Intelligent liquid flow meter reading and metering device based on visual identification

Abstract

The invention relates to an intelligent liquid flow meter reading and metering device based on visual identification. The intelligent liquid flow meter reading and metering device comprises a front-end acquisition module, a data processing module and a data transmission module, the front-end acquisition module is used for acquiring a dial plate image of the mechanical flowmeter; the data processing module is used for preprocessing the dial plate image, identifying the dial plate and calculating the numerical value; and the data transmission module is used for encrypting and transmitting data to the measurement monitoring system. Low-cost intelligent upgrading of a traditional mechanical flowmeter is achieved, the problem of manual meter reading is solved, meanwhile, low-cost intelligent flow monitoring refitting is achieved, the refitting cost is reduced, the recognition accuracy rate is high, the flowmeter can rapidly enter the stock market, and the flowmeter has the large-scale potential.

Description

Technical Field

[0001] This invention relates to the field of intelligent flow monitoring technology, and more specifically, to an intelligent liquid flow meter reading and measuring device based on visual recognition. Background Technology

[0002] Traditional mechanical flow meters rely on manual meter reading, which is inefficient, error-prone, and unable to monitor flow data in real time. In some special application scenarios, such as when flow meters operate in harsh environments with high temperature, high pressure, vibration, and corrosion, electronic flow meters are prone to damage and require frequent replacement, resulting in a significant waste of manpower, material resources, and financial resources. Mechanical flow meters are more reliable and durable than electronic flow meters and are therefore more widely used. For convenient monitoring and intelligent integration, it is necessary to maintain the functionality of mechanical flow meters while also possessing the functions of intelligent electronic flow meters. This can be achieved through a built-in clock-triggered wake-up function for meter reading, measurement, and data transmission, meeting the needs of terminal monitoring and reducing the frequency of manual maintenance.

[0003] The manual operation suffers from inefficiencies and reliability issues. Relying on manual on-site data recording of meters leads to poor timeliness, high labor costs, and the risk of data recording errors, making it difficult to meet the rigid requirements of real-time monitoring and data traceability in industrial scenarios. Electronic equipment lacks environmental adaptability. In harsh environments such as high temperature, high pressure, or corrosive conditions, electronic flow meters are prone to frequent failures, significantly shortening equipment replacement cycles and multiplying maintenance costs. While mechanical flow meters have advantages in environmental tolerance, they cannot directly meet the needs of digital management. There is a conflict between the cost and compatibility of intelligent transformation. Existing mechanical meter digitization solutions often require complete replacement with intelligent flow meters or the installation of invasive sensors, resulting in high transformation costs and damage to the original equipment structure, increasing the risk of seal failure. Image recognition stability is insufficient under complex conditions. Conventional visual recognition systems are prone to feature extraction errors under conditions of meter contamination, abnormal lighting (such as low light or strong backlight), or mechanical vibration interference, leading to a significant reduction in reading accuracy and failing to meet industrial metering accuracy standards. Finally, there is the challenge of balancing energy consumption and battery life in remote transmission. Traditional wireless communication modules consume too much power in continuous operation mode, which severely limits the battery life of battery-powered devices; if the data transmission frequency is reduced, the timeliness of traffic monitoring data will deteriorate. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a liquid flow intelligent reading and metering device based on visual recognition. Through visual recognition technology, it realizes a low-cost intelligent upgrade of traditional mechanical flow meters, solves the pain point of manual meter reading and reading, and realizes low-cost intelligent flow monitoring retrofit.

[0005] The technical solution adopted by the present invention to solve its technical problem is: to construct a liquid flow intelligent meter reading and measuring device based on visual recognition, including a front-end acquisition module, a data processing module and a data transmission module;

[0006] The front-end acquisition module is used to acquire the dial image of the mechanical flow meter;

[0007] The data processing module is used to preprocess the dial image, identify the dial, and perform numerical calculations.

[0008] The data transmission module is used to encrypt and transmit data to the measurement and monitoring system.

