threshold is determined based on experimental statistical analysis of water meter images at different shooting distances and can be fine-tuned according to specific scenarios in actual deployment.

[0106] S5: Input the standard image data into the preset water meter reading recognition model, output the initial recognition result, and determine the trend matching coefficient to characterize the consistency between the initial recognition result and the historical data trend based on the time series characteristics of the user's historical meter reading data.

[0107] In this embodiment, the standard image data is input into the pre-trained water meter reading recognition model, and the initial water meter parameter recognition result is output based on the layout information of the number of digits and the position of the decimal point contained in the reading frame structure template. The initial recognition result includes a digital reading with decimal places. The water meter reading recognition model is obtained by supervised learning training through a convolutional neural network structure based on a large amount of manually annotated water meter reading image data.

[0108] At the same time, the user's historical meter reading data for the past six months was retrieved, and its time series characteristics, including average daily water consumption, were extracted. Slope of water volume change trend and the upward period to The slope of the water volume change trend is obtained by linear fitting of historical meter reading data;

[0109] Based on the time-series characteristics of the historical meter reading data and the initial identification results, a trend matching coefficient is calculated using the following formula:

[0110]

[0111] In the formula, This represents the trend matching coefficient, with a value range of [value range missing]. ; These are the readings from the initial identification results; This represents the average daily water volume. To determine the slope based on historical water volume change trends With the upward period to The calculated current forecast reading is as follows: ,in This is the last valid meter reading; The maximum daily water volume threshold is set based on historical data; This is the tolerance parameter; , , The preset weights are determined using the analytic hierarchy process (AHP) and satisfy the following conditions: ; For indicator function: when the initial recognition result Less than or equal to When the value is 1, the indicator function takes the value 1; when Greater than When the value is 0, the indicator function takes the value 0.

[0112] For this formula, the dependent variable trend matching coefficient This is used to quantify the consistency between the currently identified water meter reading and the user's historical water usage trends; its value range is [value range missing]. The larger the value, the higher the degree of matching between the current identification results and historical water use patterns, and the more reliable the results; conversely, if... The smaller the value, the greater the deviation between the current reading and the historical trend. This may be due to identification errors, sudden abnormal water usage, or data anomalies, and requires verification or elimination.

[0113] The item reflects the current reading. Compared with historical daily average water volume The relative similarity between the two is such that if they are close, the item approaches 1, indicating that the current water use is in line with the daily level and the trend matching is high; if the difference is large, the item decreases and the matching decreases. The item predicts the current reading based on historical trends. The deviation between the identification results and the predicted values ​​is evaluated. The smaller the deviation, the closer the index term is to 1, indicating that the identification results are consistent with the water use trend. When the deviation is too large, the term decays rapidly, the matching coefficient decreases, and it reflects possible abnormal readings or sudden changes in the trend. The item is a Boolean judgment on the maximum daily water volume limit; if the current reading exceeds the historical reasonable upper limit... The value of 0 directly reduces the matching coefficient, reflecting the suppression of abnormally high water usage and meeting the needs of preventing leakage and misinterpretation in actual water management.

[0114] This formula integrates three dimensions: consistency of water consumption, continuity of trends, and upper limit of reasonableness. It combines multi-source information into a single matching indicator through weighted summation, with all components normalized to a certain value. The range and weights are adjustable to adapt to different evaluation focuses in different scenarios; the exponential term applies a non-linear penalty to the deviation, enhancing the sensitivity to outliers; the indicator function introduces hard constraints to improve the system's ability to identify obvious abnormal readings.

[0115] S6: Combine the fit coefficient and trend matching coefficient to calculate the result reliability index of each initial identification result. Review the initial identification results with a result reliability index lower than the preset threshold, exclude abnormal data, and select the initial identification results that meet the requirements as the final water meter parameter monitoring results.

