An intelligent water meter system based on AI visual recognition

Abstract

The application discloses an intelligent water meter system based on AI visual identification, relates to the field of intelligent water meters, and realizes full-automatic and high-precision collection of water meter readings through installation of an AI camera terminal, and robust segmentation and identification of the digit wheel digits and the pointer angle under complex illumination and stain environments; the application innovatively internally embeds synchronous error checking logic based on a mechanical transmission principle, can mark potential faults of inconsistency between the pointer and the digit wheel readings in real time, after data convergence of a cloud platform, not only constructs a time sequence database to support fine water consumption analysis, regional leakage early warning and scheduling optimization, but also can perform deep diagnosis on error marks reported by the terminal, quantitatively evaluate the mechanical fault grade, and generate a maintenance work order; the application does not need to transform the original meter, greatly reduces deployment and operation and maintenance costs, simultaneously upgrades traditional meter reading to intelligent management with automatic identification of the whole meter dial and intelligent decision support, and significantly improves the efficiency and reliability of water operation.

Description

Technical Field

[0001] This invention relates to the field of smart water meters, specifically to a smart water meter system based on AI visual recognition. Background Technology

[0002] In traditional water management, water meter readings are mainly collected manually on-site. This method has many drawbacks, including low efficiency, high labor costs, susceptibility to misreading and incorrect recording, inability to obtain data in real time, and inconvenience in door-to-door access. To solve this problem, automatic meter reading technology has gradually developed, mainly including two categories: one is electronic water meters based on physical signal conversion (such as direct-reading and pulse-type meters), which requires changing the internal mechanical structure of the water meter to convert the physical position of the dial wheel or pointer into an electrical signal. This has problems such as complex construction, unreliable electromechanical conversion, and difficult maintenance. It is also difficult to apply directly to the huge number of traditional mechanical water meters. The other is a visual meter reading solution based on image recognition. This solution takes pictures by installing a camera on the top of the meter and sends them back to the server for recognition. Although it achieves non-contact reading, in actual deployment, complex on-site conditions such as reflection of the meter glass, drastic changes in ambient light, dirty lenses, water stains, and image blurring caused by moisture inside the meter often lead to unstable recognition rates and insufficient reliability.

[0003] Furthermore, existing technologies generally focus only on the single objective of "digit reading," lacking the ability to identify the entire water meter panel, its mechanical deformation status, and online monitoring of abnormal indications, thus failing to provide early warnings of metering inaccuracies. At the same time, the data application level is relatively shallow, failing to extract regional water usage patterns from massive amounts of terminal data, identify pipeline leaks, and assist in scheduling decisions, thereby limiting the overall value of the system. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent water meter system based on AI visual recognition to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an intelligent water meter system based on AI visual recognition, comprising an AI camera terminal and a cloud-based recognition and management platform communicatively connected to the AI ​​camera terminal;

[0006] The AI ​​camera terminal includes:

[0007] The image acquisition and preprocessing module acquires the original image of the water meter dial, performs adaptive optimization processing, and outputs an optimized dial image.

[0008] The dial area segmentation module performs region segmentation on the optimized dial image based on an image segmentation model. The segmented regions include the counting and statistics region and the pointer display region.

[0009] The identification and error verification module is used to segment the region and execute the identification verification strategy. The identification verification strategy includes: performing digit wheel digit recognition on the counting and statistics region, measuring the pointer angle and converting the reading on the pointer display region, and performing mechanical synchronization error logic verification based on the reading status of the pointer of the largest unit of measurement and the recognition status of the unit digit of the digit wheel, and generating a structured reading data packet with error markers.

[0010] The cloud-based identification and management platform includes a mechanical fault diagnosis module, which monitors the mechanical synchronization error markers of each water meter, initiates a comparative analysis strategy based on pointer images when the unit digit of the water meter changes, obtains the mechanical error amount, and generates graded fault warnings based on the error amount and its persistence.

[0011] As a preferred method, the region segmentation method includes the following steps:

[0012] The deep learning segmentation model is invoked to perform pixel-level semantic segmentation on the optimized global dial image, and the output includes a mask for the background, the counting statistics region, and the pointer display region.

[0013] Within the pointer display area, each individual pointer is further identified, and its rotation center point and pointer tip are located; within the counting and statistics area, the precise boundary of the units digit window is located; and simultaneously, the zero mark on the dial is detected.

[0014] Based on the segmentation mask, the counting and statistics area and the pointer display area are cut out respectively, and the pointer display area is geometrically corrected to output standardized data.

[0015] Preferably, the method for measuring the pointer angle and converting the reading in the pointer display area includes the following steps:

[0016] For each pointer, based on the rotation center coordinates of the pointer and its corresponding zero-scale point coordinates provided by the dial area segmentation module, a virtual reference line passing through these two points is established in the image as a reference line for angle measurement.

[0017] Identify the position of the pointer's tip and calculate the clockwise angle between the vector pointing from the center of rotation to the pointer's tip and the vector of the reference line;

[0018] Based on the unit of measurement and total scale range represented by the pointer, the included angle is proportionally converted into the corresponding actual reading value.

