CN116156359A - Intelligent edge machine vision meter reading module of Internet of things and implementation method thereof - Google Patents

Abstract

The invention discloses an intelligent edge machine vision meter reading module of the Internet of things and an implementation method thereof, which relate to the technical field of intelligent meter reading and comprise a hardware management unit, a software management unit and a structure management unit, wherein the hardware management unit, the software management unit and the structure management unit all keep data real-time intercommunication and sharing through the Internet of things.

Description

Intelligent edge machine vision meter reading module of Internet of things and implementation method thereof

Technical Field

The invention relates to the technical field of intelligent meter reading, in particular to an intelligent edge machine vision meter reading module of the Internet of things and an implementation method thereof.

Background

At present, a large number of various digital and analog meters exist in metering of water, gas, oil and the like in families and industries, most of the meters are not provided with electronic interfaces, and particularly, the meters are small in installation space and special in position, so that great difficulty, time and labor are wasted on meter reading work. The traditional manual meter reading mainly has the following problems:

(1) The working efficiency is low: the problems of difficult household access, high strength, long period, low efficiency of manual settlement mode, easy error occurrence and the like are faced.

(2) The operation cost is high: the traditional manual meter reading mode faces the problem that the labor cost increases year by year, and the operation cost cannot be effectively controlled.

The smart meter mainly has the following problems:

(1) There is a production and marketing difference: firstly, because the intelligent meter basically adopts an electronic sensor, the intelligent meter generally has drift to cause metering errors after long-time use; secondly, production and marketing errors are caused in aspects such as errors in upstream and downstream metering, precision of a gauge tool, influence of supply environment (temperature, pressure and the like) and the like;

(2) The cost is high: the intelligent meter is replaced, firstly, the pipeline is required to be closed and implemented by professionals, and the labor cost is high; secondly, most intelligent meters are deployed at the tail end of the supply side, and the problem of equipment power supply is faced, if wired power supply and network supply are used, the construction cost is high;

(3) Network coverage is incomplete: smartmeters are often installed in closed environments, such as stairways, indoors, underground, etc., not only are the installation environments complex, but also network signals are often difficult to secure;

therefore, the invention needs to design the intelligent edge machine vision meter reading module of the internet of things and the realization method thereof to solve the problems.

The invention comprises the following steps:

the invention aims to solve the problems and provide an intelligent edge machine vision meter reading module of the internet of things and an implementation method thereof, which solve the problems of production and marketing differences in the background art: firstly, because the intelligent meter basically adopts an electronic sensor, the intelligent meter generally has drift to cause metering errors after long-time use; secondly, production and marketing errors are caused in aspects of errors existing in upstream and downstream metering, precision of a meter, influence of supply environment and the like; the cost is high: the intelligent meter is replaced, firstly, the pipeline is required to be closed and implemented by professionals, and the labor cost is high; secondly, most intelligent meters are deployed at the tail end of the supply side, and the problem of equipment power supply is faced, if wired power supply and network supply are used, the construction cost is high; network coverage is incomplete: smart meters are often installed in closed environments, such as corridors, indoors, underground, etc., not only are the installation environments complex, but also network signals are often difficult to secure.

In order to solve the problems, the invention provides a technical scheme that:

the intelligent edge machine vision meter reading module of the Internet of things and the implementation method thereof comprise a hardware management unit, a software management unit and a structure management unit, wherein the hardware management unit, the software management unit and the structure management unit all keep data real-time intercommunication and sharing through the Internet of things;

the hardware management unit is used for collecting and storing normal images, and supplementing light through the LEDs in the image collecting process;

the software management unit is used for preprocessing the acquired images in an alignment, fine adjustment and cutting mode, classifying the images through a deep convolutional neural network CNN, compressing a trained digital dial reading identification model through a TensorFlow Lite, compiling in combination with a main control program firmware, and deploying the whole to edge equipment for meter reading;

the structure management unit is used for customizing and processing the adapting clamp and the module shell according to the appearance structure of the meter to be identified.

Preferably, the output end of the hardware management unit is in communication connection with the input end of the software management unit, and the hardware management unit and the software management unit are both in bidirectional communication connection with the structure management unit.