[0116] In this embodiment, based on the fitness coefficient Matching coefficient with trend The reliability index of each initial identification result is calculated using the following formula:

[0117]

[0118] in, This represents the reliability index of the results, with a value range of [value missing]. ; This is the fit coefficient; , To preset weights, And satisfy The weight settings meet the requirements. The reason is that the fit coefficient directly reflects the reliability of the image recognition front-end, such as the reading frame localization, image quality, and template matching. Its stability and accuracy are the foundation for the reliability of subsequent recognition results. If the reading frame localization is inaccurate or the image quality is poor, even if the subsequent trend matching performs well, the reliability of the image source of the reading itself is still low. Therefore, it is given higher weight. .

[0119] For this formula, the reliability index of the dependent variable outcome Used to characterize the overall reliability of a water meter reading recognition result, its value range is: The larger the value, the higher the accuracy of the reading in the image recognition stage and the better it matches historical water use trends, indicating high overall reliability and making it suitable for direct adoption as a monitoring result; conversely, if... If the value is small, it indicates that the reading may have an identification error, image quality problem, or be seriously inconsistent with water usage habits. It is a suspicious result and needs to be manually or secondarily verified to avoid false alarms.

[0120] Adaptability coefficient An increased value indicates that the water meter reading frame is more accurately positioned in the image, the template matching degree is higher, and the image preprocessing quality is better. Therefore, the source of the recognition result is more reliable, and the result reliability index is higher. Corresponding improvement; if If the image size is reduced, it may lead to errors in the identification of the reading area due to image blurring, occlusion, or abnormal shooting angle, thereby reducing the overall reliability. Consequently, it decreases; trend matching coefficient An increase indicates a high degree of consistency between the current reading and historical water usage patterns, reflecting normal user water usage behavior. Improve; if A decrease in the reading may be due to abnormal readings, sudden high water consumption, or identification errors leading to a trend mismatch, thus reducing the system's confidence in the result. The corresponding decrease.

[0121] This formula uses a linear weighted combination to integrate the two independent evaluation dimensions, the fit coefficient and the trend matching coefficient. Its structure is simple, clear, and easy to calculate and interpret. and It can be adjusted according to the actual application scenario to reflect the emphasis on different evaluation dimensions.

[0122] The reliability index of the result is lower than the preset reliability index threshold. The initial identification results are marked as suspicious results; where the reliability index threshold is... The determination method is as follows: by performing statistical analysis on the historical recognition result dataset, selecting manually verified recognition error samples under different combinations of fit coefficient and trend matching coefficient, and plotting the result reliability index. The distribution curve is given, and the cumulative proportion of erroneous samples does not exceed the preset false alarm tolerance rate. value as threshold Furthermore, by combining grid search with cross-validation fine-tuning on actual business data, we can achieve reasonable screening of suspicious results.

[0123] The initial identification results marked as suspicious are reviewed, specifically: the meter reading data of the same user for several historical meter reading cycles before the meter reading time point corresponding to the suspicious result is extracted, the historical average daily water consumption within these historical meter reading cycles is calculated, and it is compared with the current daily water consumption calculated based on the suspicious result; if the current daily water consumption does not exceed the preset multiple threshold of the historical average daily water consumption, the suspicious result is corrected to a reliable result, otherwise it is excluded.

[0124] The reliability index shall not be lower than All initial identification results, as well as the initial identification results corrected to reliable results, are output as the final water meter parameter monitoring results.

[0125] Please see Figure 4 A water meter parameter monitoring system based on image recognition, comprising:

[0126] The image enhancement processing module is used to acquire the original image data of the water meter to be identified, and to perform multi-dimensional random enhancement processing on the original image data to obtain enhanced image data.

[0127] The matching module is used to identify the water meter model corresponding to the enhanced image data based on the preset water meter type judgment rules, and match the corresponding reading frame structure template according to the water meter model, and simultaneously calculate the fit coefficient used to characterize the degree of matching between the structure template and the actual reading frame in the enhanced image.

[0128] The reading frame annotation module is used to dynamically adjust the annotation accuracy threshold based on the matched reading frame structure template, diagonal annotation rules and adaptation coefficient, to annotate the water meter reading frame area in the enhanced image data and determine the complete outline of the reading frame.