[0019] Preferably, the method for verifying mechanical synchronization error logic includes the following steps:

[0020] Obtain the reading status of the pointer of the largest unit of measurement and determine whether it indicates "0" or returns to zero within a preset threshold range;

[0021] Obtain the recognition status of the units digit in the character wheel digit recognition result, and determine whether the units digit is in a stable display state, rather than in a critically blurred state that is changing or about to change.

[0022] If the reading status of the maximum unit of measurement pointer is determined to be zero, while the recognition status of the units digit is determined to be stable and unchanged, then a mechanical synchronization error is determined to exist, and an error mark is generated for the current reading record.

[0023] Preferably, the comparative analysis strategy includes the following steps:

[0024] Retrieve at least two key images from continuously tagged water meters, including an image containing the error mark and an image showing a normal carry-over in the unit digit;

[0025] For the maximum unit of measurement pointer, subpixel-level edge detection is performed in the two key images respectively, and the actual angle θ_actual between it and the control line is accurately calculated;

[0026] Based on the mechanical transmission ratio of the water meter model, determine the theoretical angle (θ_theoretical) that the pointer of the maximum measuring unit should have when the units digit is normally carried over.

[0027] The difference between the actual included angle and the theoretical included angle, Δθ = |θ_actual - θ_theoretical|, is calculated as the mechanical error.

[0028] Based on the mechanical error Δθ and the measurement unit and total scale range represented by the maximum measurement unit pointer, the error water volume is calculated.

[0029] Preferably, the graded fault early warning includes:

[0030] When the mechanical error is within the set first-level error threshold range and is stable, and the error markers appear continuously and regularly, it is determined to be a first-level warning. The first-level error threshold range is when the error is relatively small.

[0031] If the mechanical error is within the set secondary error threshold range and shows a slow increasing trend during continuous monitoring, it is determined to be a secondary warning, which is inferred to be obvious wear of the gear set or excessive clearance of the transmission shaft. The secondary error threshold range is where the error is relatively large.

[0032] If the mechanical error is within the set three-level error threshold range, and the frequency of error markers increases sharply in a short period of time, it is determined to be a three-level alarm, where the three-level error threshold range is the range where the error is extremely large.

[0033] Preferably, the image acquisition and preprocessing module is specifically used for:

[0034] After waking up, the camera quickly perceives the environment, assesses the overall brightness, contrast, and presence of strong light spots, and dynamically adjusts the camera parameters based on the assessment results.

[0035] In scenes with extremely low light or slight vibration, multiple images are captured rapidly and continuously, and multi-frame noise reduction and image fusion algorithms are used to synthesize images with clearer details and less noise.

[0036] Targeted image enhancement includes de-reflection processing using deep learning-based restoration models or polarization analysis principles, enhancing local contrast using adaptive histogram equalization algorithms, and weakening interference from fixed stains through color and texture analysis.

[0037] Preferably, the AI ​​camera terminal further includes a first communication module, which is used to perform the following steps:

[0038] The structured data packet generated by the identification and error verification module, which includes the final reading, error mark status, identification confidence and timestamp, is compressed and encapsulated together with key area image fragments and device status information.

[0039] The compressed and packaged data is uploaded to the cloud-based identification and management platform via a wireless communication network.

[0040] After the data upload is completed, the AI ​​camera terminal is controlled to enter a deep sleep state to reduce power consumption.

[0041] Preferably, the cloud-based identification and management platform further includes a regional water use analysis module, which includes the following operating process:

[0042] Using independent metering areas or communities as units, aggregate the water consumption of all water meters within the area during the same time period to plot the regional water load curve;

[0043] The curves are analyzed using an unsupervised learning algorithm to identify normal water usage patterns and detect abnormal water usage patterns.

[0044] When an abnormally high minimum flow rate is detected in a region, the leaking pipe section is initially located by combining the pipeline topology and water meter location through correlation analysis, and a leak detection work order is generated.

[0045] The peak water usage forecast results for the region are sent to the SCADA system or smart valve controller to perform time-of-use pressure regulation.

[0046] Preferably, the cloud-based identification management platform further includes a data aggregation and storage module for performing the following steps:

[0047] The high-concurrency access service receives structured data packets from various AI camera terminals, containing readings, timestamps, error markers, and device status.

[0048] The received data is cleaned and verified, its format and integrity are checked, and invalid data caused by transmission errors is filtered out and converted into a unified spatiotemporal data model.

[0049] The cleaned reading data, timestamps, error markers, and equipment status information are linked and integrated with water meter asset information and GIS geographic information; the water meter asset information includes water meter model, installation location, and user information, and the GIS geographic information includes the pipeline network zone and pressure zone to which it belongs;

[0050] The integrated data is stored in a spatiotemporal database, and a time-series index is created for each water meter reading record to support efficient time-based queries and trend analysis.

[0051] In summary, the beneficial effects of this invention are:

[0052] This invention achieves fully automated, high-precision acquisition and intelligent management of water meter readings through AI visual recognition and cloud-edge-device collaboration. It can be quickly deployed without modifying the original mechanical meter structure, significantly reducing installation and maintenance costs. Through adaptive image optimization and deep learning segmentation technology, it can stably identify readings under various lighting conditions and dirt interference, improving the reliability and accuracy of data acquisition. The innovative introduction of a mechanical synchronization error marking mechanism enables real-time detection of potential faults in the pointer and digit wheel transmission. Deep diagnostics in the cloud provide early warning and quantitative assessment of faults, effectively preventing metering inaccuracies. Simultaneously, the platform-level regional water use analysis function can identify abnormal water use patterns and pipeline leakage, assisting in optimizing water supply scheduling and water conservation management, thereby comprehensively improving the refinement and intelligence of water management. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of the invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 This is a schematic diagram of the overall process framework of an AI-based intelligent water meter system according to the present invention.