[0009] According to the above scheme, the front-end acquisition module includes a recognition camera and a flash;

[0010] The identification camera is a short focal length camera used to capture images of the dial of the mechanical flow meter;

[0011] The flashlight uses a low-brightness flashlight circuit to meet the brightness requirements for image recognition, and the lighting time is set according to the interval of the meter reading.

[0012] According to the above scheme, the data processing module uses data processing algorithms to perform dial image preprocessing, dial recognition, and numerical calculation, specifically including the following steps:

[0013] S1. The median filtering method is used to preprocess the flow meter image to complete the noise reduction process. Then, the standard deviation threshold tracking method is used to separate the background. The barcode image is divided into blocks, and the standard deviation of each block is used to determine whether it is a barcode area or a background area.

[0014] S2. A wavelet denoising method based on edge detection is adopted. Taking advantage of the good time-frequency characteristics of wavelet transform, the local maxima of the wavelet transform coefficient modulus are detected along the gradient direction to obtain the edge points of the image. At the same time, the wavelet coefficients corresponding to the noise points are removed to solve the image edge blurring problem caused by denoising. The features of numbers and pointers are extracted for subsequent recognition and reading analysis.

[0015] S3. The data is calibrated by fusing historical readings with real-time data and eliminating errors through deep learning algorithms to ensure an accuracy of ±0.5%.

[0016] According to the above scheme, the data processing module uses an image recognition algorithm to improve the robustness of dial image recognition of the mechanical flow meter, specifically including the following steps:

[0017] S1. The image is preprocessed by using a two-dimensional median filter with a dynamic window size to eliminate impulse noise, as shown in equation (1):

[0018] I'(x, y)=median{I(x+i, y+j)}, i, j∈[-2, 2]. (1)

[0019] In the formula, I′(x,y) is the pixel gray value of the filtered output image; I(x,y) is the gray value of the original image at pixel coordinates (x,y);

[0020] S2. Perform illumination compensation and background separation. Background separation is performed based on the local standard deviation threshold method, as shown in equation (2):

[0021]

[0022] In the formula, σ k is the sub-block standard deviation, which measures the dispersion of pixel values ​​within a sub-block; n is the total number of pixels in the sub-block; x i μ is the grayscale value of the i-th pixel; k The average gray level of the sub-block;

[0023] S3. For feature extraction and pointer recognition, the edges are extracted by wavelet transform, as shown in equation (3):

[0024]

[0025] In the formula, C edge (x,y) is the set of spatial coordinates of edge points in the image; W ψ f(x,y,θ) represents the directional wavelet transform coefficients; argmax θ To find the directional maximum, compare the gradient magnitudes in eight directions and select the θ corresponding to the maximum value as the principal direction of the edge.

[0026] Lines are detected based on Hough transform, and line segments with a length greater than 60% of the dial radius are selected as pointers;

[0027] S4. Perform data calibration algorithm and establish reading error compensation model, as shown in equation (4):

[0028] V true =V read +β·ΔV hist +γ·ΔT (4)

[0029] In the formula, V read ΔV is the current identification value. hist ΔT is the moving average of the changes in the previous three readings, ΔT is the temperature difference collected by the temperature sensor, β = 0.35, γ = 0.002.

[0030] According to the above scheme, the data transmission module adopts a timed wake-up reading method to reduce the power demand during the meter reading process. Specifically, the mechanical flow meter is normally in sleep mode. According to the externally set meter reading cycle, the control circuit board is triggered to wake up. After waking up, the flashlight is triggered in sequence, the recognition camera captures and caches the image, and then the flashlight and recognition camera are turned off. The edge computing chip of the control circuit board performs image recognition on the cached image and calculates the flow rate. The calculation result is communicated with the external system through the data transmission module. After successful communication, the control circuit board returns to sleep mode, all components are powered off, and the clock chip is retained for timing.