Preferably, the hardware management unit comprises a control and communication unit, a first image acquisition unit and a data storage unit, wherein the first image acquisition unit and the data storage unit are both in bidirectional communication connection with the control and communication unit, and the output end of the first image acquisition unit is in communication connection with the input end of the data storage unit.

Preferably, the control and communication unit is used for managing the first image acquisition unit and the data storage unit to work normally, and the control and communication unit is also used for adopting an internet of things solution;

the first image acquisition unit is used for supplementing light through the LEDs in the image acquisition process;

the data storage unit is used for receiving the acquired data sent by the first image acquisition unit, and is also used for storing data and service programs by adopting an SD card.

Preferably, the software management unit comprises a second image acquisition unit, an image preprocessing unit, a digital dial reading identification unit, an analog dial reading identification unit, a main control and communication unit, wherein the output end of the second image acquisition unit is in communication connection with the input end of the image preprocessing unit, and the second image acquisition unit, the image preprocessing unit, the digital dial reading identification unit and the analog dial reading identification unit are in bidirectional communication connection with the main control and communication unit.

Preferably, the second image acquisition unit is used for receiving the acquired data sent by the hardware management unit, and the second image acquisition unit is also used for sending the acquired data to the inside of the image preprocessing unit for preprocessing;

the image preprocessing unit is used for receiving the acquired data sent by the second image acquisition unit, and the preprocessing mode of the image preprocessing unit comprises alignment, fine adjustment and cutting;

alignment: aligning the input dial image, identifying a subsequent model, and rotating the input image by a fixed angle by the image preprocessing unit in the aligning process;

fine tuning: fine tuning is to place the pre-aligned image in a relatively well-defined position to align the image with a fixed ROI for final segmentation;

cutting: the trimmed image is in the framed position, and the ROI is segmented and stored as a single picture;

the digital dial reading identification unit is used for classifying images through a deep convolutional neural network CNN, compressing a trained digital dial reading identification model through a TensorFlow Lite, compiling by combining with a main control program firmware, and finally integrally deploying to edge equipment;

the analog dial reading identification unit is used for identifying an analog dial target area, reading analog pointer display and converting the analog pointer display into a digital value, adopting a neuron training target value, compressing a trained analog dial reading identification model through a TensorFlow Lite, compiling by combining with a main control program firmware, and finally arranging the analog dial reading identification model to an edge;

the main control and communication unit is used for four processes including reading synthesis, communication program, file management and configuration management;

reading synthesis: synthesizing a final meter reading according to the output result of the digital dial reading identification model, the output result of the analog dial reading identification model and other information recognized by OCR;

communication program: the timing period release of meter reading and original photographed images is completed through the MQTT protocol of the Internet of things, and meanwhile, inquiry and control commands of a server side can be received;

file management: the system has the functions of a file server, supports the function of network disconnection data storage and supports the functions of OTA upgrading and REST API;

configuration management: the working parameters and the network communication parameters of the module are configured directly through the built-in Web interface.

Preferably, the structure management unit comprises a fixture customizing unit and a module shell customizing unit, and the fixture customizing unit and the module shell customizing unit are in bidirectional communication connection with the structure management unit.

Preferably, the fixture customizing unit is used for customizing the processing adapting fixture according to the appearance structure of the meter to be identified;

the module shell customizing unit is used for customizing and processing the adaptation module shell according to the appearance structure of the meter to be identified.

The realization method of the intelligent edge machine vision meter reading module of the internet of things comprises the following specific steps:

s1, firstly, generating the edge machine vision meter reading module, and acquiring a large amount of dial image data by using edge hardware equipment;

s2, modeling dial image data based on a machine vision model, and accurately classifying images by using a convolutional neural network CNN to obtain data in the dial;

s3, compressing the training model by using a TensorFlow Lite, and finally integrating the training model into the firmware of the edge equipment, so that the rapid intelligent meter reading function of the edge side is realized.

Preferably, the structure of the convolutional neural network CNN in the step S2 includes a plurality of Conv2D layers, maxPooling and flame layers.