[0055] Figure 2 This is a schematic diagram of the process framework structure of the dial area segmentation module in an AI-based intelligent water meter system according to the present invention.

[0056] Figure 3 This is a schematic diagram of the process framework structure of the identification and error verification module in an AI-based intelligent water meter system according to the present invention.

[0057] Figure 4 This is a schematic diagram of the process framework of the mechanical fault diagnosis module in an AI-based intelligent water meter system according to the present invention.

[0058] Figure 5 This is a schematic diagram of the installation of an AI camera terminal in a smart water meter system based on AI visual recognition according to the present invention.

[0059] Figure 6 This is a schematic diagram of the water meter panel interface structure in an AI-based intelligent water meter system according to the present invention.

[0060] Figure 7 This is a schematic diagram showing the recognition interface of the cloud-based recognition management platform in an AI-based intelligent water meter system according to the present invention.

[0061] Figure 8 This is a schematic diagram showing the APP-side recognition interface in an AI-based visual recognition smart water meter system according to the present invention.

[0062] Figure 9 This is a schematic diagram of water meter installation in a specific operating example of an AI-based visual recognition smart water meter system according to the present invention. Detailed Implementation

[0063] The present invention will now be described in further detail with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. These drawings are simplified schematic diagrams, which are only used to illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0064] To facilitate understanding of the present invention, a more complete description of the invention will be given below with reference to the accompanying drawings, which illustrate several embodiments of the invention. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of the invention will be more thorough and complete.

[0065] All features disclosed in this specification, or all steps in all disclosed methods or processes, may be combined in any way, except for mutually exclusive features and / or steps.

[0066] Any feature disclosed in this specification (including any appended claims, abstract, and drawings) may be replaced by other equivalent or similar features for a similar purpose, unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is merely one example of a series of equivalent or similar features.

[0067] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection, a direct connection, or an indirect connection through an intermediate medium; they can refer to the internal communication of at least two elements or the interaction relationship of at least two elements, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0068] Please see Figures 1-9 This invention provides an embodiment of an AI-based intelligent water meter system. Through a collaborative process of "edge sensing – edge recognition – cloud intelligence," it achieves fully automatic, high-precision, and highly reliable acquisition and intelligent management of water meter readings. Specifically, it includes an AI camera terminal and a cloud-based recognition management platform. The AI ​​camera terminal is fixed to the outside of the dial glass of an existing mechanical water meter using a snap-on or magnetic method. Figure 5 Without modifying the original table structure, the terminal automatically completes network registration after power-on, reporting the device ID, location information, and initial status to the cloud management platform to complete binding and activation.

[0069] Specifically, the AI ​​camera terminal includes an image acquisition and preprocessing module, a dial area segmentation module, a recognition and error verification module, and a first communication module;

[0070] The image acquisition and preprocessing module is used to acquire the original image of the watch face according to a preset period or remote command wake-up, and perform adaptive optimization processing. The process is as follows:

[0071] After the camera is woken up, it first performs a quick environmental perception, assessing the overall brightness, contrast, and presence of strong light spots in the current environment.

[0072] Based on the evaluation results, camera parameters such as exposure time, gain, and software exposure value are dynamically adjusted to obtain original images with balanced brightness in dimly lit water meter wells or in direct sunlight outdoors, avoiding overexposure or underexposure.

[0073] Multi-frame fusion and deblurring:

[0074] In extremely low light conditions or in the presence of slight vibrations, such as water flow in a pipe, the terminal will quickly and continuously capture multiple images.

[0075] By employing multi-frame noise reduction and image fusion algorithms, a more detailed and less noisy image is synthesized, effectively overcoming the blurring problem.

[0076] Targeted image enhancement:

[0077] Anti-glare processing: Identify highlight areas in the image, use deep learning-based restoration models or polarized light analysis principles to infer and remove reflections, and restore the obscured dial numbers or hands.

[0078] Contrast and Sharpening: Adaptive histogram equalization and other algorithms are used to enhance the local contrast of the text wheel edges, pointers and tick marks, making them more distinct from the background.

[0079] Dirt / water stain suppression: Through color and texture analysis, it distinguishes between fixed stains attached to the dial glass and dial content that needs to be identified, and weakens the interference of stains in the pre-processing.

[0080] It should be noted that the dial area segmentation module uses a built-in image segmentation model to accurately segment the counting and statistics area and the pointer display area from the optimized image. Specifically:

[0081] After obtaining the optimized global dial image, the terminal calls the built-in lightweight deep learning segmentation model, such as an encoding / decoding structure based on MobileNet or CNN, to perform pixel-level classification:

[0082] The model classifies each pixel in the image into a predefined category, referencing... Figure 6 The core outputs include:

[0083] Background: Watch case, environment, etc.

[0084] Counting and Statistics Area: A display window containing the numbers of all the number wheels.