[0031] According to the above scheme, the timed wake-up table reading method includes the following steps:

[0032] S1, Sleep mode, RTC clock running, power consumption 2μA, timing period T = 1 hour;

[0033] S2 and RTC trigger interrupts to wake up the main controller and start the flashlight. The lighting time is Δt = 200ms.

[0034] S3, camera captures images at a resolution of 1280×720, with an exposure time of 10ms;

[0035] S4. Turn off the flash and cache the image to SRAM;

[0036] S5, the edge computing unit executes the recognition algorithm within 500ms;

[0037] The data transmission format of the S6 and 5G modules is shown in Table 1.

[0038] S7. The system enters deep sleep, and the power consumption drops to 3μA.

[0039] According to the above scheme, the data transmission module uses the 5G communication data transmission method to transmit data, and the 5G communication data transmission method uses the standard short message data format for transmission.

[0040] According to the above scheme, the short message content includes: the meter reading device number, the meter reading time, the meter reading time and the dial value, the fluid consumption during the interval, and the check code.

[0041] According to the above scheme, the 5G communication data transmission method includes the following steps:

[0042] S1. Initialize communication. After the device wakes up, the 5G module first sends a connection request containing the device ID and authentication key to the remote monitoring platform. If no response is received from the platform within 10 seconds, the first retry is triggered. Each retry interval increases exponentially by $t = 2^{n-1}$ seconds, where $n$ is the current retry sequence number.

[0043] S2. After the connection is successfully established, encapsulate the data packet (see Table 1 for the format) and send it via the CoAP protocol. Start the transmission timeout timer. If no ACK confirmation is received within the timeout, start the secondary retry mechanism and retransmit the packet at a fixed interval of 1 second, up to a maximum of 10 times. If the cumulative retries exceed 10, it is determined to be a communication failure. After recording the error log, it enters sleep mode.

[0044] S3. Perform abnormal handling. If the cumulative failures exceed 10 times in any stage, immediately terminate communication and execute the operation. The data to be transmitted is temporarily stored in FRAM non-volatile memory. In the next wake-up cycle, historical data is preferentially retransmitted, and a fault code is triggered to send an alarm to the operation and maintenance terminal via SMS.

[0045] Table 1 5G Communication Data Message Content

[0046]

[0047] According to the above scheme, it also includes an installation housing and an isolation protective glass. The installation housing is fixedly installed on the mechanical flow meter, and the isolation protective glass is installed between the installation housing and the mechanical flow meter. The installation housing is equipped with a front-end acquisition module, a data processing module and a data transmission module.

[0048] According to the above scheme, the identification camera is a short focal length camera used to acquire the dial image of the mechanical flow meter; the flash uses a low-brightness flash circuit to meet the brightness requirements for image recognition, and the lighting time is set according to the meter reading interval; the control circuit board is powered by a built-in battery or an external power supply and has a built-in clock chip; the control circuit board is equipped with an edge computing chip for image preprocessing, dial recognition, and numerical calculation; the data transmission module is used to encrypt and transmit data to the measurement and monitoring system; and the protective glass is used to isolate the electrical area from the external humid area.

[0049] The intelligent liquid flow metering device based on visual recognition of the present invention has the following beneficial effects:

[0050] 1. This invention automatically acquires dial images and analyzes readings through visual recognition technology. It can be combined with 5G communication to achieve real-time remote data transmission, completely eliminating the need for manual intervention and ensuring the timeliness and accuracy of data acquisition. It uses a non-invasive external device to modify the mechanical flow meter, retaining its environmental tolerance advantage while realizing digital readings through an external visual recognition module, avoiding damage to the original equipment structure.

[0051] 2. This invention features a standardized adapter interface and modular components, enabling intelligent upgrades without disassembling the original equipment or compromising its sealing. This significantly reduces modification costs and construction complexity. It develops an adaptive image processing algorithm and integrates multi-scene noise reduction, dynamic lighting compensation, and vibration blur correction technologies to ensure stable recognition accuracy under harsh conditions. It employs a timed wake-up mechanism and a deep sleep mode for coordinated control, activating the camera and communication module only during the moment of reading the meter, while maintaining ultra-low power consumption at other times, achieving compatibility between high-frequency data transmission and long battery life.