The beneficial effects of the invention are as follows: the invention successfully deploys the machine vision meter reading module based on deep learning to the edge end for the first time in the automatic meter reading field, and compared with the current meter reading scheme, the scheme has the following advantages: firstly, the installation and the deployment are convenient, disassembly, assembly and transformation are not needed, the intelligent meter reading function of the traditional meter can be realized rapidly, and time and labor are saved; secondly, the measurement accuracy and reliability of the original mechanical meter are maintained, and meanwhile meter reading errors can be corrected through a machine vision AI model, so that the measurement accuracy is ensured; thirdly, compared with the replacement of the intelligent meter, the pipeline does not need to be closed, and normal use of a user is not affected; secondly, the equipment and the installation labor cost are low; fourthly, the network communication system has the functions of the Internet of things such as WIFI, NB-IoT and the like, and can meet the requirements of network communication in various application scenes; fifthly, the intelligent meter reading system has the functions of parameter configuration, OTA firmware upgrading function, third party integration API and the like, and through designing and developing the hardware of the peripheral machine vision meter reading module of the Internet of things and combining a meter reading model, the peripheral side deployment is completed, the intelligent meter reading function of the traditional analog meter and the traditional digital meter is realized, the problems that time and labor are wasted, the cost is too high and the metering precision is difficult to guarantee in the traditional manual meter reading and intelligent meter changing method are solved, and meanwhile, the intelligent level of the whole meter reading management is improved.

Description of the drawings:

for ease of illustration, the invention is described in detail by the following detailed description and the accompanying drawings.

FIG. 1 is an overall topology diagram of an intelligent edge machine vision meter reading module of the Internet of things and an implementation method thereof;

FIG. 2 is a flowchart of an intelligent edge machine vision meter reading module of the Internet of things and an implementation method thereof;

FIG. 3 is a topology diagram of a hardware management unit of the intelligent peripheral machine vision meter reading module of the Internet of things and the implementation method thereof;

FIG. 4 is a topology diagram of a software management unit of an intelligent edge machine vision meter reading module of the Internet of things and an implementation method thereof;

fig. 5 is a topology diagram of a structural management unit of the intelligent edge machine vision meter reading module of the internet of things and the implementation method thereof.

The specific embodiment is as follows:

as shown in fig. 1, 2, 3, 4 and 5, the present embodiment adopts the following technical solutions:

the intelligent edge machine vision meter reading module of the Internet of things and the implementation method thereof comprise a hardware management unit, a software management unit and a structure management unit, wherein the hardware management unit, the software management unit and the structure management unit all keep data real-time intercommunication and sharing through the Internet of things; the system comprises a hardware management unit, a software management unit and a structure management unit, wherein the output end of the hardware management unit is in communication connection with the input end of the software management unit, and the hardware management unit and the software management unit are both in bidirectional communication connection with the structure management unit;

the hardware management unit is used for collecting and storing normal images, and supplementing light through the LEDs in the image collecting process;

the software management unit is used for preprocessing the acquired images in an alignment, fine adjustment and cutting mode, classifying the images through a deep convolutional neural network CNN, compressing a trained digital dial reading identification model through a TensorFlow Lite, compiling in combination with a main control program firmware, and deploying the whole to edge equipment for meter reading;

the structure management unit is used for customizing and processing the adapting clamp and the module shell according to the appearance structure of the meter to be identified.

Further, the hardware management unit comprises a control and communication unit, a first image acquisition unit and a data storage unit, wherein the first image acquisition unit and the data storage unit are both in bidirectional communication connection with the control and communication unit, and the output end of the first image acquisition unit is in communication connection with the input end of the data storage unit.

Furthermore, the control and communication unit is used for managing the first image acquisition unit and the data storage unit to work normally, the control and communication unit is also used for adopting an Internet of things solution, and the control and communication unit supports WIFI and Bluetooth communication modes simultaneously, supports InfluxDb, MQTT and REST APIs and facilitates third party integration application;

the first image acquisition unit is used for avoiding interference of image acquisition noise through LED light supplementing in the image acquisition process, and an ESP32-CAM or Arducam smart camera is adopted to realize a photographing function;

the data storage unit is used for receiving the acquired data sent by the first image acquisition unit, and is also used for storing data and service programs by adopting an SD card.