[0085] Pointer display area: contains all pointers and their corresponding circular dials.

[0086] Brand / logo area (optional): Used to assist in determining the water meter model.

[0087] Instance segmentation and keypoint localization are used to achieve error detection.

[0088] Within the pointer display area, the model further identifies each individual pointer and locates its rotation center point and pointer tip.

[0089] Within the counting and statistical region, the model pinpoints the precise boundary of the units digit window. This is a crucial prerequisite for executing the subsequent "error marking" logic.

[0090] At the same time, the model will detect the zero mark on the dial.

[0091] Region extraction and normalization:

[0092] Based on the segmentation mask, the "counting and statistics area" sub-image and the "pointer display area" sub-image are cropped from the original image respectively.

[0093] Perform geometric correction on the "pointer display area" to ensure that the dial is a perfect circle, so as to accurately establish the subsequent polar coordinate system for calculating the angle between the pointer and the "reference line".

[0094] This will output the following standardized data:

[0095] Sub-image A (Counting Statistics Area): A rectangular image containing only the digits of the wheel, which has been angled.

[0096] Sub-image B (pointer display area): A circular area image containing the complete pointer and tick marks, with the center aligned.

[0097] Key metadata: coordinates of the pointer rotation center, coordinates of the units digit window, and coordinates of the zero mark.

[0098] It is worth mentioning that, in this embodiment, the recognition and error verification module is used to perform recognition operations on the segmented regions and execute a numerical extraction strategy with error detection. The specific process is as follows:

[0099] 1. Digit recognition using a digit wheel:

[0100] Input: The segmented "counting statistics region" subgraph.

[0101] First, based on prior knowledge or a lightweight detection network, accurately locate the boundaries of each digital window.

[0102] For each image within a window, a high-precision OCR model, typically a quantized CNN, is used for classification, outputting the probability of a number from 0 to 9.

[0103] The model specifically records and outputs the recognition results of the "units digit window," which will serve as a key input for subsequent logical verification.

[0104] Output: An ordered sequence of numbers, such as [1,2,3,4,5,6], representing 123.456m³, with the units digit explicitly identified.

[0105] 2. Pointer angle measurement and reading conversion:

[0106] Input: The segmented "pointer display area" sub-image, and the coordinates of the "pointer rotation center" and the "zero scale point" corresponding to each pointer obtained in the segmentation stage.

[0107] Establish a "reference line": In the image coordinate system, draw a virtual straight line for each pointer, which strictly passes through the "pointer rotation center" and the "zero mark point of the pointer". This line is the baseline (0-degree line) for angle measurement.

[0108] Identify the position of the pointer tip. Calculate the clockwise angle between the vector from the "rotation center" to the "pointer tip" and the "reference line" vector. This calculation is usually performed in polar coordinates with an accuracy of over 0.1 degrees.

[0109] Convert readings: Based on the unit of measurement represented by the pointer, such as 0.1m³ or 0.01m³, convert the angle value, such as 36 degrees, to the actual reading, such as 0.036m³, according to the total scale range. At the same time, determine the pointer with the largest unit of measurement among all pointers, such as the 0.1m³ pointer.

[0110] Output: The reading of each pointer, such as [0.0, 0.3, 0.06], and the current state of the pointer with the largest unit of measurement.

[0111] 3. Mechanical Synchronization Error Marker: Based on the mechanical principle of the water meter, when the pointer of the largest measuring unit (e.g., 0.1m³) rotates a full circle (from scale 9 to 0), it should drive the units digit of the counting area forward by one value. Therefore, at the instant the image is captured, if the pointer happens to point to "0", that is, coincide with the "control line", allowing for a very small error threshold, such as ±2 degrees, then at the next instant, the units digit should change or be in a critical state of about to change.

[0112] The algorithm checks the following two conditions in real time:

[0113] First condition: Is the reading of the pointer of the largest unit of measurement "0", that is, does it coincide with the control line?

[0114] The second condition is whether the currently identified "unit digit" is stable, not in a blurred or changing state, and has not reached the critical value for carry-over, such as the absence of overlapping at the edges of the digit.

[0115] If the first condition is true and the second condition is false, that is, the pointer has returned to zero, but the units digit does not appear to have changed or show any signs of change, the algorithm determines that there is a potential mechanical failure risk of "dissynchronization of the digit wheel and pointer transmission" in this reading.

[0116] Generate a flag: Immediately append a status bit of "mechanical synchronization error flag" to this reading record, such as error_flag=1.

[0117] 4. Result fusion and output: Logically concatenate the integer part of the reading from the digit wheel with the decimal part of the reading from the pointer to generate a complete and accurate water meter reading, such as 123.456.

[0118] The final reading, whether the identification includes a "mechanical synchronization error mark", the identification confidence level, and the timestamp are encapsulated into a structured data packet.

[0119] Local cache: To cope with momentary network interruptions, the data packet will be temporarily stored in local flash memory.

[0120] The technical value and output of this step:

[0121] Output: A structured data object containing {timestamp, instrument ID, reading, error flag, confidence level}.

[0122] It should be noted that in this embodiment, the first communication module compresses and encapsulates the reading results, error marker status, key area image segments, and device status, and uploads them to the cloud-based identification management platform via a wireless network. Figure 7 It then enters a deep hibernation state.