[0052] 3. This invention utilizes visual recognition technology and 5G communication technology to achieve a low-cost intelligent upgrade of traditional mechanical flow meters, solving the pain points of manual meter reading and reading, while simultaneously enabling low-cost intelligent flow monitoring retrofitting. Retrofitting costs are reduced by 80%, maintenance by 70%, and the recognition accuracy is 99.2%. It can quickly penetrate the existing market and has scalability potential. For intelligent applications in special scenarios, it uses reliable and durable mechanical flow meters, avoiding the high costs of frequent replacements of damaged electronic flow meters. The non-invasive retrofitting eliminates the need to replace traditional mechanical flow meters; intelligent transformation is achieved through external devices, reducing retrofitting costs by over 80%. High-precision recognition is achieved; the deep learning model achieves a recognition accuracy of 99.2% in complex environments. The low-power design utilizes a low-power 5G communication module and a low-power flashlight, triggered by a timer clock, resulting in industry-leading device battery life. Attached Figure Description

[0053] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings:

[0054] Figure 1 This is a schematic diagram of the intelligent liquid flow meter reading device based on visual recognition according to the present invention.

[0055] Figure 2 This is a schematic diagram of the 5G transmission process of the present invention;

[0056] In the diagram: 1. Protective glass, 2. Identification camera, 3. Flash, 4. Control circuit board, 5. Data transmission module. Detailed Implementation

[0057] To provide a clearer understanding of the technical features, objectives, and effects of the present invention, specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0058] like Figure 1As shown, the intelligent liquid flow metering device based on vision recognition of the present invention includes a front-end acquisition module, a data processing module, and a data transmission module. The front-end acquisition module is used to acquire the dial image of the mechanical flow meter; the data processing module is used to preprocess the dial image, identify the dial, and perform numerical calculations; the data transmission module is used to encrypt and transmit the data to the measurement and monitoring system.

[0059] The front-end acquisition module includes a recognition camera 2 and a flash 3. The recognition camera 2 is equipped with autofocus and infrared illumination, supporting full-scene dial image acquisition, including low-light, backlight, and dial stain / damage environments. The recognition camera 2 uses a short focal length for acquiring dial images from traditional mechanical flow meters; the flash 3 uses a low-brightness flash circuit to meet the brightness requirements for image recognition. The control circuit board 4 is equipped with an image processing chip to perform dial image preprocessing, pointer recognition, and numerical calculation functions.

[0060] The data processing module uses data processing algorithms to preprocess the dial image, identify the dial, and perform numerical calculations, specifically including the following steps:

[0061] S1. The median filtering method is used to preprocess the flow meter image to complete the noise reduction process. Then, the standard deviation threshold tracking method is used to separate the background. The barcode image is divided into blocks, and the standard deviation of each block is used to determine whether it is a barcode area or a background area.

[0062] S2. A wavelet denoising method based on edge detection is adopted. Taking advantage of the good time-frequency characteristics of wavelet transform, the local maxima of the wavelet transform coefficient modulus are detected along the gradient direction to obtain the edge points of the image. At the same time, the wavelet coefficients corresponding to the noise points are removed to solve the image edge blurring problem caused by denoising. The features of numbers and pointers are extracted for subsequent recognition and reading analysis.

[0063] S3. The data is calibrated by fusing historical readings with real-time data and eliminating errors through deep learning algorithms to ensure an accuracy of ±0.5%.