Further, the software management unit comprises a second image acquisition unit, an image preprocessing unit, a digital dial reading identification unit, an analog dial reading identification unit, a main control and communication unit, wherein the output end of the second image acquisition unit is in communication connection with the input end of the image preprocessing unit, and the second image acquisition unit, the image preprocessing unit, the digital dial reading identification unit and the analog dial reading identification unit are in bidirectional communication connection with the main control and communication unit.

Further, the second image acquisition unit is used for receiving the acquired data sent by the hardware management unit, and the second image acquisition unit is also used for sending the acquired data to the inside of the image preprocessing unit for preprocessing;

the image preprocessing unit is used for receiving the acquired data sent by the second image acquisition unit, and the preprocessing mode of the image preprocessing unit comprises alignment, fine adjustment and cutting; alignment: aligning an input dial image, identifying a subsequent model, reducing interference of noise data, and rotating the input image by a fixed angle by the image preprocessing unit in the aligning process, for example, correcting an image with a downward head or an image rotated by 90 degrees, wherein the aligning angle supports configuration according to a scene; fine tuning: fine tuning is to place the pre-aligned image in a relatively well-defined position to align the image with a fixed ROI for final segmentation, which is based on OpenCV image processing, identify the positions of multiple reference structures in the image, and transform the image to the final position by affine transformation; cutting: the trimmed image is in the already framed position, the ROI can be segmented and stored as separate pictures, any number of ROIs can be defined and stored in each category for different dials, and for each ROI, the name, position (x, y) and size (dx, dy) need to be specified;

the digital dial reading identification unit is used for classifying images through a deep convolutional neural network CNN, compressing a trained digital dial reading identification model through a TensorFlow Lite, compiling by combining a main control program firmware, and finally integrally deploying to edge equipment, wherein the digital dial reading identification is a classical classification network, 11 classes represent numbers 0, 1, 9 and special classes 'NaN', the digital dial reading identification unit is very suitable for various rolling digital display dials, a non-digital special state exists between two images at times, the NaN represents a non-digital condition, namely the images cannot be uniquely classified into one number, for example, because the NaN is positioned between the two numbers, for the type, the required training data amount of each type is minimum, a large number of types are already part of a training set, in order to ensure the generalization capability of the identification model, the training model uses scattering input images, the input images are scattered through brightness, pixel shift, rotation and rotation, the input images are scattered, the stability of the network is greatly improved, and the specific implementation measures comprise the following aspects: 1) Luminance dispersion +/-30%, 2) pixel positions are interspersed by +/-1 pixel in each direction. Since the original picture is almost 3 times larger (55 x90 pixels), ensuring sufficient uncertainty in the upstream image alignment and cropping process, 3) rotation spread +/-10 °, 4) scaling from 60% to 140%;

the analog dial reading identification unit is used for identifying an analog dial target area, reading analog pointer display and converting the analog pointer display into a digital value, adopting a neuron training target value, compressing a trained analog dial reading identification model through a TensorFlow Lite, compiling by combining with a main control program firmware, and finally integrally deploying to an edge, wherein the pointer reading identification unit can identify pointer tables of different types, the analog dial reading identification unit is not limited to a fixed type, the analog dial reading identification is a classical image classification task, the classical method is to use a convolutional neural network model with 10 output categories, representing 10 categories from 0 to 9, and further accurately reading the two latter three categories of decimal points, meaning that the model is output with 100 categories (0.0, 0.1, 9.8 and 9.9), a large number of pictures are needed for training, and in order to avoid a large number of images and final working images caused by 100 multi-classification tasks, the training is realized by the fact that the target value is scattered by a plurality of the neural network, and the training power is improved by the fact that the training is realized by the fact that the neural network is scattered by using a specific neural network, and the brightness is greatly shifted, the training power is realized by the fact that the training of the neural network is input to the specific image is greatly, and the brightness is improved in the aspect of the aspect that the training is realized by using a CNnetwork, and the training is greatly shifted by the target value: 1) And (3) brightness dispersion: +/-30%, 2) pixel locations are interspersed with +/-1 pixel in each direction. Since the original picture is 4 times larger (142 x142 pixels), it is ensured that there is sufficient uncertainty in the upstream image alignment and cropping process;