[0123] As for the cloud-based identification and management platform, which includes a data aggregation and storage module, a regional water use analysis module, and a mechanical fault diagnosis module, specifically:

[0124] The data aggregation and storage module is used to receive and store data from various AI camera terminals, and to build a time-series database of time and water volume, specifically including:

[0125] Data access and cleaning: Receive structured data packets uploaded from millions of terminals through a high-concurrency access service;

[0126] First, data cleaning and verification are performed to check the data format and integrity, filter out invalid data caused by transmission errors, and convert the data into a unified spatiotemporal data model.

[0127] Association storage and index construction: Information such as readings, timestamps, error markers, and equipment status are associated with water meter asset information and GIS geographic information and stored in a spatiotemporal database. Water meter asset information includes model, installation location, and user information; GIS geographic information includes the pipeline network zone and pressure zone to which it belongs.

[0128] A high-performance time-series index is built for each water meter, supporting millisecond-level queries of all its historical reading records, laying the foundation for trend analysis and pattern recognition.

[0129] It is worth mentioning that, in this embodiment, the regional water use analysis module is used to perform cluster analysis on the data of water meters within the same geographical or logical partition, identify regional water use patterns and anomalies, and generate corresponding water use scheduling strategies, specifically including the following steps:

[0130] Step 1: Spatiotemporal Clustering and Pattern Recognition

[0131] Regional analysis: Taking independent metering areas or communities as units, the platform automatically aggregates the water consumption of all water meters in the same area during the same period and plots the "regional water load curve".

[0132] Algorithm Engine: Utilizes unsupervised learning algorithms, such as cluster analysis and anomaly detection algorithms, to analyze the curves.

[0133] Identify normal water usage patterns: such as morning and evening peak hours in residential areas, and daytime peak hours in office areas.

[0134] Detect abnormal water usage patterns: If a region experiences a continuous, stable, low flow rate in the early morning hours, it may indicate background leakage; or if there is a short-term peak flow rate, it may indicate a pipe burst or water theft.

[0135] Step 2: Strategy Generation and Automatic Scheduling

[0136] Leakage location and early warning: When an abnormal increase in the minimum flow rate in a region is detected, the platform can combine the pipeline topology and water meter location to preliminarily locate the pipe section where a leak may occur through correlation analysis, and generate a work order to be dispatched to the leak detection team.

[0137] Pressure optimization scheduling: The platform can send the peak water consumption forecast results of the region to the SCADA (Supervisory and Data Acquisition) system or smart valve controller through a standard interface (such as API) to suggest or automatically execute time-sharing pressure regulation, thereby reducing the risk of pipe bursts and energy consumption while ensuring water supply.

[0138] Water usage report generation: Automatically generates periodic water usage analysis reports for commercial users or properties to assist in their water conservation management.

[0139] It should be noted that, in this embodiment, the mechanical fault diagnosis module converts the "error markers" reported by the terminal into executable maintenance insights, specifically including the following steps:

[0140] 1. Continuous error monitoring and case library construction:

[0141] The platform maintains an "error status window" for each water meter. For example, "three consecutive readings are marked" is defined as the threshold for triggering in-depth diagnostics.

[0142] Once triggered, the system automatically adds the water meter to the "suspected fault list" and creates a diagnostic case to begin collecting relevant data.

[0143] 2. Multi-timepoint image retrospective and quantitative analysis:

[0144] Keyframe retrieval: The platform intelligently retrieves image data from two key moments in the water meter's historical storage:

[0145] The image at the recent "error mark" moment: the pointer returns to zero while the units digit remains unchanged.

[0146] Images of the moments before and after the last "normal carry-over of the units digit": for example, the position of the pointer when the units digit changes from 3 to 4, especially the pointer of the largest unit.

[0147] Precise angle calculation: On the more powerful computing resources in the cloud, subpixel-level edge detection and angle analysis are performed on the retrieved image to accurately calculate the actual angle θ_actual between the maximum unit pointer and the "control line".

[0148] Theoretical value comparison: Based on the mechanical transmission ratio of the water meter model, determine the theoretically correct included angle θ_theoretical when the largest unit pointer should be when carrying over to the units digit. It should usually be very close to 0 degrees, that is, strictly pointing to the 0 mark.

[0149] 3. Fault diagnosis, classification, and early warning:

[0150] Error calculation: Calculate the difference Δθ = |θ_actual - θ_theoretical|, and convert it into a specific measurement error volume through the angle-flow relationship. For example, if the pointer deviates by 5 degrees each time it returns to zero, it is equivalent to under-counting by 0.014 m³ each time.

[0151] 4. Fault Root Cause Inference and Classification:

[0152] Level 1 Warning (Slight Inaccuracy): Δθ is small and stable. This is likely due to slight pointer loosening or initial gear wear. Replacement is recommended at the next cycle check.

[0153] Level 2 Warning (Moderate Fault): Δθ is large or slowly increasing. This is inferred to be due to significant wear of the gear set or excessive clearance in the drive shaft. Planned replacement is recommended.

[0154] Level 3 Alarm (Severe Fault / Failure): Δθ is extremely high, or the error marker frequency increases sharply. This is inferred to be due to gear slippage, pointer jamming, or transmission mechanism breakage. The system immediately generates a high-priority repair work order and marks the water meter data as unreliable.