[0064] The data processing module employs image recognition algorithms to improve the robustness of dial image recognition in mechanical flow meters, specifically including the following steps:

[0065] S1. The image is preprocessed by using a two-dimensional median filter with a dynamic window size to eliminate impulse noise, as shown in equation (1):

[0066] I'(x, y)=median{I(x+i, y+j)}, i, j∈[-2, 2]. (1)

[0067] In the formula, I′(x,y) is the pixel gray value of the filtered output image; I(x,y) is the gray value of the original image at pixel coordinates (x,y);

[0068] S2. Perform illumination compensation and background separation. Background separation is performed based on the local standard deviation threshold method, as shown in equation (2):

[0069]

[0070] In the formula, σ k is the sub-block standard deviation, which measures the dispersion of pixel values ​​within a sub-block; n is the total number of pixels in the sub-block; x i μ is the grayscale value of the i-th pixel; k The average gray level of the sub-block;

[0071] S3. For feature extraction and pointer recognition, the edges are extracted by wavelet transform, as shown in equation (3):

[0072]

[0073] In the formula, C edge (x, y) is the set of spatial coordinates of edge points in the image; W ψ f(x,y,θ) represents the directional wavelet transform coefficients; argmax θ To find the directional maximum, compare the gradient magnitudes in eight directions and select the θ corresponding to the maximum value as the principal direction of the edge.

[0074] Lines are detected based on Hough transform, and line segments with a length greater than 60% of the dial radius are selected as pointers;

[0075] S4. Perform data calibration algorithm and establish reading error compensation model, as shown in equation (4):

[0076] V true =V read +β·ΔV hist +γ·ΔT (4)

[0077] In the formula, V read ΔV is the current identification value. hist ΔT is the moving average of the changes in the previous three readings, ΔT is the temperature difference collected by the temperature sensor, β = 0.35, γ = 0.002.

[0078] The historical reading sequence [100.1, 100.3, 100.5]m 3 For example, suppose the current identification value V read =100.7, temperature sensor collects temperature difference ΔT = -5℃. Calculate.

[0079] ΔV hist =[(100.3-100.1)+(100.5-100.3)] / 2=0.2,

[0080] V true=100.7+0.35×0.2+0.002×(-5)=100.77m 3 .

[0081] The data transmission module employs a low-power 5G communication module, supporting timed wake-up for data transmission and ensuring extended battery life. Whether powered by battery or external power, it provides real-time data transmission and monitoring. The data transmission module utilizes Dynamic Voltage Scaling (DVFS) technology combined with a deep sleep mode, supporting up to 12 months of device operation in battery mode. Data is encrypted before being transmitted to the cloud via the 5G network, and blockchain technology ensures data immutability.

[0082] The data transmission module 5 uses a timed wake-up reading method to reduce the power consumption during the meter reading process. Specifically, the mechanical flow meter is normally in sleep mode. According to the externally set meter reading cycle, the control circuit board 4 is triggered to wake up. After waking up, the flashlight 3 is triggered to light up in sequence, the recognition camera 2 takes pictures and caches the images, and then the flashlight 3 and the recognition camera 2 are turned off. The edge computing chip of the control circuit board 4 performs image recognition on the cached images and calculates the flow rate. The calculation results are communicated with the external system through the data transmission module 5. After successful communication, the control circuit board 4 returns to sleep mode, all components stop receiving power, and the clock chip is retained for timing.

[0083] The timed wake-up meter reading method works as follows: Normally, the intelligent meter reading device of the mechanical flow meter is in sleep mode. Based on the externally set meter reading cycle, the control circuit board 4 is triggered to wake it up. After waking up, the flashlight 3 is activated sequentially, and the recognition camera 2 captures and caches an image. The flashlight 3 and recognition camera 2 are then turned off. The edge computing chip on the control circuit board 4 performs image recognition on the cached image and calculates the flow rate. The calculation result is communicated to the external system via the data transmission module 5. After successful communication, the control circuit board 4 returns to sleep mode, all components are powered off, and only the clock chip remains powered on for timing. The entire process is completed within 1-3 seconds, significantly reducing the power consumption of the meter reading process. The timed wake-up meter reading method includes the following steps:

[0084] S1, Sleep mode, RTC clock running, power consumption 2μA, timing period T = 1 hour;

[0085] S2 and RTC trigger interrupt to wake up the main controller and start flashlight 3, with a lighting time Δt = 200ms;

[0086] S3, camera captures images at a resolution of 1280×720, with an exposure time of 10ms;

[0087] S4. Turn off flash 3, and cache image to SRAM;

[0088] S5, the edge computing unit executes the recognition algorithm within 500ms;

[0089] The data transmission format of the S6 and 5G modules is shown in Table 1.