[0129] The adaptive scaling adjustment module is used to scale the enhanced image data with a complete outline reading frame by combining fixed-point scaling and center-point scaling to obtain standard image data that matches the preset input size of the water meter reading recognition model.

[0130] The identification and trend matching module is used to input the standard image data into a preset water meter reading identification model, output the initial identification result, and determine the trend matching coefficient to characterize the consistency between the initial identification result and the historical data trend based on the time series characteristics of the user's historical meter reading data.

[0131] The comprehensive verification module is used to combine the fit coefficient and the trend matching coefficient to calculate the result reliability index of each initial identification result, and to review the initial identification results with a result reliability index lower than a preset threshold, exclude abnormal data, and select the initial identification results that meet the requirements as the final water meter parameter monitoring results.

[0132] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0133] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0134] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0135] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for monitoring water meter parameters based on image recognition, characterized in that, Specifically, it includes: The original image data of the water meter to be identified is obtained, and the original image data is subjected to multi-dimensional random enhancement processing to obtain enhanced image data. Based on the preset water meter type judgment rules, the water meter model corresponding to the enhanced image data is identified, and the corresponding reading frame structure template is matched according to the water meter model. Simultaneously, the fit coefficient used to characterize the matching degree between the reading frame structure template and the actual reading frame in the enhanced image is calculated. The annotation accuracy threshold is dynamically adjusted based on the adaptation coefficient, and the water meter reading frame area in the enhanced image data is annotated according to the standard aspect ratio and diagonal annotation rules determined by the matched reading frame structure template, so as to determine the complete outline of the reading frame. For the enhanced image data with a complete outline reading frame marked, a scaling mechanism combining fixed-point scaling and center-point scaling is used to obtain standard image data that matches the input size of the preset water meter reading recognition model. The standard image data is input into a preset water meter reading recognition model, the initial recognition result is output, and a trend matching coefficient is determined based on the time series characteristics of the user's historical meter reading data to characterize the consistency between the initial recognition result and the historical data trend. Combining the fit coefficient and trend matching coefficient, the result reliability index of each initial identification result is calculated. The initial identification results with a result reliability index lower than the preset threshold are reviewed, abnormal data is excluded, and the initial identification results that meet the requirements are selected as the final water meter parameter monitoring results.

2. The water meter parameter monitoring method based on image recognition according to claim 1, characterized in that: The specific execution process of performing multi-dimensional random enhancement processing on the original image data is as follows: The original image data of the water meter to be identified is acquired. While keeping the reading frame intact, the cropping area is randomly selected and the cropping ratio is adjusted. Then, Gaussian blur processing is applied to the cropped image. The image is blurred to different degrees by randomly generating Gaussian kernel standard deviation. Next, random noise that conforms to a normal distribution is superimposed on the image pixel values. Then, the effect of natural light occlusion is simulated by generating a mask layer with random shape and transparency. Then, different lighting environments are simulated by adjusting the image brightness, contrast, and color temperature; then, random rotation and small-angle oscillation operations are performed to make the image rotate within a preset angle range; finally, the image's hue, saturation, and brightness parameters are randomly fine-tuned to simulate the color performance differences of different shooting devices. The execution order and intensity parameters of each enhancement operation are randomly generated in each round of processing, and the final output is enhanced image data with diverse shooting conditions. The intensity parameters include cropping ratio, Gaussian kernel standard deviation, noise intensity, mask transparency, brightness adjustment range, contrast adjustment range, color temperature offset, rotation angle range, hue offset, saturation adjustment range, and brightness adjustment range.