[0155] Warning issuance: Diagnostic conclusions, error amounts, and recommended measures are automatically pushed to the water company's operation and maintenance work order system or the mobile APP of relevant personnel, and the location of the fault table is highlighted on the geographic information system (GIS) map.

[0156] Below is a specific operational example: A water company has installed AI-powered camera smart water meters for 5,000 households in the communities it manages. (Refer to...) Figure 9 The installation process will be demonstrated using user A's water meter (device ID: Meter_A001) as an example, showing the complete process from routine meter readings to the diagnosis of mechanical faults.

[0157] Phase 1: Normal and stable operation, from day 1 to day 10;

[0158] Day 1, Installation and Activation;

[0159] Step 1: Installation and Activation

[0160] The technicians magnetically attached the terminal device to the outside of the dial of User A's mechanical water meter (model: LXS-15E, a combination of digit wheel and pointer).

[0161] The device automatically powers on, registers with the cloud platform via the NB-IoT network, and reports the following information: {Device ID: Meter_A001, Location: Unit 1, Building 3, Yangguang Community, Status: Normal}.

[0162] The cloud platform binds it to user A's account and issues the default parameter: collect data once a day at 2:00 AM.

[0163] Day 2 to Day 10 (Daily readings collected)

[0164] Perform steps 2 and 3 every day at 2:00 AM:

[0165] Step 2: Image Acquisition and Segmentation

[0166] The Meter_A001's sensor detects darkness and automatically turns on the miniature LED fill light.

[0167] The camera captures a full-view image. Due to slight water stains, the algorithm activates a stain removal module for processing.

[0168] Precise positioning of the segmentation model: ① Word wheel area (5 integer digits, 3 decimal digits), ② Pointer area (3 pointers: 0.1m³, 0.01m³, 0.001m³).

[0169] Output normalized word wheel subgraph and pointer subgraph, and extract the coordinates of the pointer rotation center and zero mark.

[0170] Step 3: Identification and Verification

[0171] Character wheel recognition: The result is [0,1,2,3,4,5,6,7], which means the reading is 123.456m³, and the units digit is 3.

[0172] Pointer reading:

[0173] Establish a reference line for the 0.1m³ pointer, calculate its included angle as 324 degrees, and the reading is 0.090m³ (324 / 360*0.1).

[0174] The pointer reading for 0.01m³ is 0.005m³, and the pointer reading for 0.001m³ is 0.0007m³.

[0175] Logical verification: The maximum unit pointer (0.1m³) reading is 0.090 (non-zero), therefore the error flag is not triggered.

[0176] Output result: Overall reading 123.456 + 0.090 + 0.005 + 0.0007 = 123.5517 ≈ 123.552 m³, consistent with the logic of the character wheel display. Data packet generated: {Time: 2023-10-02 02:00:00, Reading: 123.552, Error flag: 0}.

[0177] Step 4: Compress the data and upload it to the cloud.

[0178] Step 5: The cloud platform stores data normally and aggregates water consumption for the community; no anomalies are found. The error flag counter for Meter_A001 remains at 0.

[0179] Phase 2: Initial Manifestation of Mechanical Failure and Continuous Marking (Day 11 to Day 13)

[0180] Background: Inside user A's water meter, the gear set driving the units digit wheel has experienced slight wear, resulting in a minor delay in transmission.

[0181] 2:00 AM on Day 11

[0182] Steps 2 and 3 were executed normally.

[0183] Key identification results:

[0184] Reading of the digit wheel: 124.067m³ (the units digit has changed from 3 to 4).

[0185] Pointer reading: 0.1 m³. The angle between the pointer and the reference line is 2 degrees, and the reading is 0.0006 m³ (close to 0).

[0186] Logical verification triggered: The algorithm determines that the maximum unit pointer has basically returned to zero (within the set ±2 degree error threshold), but at this time the units digit has changed from 3 to 4. Theoretically, the action of the pointer returning to zero should "drive" the change of the units digit, but in reality, the change of the digit wheel occurs first.

[0187] System action: Generate a "mechanical synchronization error flag" for this reading. The data packet is: {Time: 2023-10-11 02:00:00, Reading: 124.068, Error flag: 1}.

[0188] After receiving the data, the cloud platform sets the error continuity counter of Meter_A001 to 1.

[0189] Day 12, Day 13

[0190] This situation occurred repeatedly. Each time the 0.1m³ pointer pointed to 0, the units digit had already been carried over.

[0191] The cloud platform error counter has accumulated to 3, triggering the "3 consecutive markings" deep diagnostic threshold.

[0192] Phase 3: In-depth cloud-based diagnostics and early warning (Day 13, daytime)

[0193] Step 5 (In-depth diagnostic process) starts automatically:

[0194] Case creation: The platform creates a fault diagnosis case for Meter_A001 and retrieves key images from its historical storage.

[0195] Backtracking analysis:

[0196] Retrieve the image from day 10 (normal state): When the reading is 123.999 m³, the 0.1 m³ pointer reading is 0.099 m³, which is about to return to zero.

[0197] Retrieve image from day 11 (first error): reading is 124.068 m³, 0.1 m³ pointer reading is 0.0006 m³ (angle 2 degrees).