[0090] S7. The system enters deep sleep, and the power consumption drops to 3μA.

[0091] The data transmission module 5 uses 5G communication data transmission method for data transmission. The 5G communication data transmission method uses standard short message data format for transmission. The message content includes: meter reading device number, meter reading time, meter reading time dial value, fluid consumption during the interval, and check code.

[0092] This invention employs a tiered retry mechanism to achieve reliable transmission in industrial environments, such as... Figure 2 As shown, the 5G communication data transmission method includes the following steps:

[0093] S1. Communication initialization: After the device wakes up, the 5G module first sends a connection request containing the device ID and authentication key to the remote monitoring platform. If no response is received from the platform within 10 seconds, the first retry is triggered. The maximum number of retries is 10, and the interval between each retry increases exponentially by $t = 2^{n-1}$ seconds, where $n$ is the current retry sequence number.

[0094] S2. After the connection is successfully established, encapsulate the data packet (see Table 1 for the format) and send it via the CoAP protocol. Start the transmission timeout timer. If no ACK confirmation is received within the timeout, start the secondary retry mechanism and retransmit the packet at a fixed interval of 1 second, up to a maximum of 10 times. If the cumulative retries exceed 10, it is determined to be a communication failure. After recording the error log, it enters sleep mode.

[0095] S3. Perform abnormal handling. If the cumulative failures exceed 10 times at any stage, immediately terminate communication and execute the operation. The data to be transmitted is temporarily stored in FRAM non-volatile memory. In the next wake-up cycle, historical data is preferentially transmitted. Trigger fault code (0xE001) and send an alarm to the operation and maintenance terminal via SMS.

[0096] Table 1 5G Communication Data Message Content

[0097]

[0098] The present invention also provides a liquid flow intelligent meter reading device based on visual recognition, including a mounting shell and an isolation protective glass 1. The mounting shell is fixedly mounted on the mechanical flow meter, and the isolation protective glass 1 is disposed between the mounting shell and the mechanical flow meter. Inside the mounting shell, a recognition camera 2, a flash 3, a control circuit board 4 and a data transmission module 5 are arranged through a mounting interface.

[0099] The identification camera 2 uses a short focal length camera to capture images of the mechanical flow meter's dial; the flash 3 uses a low-brightness flash circuit to meet the brightness requirements for image recognition, and the lighting time is set according to the interval of meter reading; the control circuit board 4 is powered by a built-in battery or an external power supply and has a built-in clock chip; the control circuit board 4 is equipped with an edge computing chip for image preprocessing, dial recognition, and numerical calculation; the 5G communication module is used for encrypted data transmission to the measurement and monitoring system; the protective glass is used to isolate the electrical area from the external humid area.

[0100] Working principle of the invention:

[0101] This invention is applicable to the non-intrusive intelligent transformation of traditional mechanical flow meters. The intelligent meter reading device is installed using a non-intrusive design and an interface compatible with traditional mechanical flow meters. The intelligent meter reading device is powered by a built-in battery or an external power supply. It periodically wakes up according to the meter reading cycle, illuminating a light at the time of reading and capturing images with a camera for digital recognition. The recognized measurement data is transmitted via 5G communication signals in a standard short message data format. After transmission, the intelligent meter reading device enters a low-power sleep mode. This invention solves the problems of low efficiency and error-proneness associated with manual meter reading of traditional mechanical flow meters, and allows for the transformation of mechanical flow meters into intelligent meter reading devices based on visual recognition technology without replacing the original flow meter. In some special application scenarios, such as when flow meters operate in harsh environments with high temperature, high pressure, vibration, and corrosion, electronic flow meters are prone to damage and require frequent replacement, resulting in a significant waste of manpower, material resources, and financial resources. Mechanical flow meters are more reliable and durable than electronic flow meters and are therefore more widely used. However, for convenient monitoring and intelligent integration, it is necessary to maintain the use of mechanical flow meters while also having the functions of intelligent electronic flow meters. This can be achieved through built-in clock-triggered wake-up for meter reading, measurement, and data transmission, meeting the terminal monitoring needs and reducing the frequency of manual maintenance.