3. The water meter parameter monitoring method based on image recognition according to claim 2, characterized in that: Based on the enhanced image data, the dial shape, digital area distribution, scale line style and brand logo of the water meter are extracted. These features are then compared with a database of known water meter models using a pre-set classification model to identify the corresponding water meter model. After identifying the water meter model, the system matches the corresponding model's structure template from the preset reading frame structure template library based on the model index. This template includes the layout information of the length-width ratio, number of digits, and decimal point position of the water meter reading frame for that model. The fitting coefficient, which characterizes the degree of matching between the reading frame structure template and the actual reading frame in the enhanced image, is calculated synchronously, specifically including: The fit coefficient is calculated by weighted summation based on structural similarity, normalized positional deviation distance, and directional consistency parameter. Specifically: the structural similarity is obtained by extracting the Canny edge feature maps of the reading frame template and the actual reading frame of the image, and calculating their cosine similarity; the normalized positional deviation distance is the ratio of the Euclidean distance between the geometric center point of the reading frame template and the center point of the actual reading frame of the image, divided by the length of the image diagonal; the directional consistency parameter is obtained by calculating the absolute value of the angle between the principal directions of the template reading frame and the actual reading frame of the image, and normalizing it by dividing by 90 degrees; the fit coefficient is the weighted sum of structural similarity, the difference between one and the normalized positional deviation distance, and the directional consistency parameter, where each weighting coefficient is a preset weight value, and the sum of these weight values ​​equals 1.

4. The water meter parameter monitoring method based on image recognition according to claim 3, characterized in that: The annotation accuracy threshold is dynamically set based on the fit coefficient, specifically as follows: The annotation accuracy threshold is the sum of the difference between the basic annotation accuracy threshold and the adjustment coefficient multiplied by one minus the fit coefficient; wherein, the lower the fit coefficient, the higher the annotation accuracy threshold. Subsequently, based on the standard aspect ratio determined by the matching reading frame structure template, i.e., the ratio of template width to template height, the upper left corner of the water meter reading frame area is automatically detected and marked in the enhanced image data using a diagonal annotation method. and the bottom right corner ; Finally, based on the labeled top-left and bottom-right corner points, and according to the standard aspect ratio, the coordinates of the other two corner points of the reading frame are calculated using the following geometric relationships: The x-coordinate of the top right corner point is the same as that of the bottom right corner point, and the y-coordinate is equal to the y-coordinate of the top left corner point plus the difference between the x-coordinates of the bottom right corner point and the top left corner point, divided by the standard aspect ratio. The x-coordinate of the bottom left corner point is the same as that of the top left corner point, and the y-coordinate is equal to the y-coordinate of the bottom right corner point minus the difference between the x-coordinates of the bottom right corner point and the top left corner point, divided by the standard length-width ratio. This determines the complete outline of the reading frame formed by the four corner points: the upper left corner, the lower right corner, the upper right corner, and the lower left corner.

5. The water meter parameter monitoring method based on image recognition according to claim 1, characterized in that: For the enhanced image data with a complete outline of the reading frame, a scaling mechanism combining fixed-point scaling and center-point scaling is used to obtain standard image data that matches the preset input size of the water meter reading recognition model. The specific execution process is as follows: First, obtain the standard input dimensions of the preset water meter reading recognition model. Then, calculate the actual pixel percentage of the reading frame region in the enhanced image data with the complete outline of the reading frame. If this actual pixel percentage is less than a preset percentage threshold... The fixed-point scaling mechanism is adopted, which uses the geometric center point of the reading frame as a fixed reference and enlarges it proportionally according to the ratio coefficient between the standard input size and the actual reading frame size to ensure that the numbers and scales in the reading frame are completely preserved without distortion. If the actual pixel percentage is greater than or equal to the preset percentage threshold The center point scaling mechanism is adopted to keep the geometric center point of the reading frame unchanged and adjust the overall scaling ratio of the image according to the standard input size so that the reading frame area is adapted to the model input requirements. During the scaling process, bilinear interpolation algorithm is used to supplement pixels at the image edges to avoid jagged distortion. Finally, the output size is a standard image data that is completely matched with the model input size.