[0198] Quantitative calculation: In the cloud-based analysis engine, sub-pixel-level analysis was performed on the image from day 11. The actual angle between the 0.1m³ pointer and the control line was measured to be 1.8 degrees, while the theoretical angle should be 0 degrees. Therefore, the error angle Δθ = 1.8 degrees.

[0199] Converted to measurement error: Each time the pointer is zeroed, it results in an under-counting of water volume of (1.8 / 360)*0.1≈0.0005m³.

[0200] Fault diagnosis and early warning:

[0201] Diagnostic conclusion: The error angle is fixed and small (1.8 degrees), which is consistent with the characteristics of "slight wear of gear set", causing the pointer movement to lag behind the number wheel advance.

[0202] Impact assessment: For every 1 m³ of water used by the user, this fault will result in approximately 0.005 m³ (0.5%) of water volume not being counted (because the 0.1 m³ pointer has an error of 0.0005 m³ per revolution).

[0203] Early Warning Generation: The platform automatically generates a Level 2 early warning work order, with the following content:

[0204] Title: Water meter mechanical failure warning, gear wear;

[0205] Equipment: Meter_A001, Sunshine Community 3-1-101;

[0206] Fault type: The pointer and the number wheel are out of sync;

[0207] Quantization error: Each zeroing operation results in an underestimation of approximately 0.0005 m³, with a cumulative impact of approximately 0.5%.

[0208] Recommended measures: Planned replacement, preferably within 3 months;

[0209] Urgency level: Medium

[0210] Strategy output: The work order is automatically pushed to the water company's mobile operation and maintenance APP, and the Meter_A001 icon is turned yellow on the geographic information system (GIS) map.

[0211] Meanwhile, the platform added a note to the regional water use analysis report of "Sunshine Community": There is one faulty meter, which has a slight impact of about 0.01 m³ / h on the accuracy of the minimum nighttime flow calculation for the independent metering area.

[0212] Phase 4: Operations and Maintenance Response and System Learning (Day 14 and beyond)

[0213] After receiving the work order, the maintenance personnel will include it in the batch replacement plan for the next quarter.

[0214] On the 90th day, the maintenance personnel replaced Meter_A001 on-site. After the new meter was installed, the error markers no longer appeared.

[0215] The cloud platform removed Meter_A001 from the fault list and automatically archived this case (including all images and diagnostic data) into the "gear wear" training sample library.

[0216] In the next model iteration training, this case will be used to optimize the AI ​​segmentation model and error detection logic, making it more sensitive to the identification of similar early faults in the future.

[0217] In summary, by leveraging AI visual recognition and cloud-edge-device collaboration, fully automated, high-precision water meter reading acquisition and intelligent management have been achieved. The key benefits include: rapid deployment without modifying existing mechanical meter structures, significantly reducing installation and maintenance costs; stable reading recognition under various lighting and dirt conditions through adaptive image optimization and deep learning segmentation technologies, improving data acquisition reliability and accuracy; the innovative introduction of a mechanical synchronization error marking mechanism, enabling real-time detection of potential faults in the pointer and digit wheel transmission, and providing early warning and quantitative assessment of faults through cloud-based deep diagnostics, effectively preventing metering inaccuracies; and platform-level regional water usage analysis identifying abnormal water usage patterns and pipeline leaks, assisting in optimizing water supply scheduling and water conservation management, thereby comprehensively improving the refinement and intelligence of water management.

[0218] The above description is merely a specific embodiment of the invention, but the scope of protection of the invention is not limited thereto. Any variations or substitutions conceived without inventive effort should be included within the scope of protection of the invention. Therefore, the scope of protection of the invention should be determined by the scope defined in the claims.

Claims

1. A smart water meter system based on AI visual recognition, characterized in that: Includes an AI camera terminal and a cloud-based recognition and management platform that is communicatively connected to the AI ​​camera terminal; The AI ​​camera terminal includes: The image acquisition and preprocessing module acquires the original image of the water meter dial, performs adaptive optimization processing, and outputs an optimized dial image. The dial area segmentation module performs region segmentation on the optimized dial image based on an image segmentation model. The segmented regions include the counting and statistics region and the pointer display region. The identification and error verification module is used to execute an identification verification strategy in the segmented region. The identification verification strategy includes: performing digit recognition on the counting and statistics region, measuring the pointer angle and converting the reading on the pointer display region, and performing mechanical synchronization error logic verification based on the reading status of the pointer of the largest unit of measurement and the recognition status of the units digit of the digit wheel. Specifically, it includes: Obtain the reading status of the pointer of the largest unit of measurement and determine whether it indicates "0" or returns to zero within a preset threshold range; Obtain the recognition status of the units digit in the character wheel digit recognition result, and determine whether the units digit is in a stable display state, rather than in a critically blurred state that is changing or about to change. If the reading status of the maximum unit of measurement pointer is determined to be zero, and the recognition status of the units digit is determined to be stable and unchanged, then a mechanical synchronization error is determined to exist, and a structured reading data packet with error markers is generated for the current reading record. The cloud-based identification and management platform includes a mechanical fault diagnosis module, which monitors the mechanical synchronization error markers of each water meter, initiates a comparative analysis strategy based on pointer images when the unit digit of the water meter changes, obtains the mechanical error amount, and generates graded fault warnings based on the error amount and its persistence.