[0102] This invention relates to a non-invasive mechanical flow meter external attachment device and a low-power collaborative control system. It achieves seamless compatibility with existing mechanical flow meters through a customized installation interface, and uses protective isolation glass 1 to ensure the electrical components are isolated from harsh environments. A built-in clock chip triggers periodic wake-up, completing dial illumination, image acquisition, edge computing, and 5G data transmission within a millisecond-level operating window before immediately entering deep sleep, forming a complete low-power operating closed loop. An industrial-grade 5G short message encrypted transmission system is also included. The design incorporates a standard short message structure including device code, timestamp, metering data, and checksum, employing an end-to-end encrypted transmission guarantee mechanism to ensure data integrity and operational traceability while meeting the reliability requirements of data transmission in harsh industrial environments.

[0103] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A smart metering device for liquid flow based on visual recognition, characterized in that, It includes a front-end acquisition module, a data processing module, and a data transmission module; The front-end acquisition module is used to acquire the dial image of the mechanical flow meter; The data processing module is used to preprocess the dial image, identify the dial, and perform numerical calculations. The data transmission module is used to encrypt and transmit data to the measurement and monitoring system.

2. The intelligent liquid flow metering device based on visual recognition according to claim 1, characterized in that, The front-end acquisition module includes a recognition camera and a flash; The identification camera is a short focal length camera used to capture images of the dial of the mechanical flow meter; The flashlight uses a low-brightness flashlight circuit to meet the brightness requirements for image recognition, and the lighting time is set according to the interval of the meter reading.

3. The intelligent liquid flow metering device based on visual recognition according to claim 1, characterized in that, The data processing module employs data processing algorithms to preprocess the dial image, identify the dial, and perform numerical calculations, specifically including the following steps: S1. The median filtering method is used to preprocess the flow meter image to complete the noise reduction process. Then, the standard deviation threshold tracking method is used to separate the background. The barcode image is divided into blocks, and the standard deviation of each block is used to determine whether it is a barcode area or a background area. S2. A wavelet denoising method based on edge detection is adopted. Taking advantage of the good time-frequency characteristics of wavelet transform, the local maxima of the wavelet transform coefficient modulus are detected along the gradient direction to obtain the edge points of the image. At the same time, the wavelet coefficients corresponding to the noise points are removed to solve the image edge blurring problem caused by denoising. The features of numbers and pointers are extracted for subsequent recognition and reading analysis. S3. The data is calibrated by fusing historical readings with real-time data and eliminating errors through deep learning algorithms.

4. The intelligent liquid flow metering device based on visual recognition according to claim 3, characterized in that, The data processing module employs image recognition algorithms to improve the robustness of dial image recognition in mechanical flow meters, specifically including the following steps: S1. The image is preprocessed by using a two-dimensional median filter with a dynamic window size to eliminate impulse noise, as shown in equation (1): I′(x,y)=median{I(x+i,y+j)},i,j∈[-2,2]. (1) In the formula, I′(x,y) is the pixel gray value of the filtered output image; I(x,y) is the gray value of the original image at pixel coordinates (x,y); S2. Perform illumination compensation and background separation. Background separation is performed based on the local standard deviation threshold method, as shown in equation (2): In the formula, σ k is the sub-block standard deviation, which measures the dispersion of pixel values ​​within a sub-block; n is the total number of pixels in the sub-block; x i μ is the grayscale value of the i-th pixel; k The average gray level of the sub-block; S3. For feature extraction and pointer recognition, the edges are extracted by wavelet transform, as shown in equation (3): In the formula, C edge (x, y) is the set of spatial coordinates of edge points in the image; W ψ f(x, y, θ) are the directional wavelet transform coefficients; argmax θ To find the directional maximum, compare the gradient magnitudes in eight directions and select the θ corresponding to the maximum value as the principal direction of the edge. Lines are detected based on Hough transform, and line segments with a length greater than 60% of the dial radius are selected as pointers; S4. Perform data calibration algorithm and establish reading error compensation model, as shown in equation (4): V true =V read +β·ΔV hist +γ·ΔT (4) In the formula, V read ΔV is the current identification value. hist ΔT is the moving average of the changes in the previous three readings, and ΔT is the temperature difference collected by the temperature sensor.