6. The water meter parameter monitoring method based on image recognition according to claim 5, characterized in that: The standard image data is input into the pre-trained water meter reading recognition model, and the initial water meter parameter recognition result is output based on the layout information of the number of digits and the position of the decimal point contained in the reading frame structure template. The initial recognition result includes a digital reading with decimal places. The water meter reading recognition model is obtained by supervised learning training through a convolutional neural network structure based on a large amount of manually annotated water meter reading image data. At the same time, the user's historical meter reading data for the past six months was retrieved, and its time series characteristics, including average daily water consumption, were extracted. Slope of water volume change trend and the upward period to The slope of the water volume change trend is obtained by linear fitting of historical meter reading data, and the upward period refers to the number of days from the most recent valid meter reading date to the current day; Based on the time-series characteristics of the historical meter reading data and the initial identification results, a trend matching coefficient is calculated, specifically as follows: The trend matching coefficient is obtained by weighted summation of three parts: the first part is the ratio of the smaller to the larger value of the current reading and the daily average water volume; the second part is an exponential function with the natural constant e as the base, whose exponent is the square of the difference between the current reading and the predicted reading divided by twice the square of the allowable error parameter; the third part is an indicator function term, which takes a value of 1 when the current reading is less than or equal to the daily maximum water volume threshold set according to historical data, and 0 otherwise, and is used to determine whether the current reading is within the reasonable water volume upper limit range. Multiply each of these three parts by a preset weight and then add them together to obtain a trend matching coefficient with a value between 0 and 1, and the sum of the weights of the three parts is 1.

7. The water meter parameter monitoring method based on image recognition according to claim 6, characterized in that: The reliability index of each initial identification result is calculated based on the fitness coefficient and trend matching coefficient, specifically as follows: The result reliability index is obtained by multiplying the fitness coefficient and the trend matching coefficient by their respective preset weights and then adding them together. The fitness coefficient has a greater weight than the trend matching coefficient. Both weights are greater than zero and their sum is one. The result reliability index ranges from 0 to 1. The reliability index of the result is lower than the preset reliability index threshold. The initial identification results were marked as suspicious results; The initial identification results marked as suspicious are reviewed, specifically: the meter reading data of the same user for several historical meter reading cycles before the meter reading time point corresponding to the suspicious result is extracted, the historical average daily water consumption within these historical meter reading cycles is calculated, and it is compared with the current daily water consumption calculated based on the suspicious result; if the current daily water consumption does not exceed the preset multiple threshold of the historical average daily water consumption, the suspicious result is corrected to a reliable result, otherwise it is excluded. The reliability index shall not be lower than All initial identification results, as well as the initial identification results corrected to reliable results, are output as the final water meter parameter monitoring results.

8. A water meter parameter monitoring system based on image recognition, characterized in that: The image recognition-based water meter parameter monitoring system is used to execute the image recognition-based water meter parameter monitoring method according to any one of claims 1-7, comprising: The image enhancement processing module is used to acquire the original image data of the water meter to be identified, and to perform multi-dimensional random enhancement processing on the original image data to obtain enhanced image data. The matching module is used to identify the water meter model corresponding to the enhanced image data based on the preset water meter type judgment rules, and match the corresponding reading frame structure template according to the water meter model, and simultaneously calculate the fit coefficient used to characterize the degree of matching between the structure template and the actual reading frame in the enhanced image. The reading frame annotation module is used to dynamically adjust the annotation accuracy threshold based on the matched reading frame structure template, diagonal annotation rules and adaptation coefficient, to annotate the water meter reading frame area in the enhanced image data and determine the complete outline of the reading frame. The adaptive scaling adjustment module is used to scale the enhanced image data with a complete outline reading frame by combining fixed-point scaling and center-point scaling to obtain standard image data that matches the preset input size of the water meter reading recognition model. The identification and trend matching module is used to input the standard image data into a preset water meter reading identification model, output the initial identification result, and determine the trend matching coefficient to characterize the consistency between the initial identification result and the historical data trend based on the time series characteristics of the user's historical meter reading data. The comprehensive verification module is used to combine the fit coefficient and the trend matching coefficient to calculate the result reliability index of each initial identification result, and to review the initial identification results with a result reliability index lower than a preset threshold, exclude abnormal data, and select the initial identification results that meet the requirements as the final water meter parameter monitoring results.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements a water meter parameter monitoring method based on image recognition as described in any one of claims 1-7.

Images