2. The smart water meter system based on AI visual recognition according to claim 1, characterized in that: The region segmentation method includes the following steps: The deep learning segmentation model is invoked to perform pixel-level semantic segmentation on the optimized global dial image, and the output includes a mask for the background, the counting statistics region, and the pointer display region. Within the pointer display area, each individual pointer is further identified, and its rotation center point and pointer tip are located; within the counting and statistics area, the precise boundary of the units digit window is located; and simultaneously, the zero mark on the dial is detected. Based on the segmentation mask, the counting and statistics area and the pointer display area are cut out respectively, and the pointer display area is geometrically corrected to output standardized data.

3. The smart water meter system based on AI visual recognition according to claim 2, characterized in that: The method for measuring and converting pointer angles in the pointer display area includes the following steps: For each pointer, based on the rotation center coordinates of the pointer and its corresponding zero-scale point coordinates provided by the dial area segmentation module, a virtual reference line passing through these two points is established in the image as a reference line for angle measurement. Identify the position of the pointer's tip and calculate the clockwise angle between the vector pointing from the center of rotation to the pointer's tip and the vector of the reference line; Based on the unit of measurement and total scale range represented by the pointer, the included angle is proportionally converted into the corresponding actual reading value.

4. The smart water meter system based on AI visual recognition according to claim 3, characterized in that: The comparative analysis strategy includes the following steps: Retrieve at least two key images from continuously tagged water meters, including an image containing the error mark and an image showing a normal carry-over in the unit digit; For the maximum unit of measurement pointer, subpixel-level edge detection is performed in the two key images to accurately calculate its actual angle with the control line; Based on the mechanical transmission ratio of the water meter model, determine the theoretical included angle that the pointer of the maximum measuring unit should have when the units digit is normally carried over; The difference between the actual included angle and the theoretical included angle, Δθ = |actual included angle - theoretical included angle|, is calculated as the mechanical error. Based on the mechanical error Δθ and the measurement unit and total scale range represented by the maximum measurement unit pointer, the error water volume is calculated.

5. The smart water meter system based on AI visual recognition according to claim 4, characterized in that: The determination of the graded fault warning includes: When the mechanical error is within the set first-level error threshold range and is stable, and the error markers appear continuously and regularly, it is determined to be a first-level warning. If the mechanical error is within the set secondary error threshold range and shows a slow increasing trend during continuous monitoring, it is determined to be a secondary warning, and it is inferred that the gear set is obviously worn or the transmission shaft clearance is too large. If the mechanical error is within the set three-level error threshold range, and the frequency of error markers increases sharply in a short period of time, it is determined to be a three-level alarm.

6. The smart water meter system based on AI visual recognition according to claim 1, characterized in that: The image acquisition and preprocessing module performs the following steps: After waking up, the camera quickly perceives the environment, assesses the overall brightness, contrast, and presence of strong light spots, and dynamically adjusts the camera parameters based on the assessment results. In scenes with extremely low light or slight vibration, multiple images are captured rapidly and continuously, and multi-frame noise reduction and image fusion algorithms are used to synthesize images with clearer details and less noise. Targeted image enhancement includes de-reflection processing using deep learning-based restoration models or polarization analysis principles, enhancing local contrast using adaptive histogram equalization algorithms, and weakening interference from fixed stains through color and texture analysis.

7. The smart water meter system based on AI visual recognition according to claim 1, characterized in that: The AI ​​camera terminal further includes a first communication module, which is used to perform the following steps: The structured data packet generated by the identification and error verification module, which includes the final reading, error mark status, identification confidence and timestamp, is compressed and encapsulated together with key area image fragments and device status information. The compressed and packaged data is uploaded to the cloud-based identification and management platform via a wireless communication network. After the data upload is completed, the AI ​​camera terminal is controlled to enter a deep sleep state to reduce power consumption.

8. The smart water meter system based on AI visual recognition according to claim 1, characterized in that: The cloud-based identification and management platform also includes a regional water use analysis module, which includes the following operating process: Using independent metering areas as units, the water consumption of all water meters within an area during the same time period is aggregated, and a regional water load curve is plotted. The curve is then analyzed to identify normal water consumption patterns and detect abnormal water consumption patterns. When an abnormally high minimum flow rate is detected in a region, the leaking pipe section is initially located by combining the pipeline topology and water meter location through correlation analysis, and a leak detection work order is generated. The peak water usage forecast for the region is sent to the smart valve controller to perform time-of-use pressure regulation.

9. The smart water meter system based on AI visual recognition according to claim 1, characterized in that: The cloud-based identification and management platform also includes a data aggregation and storage module, used to perform the following steps: By receiving structured data packets uploaded from various AI camera terminals, the structured data packets include readings, timestamps, error markers, and device status; The received data is cleaned and verified, its format and integrity are checked, and invalid data caused by transmission errors is filtered out and converted into a unified spatiotemporal data model. The cleaned reading data, timestamps, error markers, and equipment status information are linked and integrated with water meter asset information and GIS geographic information; the water meter asset information includes water meter model, installation location, and user information, and the GIS geographic information includes the pipeline network zone and pressure zone to which it belongs; The integrated data is stored in a spatiotemporal database, and a time-series index is created for each water meter reading record to support time-based queries and trend analysis.

Images