5. The intelligent liquid flow metering device based on visual recognition according to claim 1, characterized in that, The data transmission module employs a timed wake-up reading method to reduce power consumption during meter reading. Specifically, the mechanical flow meter is normally in sleep mode. Based on the externally set meter reading cycle, the control circuit board is triggered to wake it up. After waking up, the flashlight is triggered sequentially, the recognition camera captures and caches images, and then the flashlight and recognition camera are turned off. The edge computing chip on the control circuit board performs image recognition on the cached images and calculates the flow rate. The calculation results are communicated with the external system through the data transmission module. After successful communication, the control circuit board returns to sleep mode, all components are powered off, and the clock chip is retained for timing.

6. The intelligent liquid flow metering device based on visual recognition according to claim 5, characterized in that, The timed wake-up table reading method includes the following steps: S1, Sleep mode, RTC clock is running; S2 and RTC trigger interrupts to wake up the main controller and start the flashlight; S3, The camera captures images; S4. Turn off the flash and cache the image to SRAM; S5, the edge computing unit executes the recognition algorithm; S6. The data transmission module sends data; S7, the system enters deep sleep.

7. The intelligent liquid flow metering device based on visual recognition according to claim 1, characterized in that, The data transmission module uses a 5G communication data transmission method for data transmission, and the 5G communication data transmission method uses a standard short message data format for transmission.

8. The intelligent liquid flow metering device based on visual recognition according to claim 7, characterized in that, The short message content includes: the meter reading device number, the meter reading time, the meter reading value at the time of reading, the fluid consumption during the interval, and the check code.

9. The intelligent liquid flow metering device based on visual recognition according to claim 8, characterized in that, The 5G communication data transmission method includes the following steps: S1. Initialize communication. After the device wakes up, the 5G module first sends a connection request containing the device ID and authentication key to the remote monitoring platform. If no response is received from the platform within 10 seconds, the first retry is triggered. Each retry interval increases exponentially by $t = 2^{n-1}$ seconds, where $n$ is the current retry sequence number. S2. After the connection is successfully established, encapsulate the data packet (see Table 1 for the format) and send it via the CoAP protocol. Start the transmission timeout timer. If no ACK confirmation is received within the timeout, start the secondary retry mechanism and retransmit the packet at a fixed interval of 1 second, up to a maximum of 10 times. If the cumulative retries exceed 10, it is determined to be a communication failure. After recording the error log, it enters sleep mode. S3. Perform abnormal handling. If the cumulative failures exceed 10 times at any stage, immediately terminate communication and execute the operation. The data to be transmitted is temporarily stored in FRAM non-volatile memory. In the next wake-up cycle, historical data is preferentially retransmitted, and a fault code is triggered to send an alarm to the operation and maintenance terminal via SMS.

10. The intelligent liquid flow metering device based on visual recognition according to claim 1, characterized in that, It also includes a mounting housing and an isolation protective glass. The mounting housing is fixedly mounted on the mechanical flow meter, and the isolation protective glass is disposed between the mounting housing and the mechanical flow meter. The mounting housing is equipped with a front-end acquisition module, a data processing module, and a data transmission module.

Images