the main control and communication unit is used for four processes including reading synthesis, communication program, file management and configuration management; reading synthesis: synthesizing a final meter reading according to the output result of the digital dial reading identification model, the output result of the analog dial reading identification model and other information recognized by OCR; communication program: the timing period release of meter reading and original photographed images is completed through the MQTT protocol of the Internet of things, and meanwhile, inquiry and control commands of a server side can be received; file management: the system has the functions of a file server, supports the function of network disconnection data storage and supports the functions of OTA upgrading and REST API; configuration management: the working parameters and the network communication parameters of the module are configured directly through the built-in Web interface.

Further, the structure management unit comprises a clamp customizing unit and a module shell customizing unit, the clamp customizing unit and the module shell customizing unit are connected with the structure management unit in a two-way communication mode, and the clamp customizing unit is used for customizing the processing adaptation clamp according to the appearance structure of the meter to be identified; the module shell customizing unit is used for customizing and processing the adaptation module shell according to the appearance structure of the meter to be identified.

The realization method of the intelligent edge machine vision meter reading module of the internet of things comprises the following specific steps:

s1, firstly, generating the edge machine vision meter reading module, and acquiring a large amount of dial image data by using edge hardware equipment;

s2, modeling dial image data based on a machine vision model, and accurately classifying images by using a convolutional neural network CNN to obtain data in the dial, wherein the structure of the convolutional neural network CNN comprises a plurality of Conv2D layers, maxPooling layers and flame layers;

s3, compressing the training model by using a TensorFlow Lite, and finally integrating the training model into the firmware of the edge equipment, so that the rapid intelligent meter reading function of the edge side is realized.

Examples

S1, in the process of image acquisition through a first image acquisition unit, light is supplemented through an LED, a data storage unit is used for storing data and service programs by adopting an SD card, the first image acquisition unit and the data storage unit are managed to work normally through a control and communication unit, and an Internet of things solution is adopted;

s2, receiving acquisition data sent by the hardware management unit through a second image acquisition unit, and aligning by an image preprocessing unit: aligning the input dial image, identifying a subsequent model, and rotating the input image by a fixed angle by the image preprocessing unit in the aligning process; fine tuning: fine tuning is to place the pre-aligned image in a relatively well-defined position to align the image with a fixed ROI for final segmentation; cutting: the trimmed image is in the framed position, and the ROI is segmented and stored as a single picture; the digital dial reading identification unit classifies images through a deep convolutional neural network CNN, compresses a trained digital dial reading identification model through a TensorFlow Lite, compiles by combining a main control program firmware, and finally is integrally deployed to edge equipment; the analog dial reading identification unit is used for identifying an analog dial target area, reading analog pointer display and converting the analog pointer display into a digital value, adopting a neuron training target value, compressing a trained analog dial reading identification model through a TensorFlow Lite, compiling by combining with a main control program firmware, and finally integrally deploying to an edge device, wherein the four processes of reading synthesis, a communication program, file management and configuration management are used by the main control and communication unit; reading synthesis: synthesizing a final meter reading according to the output result of the digital dial reading identification model, the output result of the analog dial reading identification model and other information recognized by OCR; communication program: the timing period release of meter reading and original photographed images is completed through the MQTT protocol of the Internet of things, and meanwhile, inquiry and control commands of a server side can be received; file management: the system has the functions of a file server, supports the function of network disconnection data storage and supports the functions of OTA upgrading and REST API; configuration management: the working parameters and the network communication parameters of the module are configured directly through the built-in Web interface;

s3, customizing and processing the adapting clamp according to the exterior structure of the meter to be identified through the clamp customizing unit, customizing and processing the adapting module shell according to the exterior structure of the meter to be identified through the module shell customizing unit, and forming the meter reading module structure by the clamp and the module hardware shell.

Specific: the meter reading module structure in actual use consists of two parts of a clamp and a module hardware shell, the machine vision meter reading module based on deep learning is successfully deployed to the edge end for the first time in the automatic meter reading field, and compared with the current meter reading scheme, the scheme has the following advantages: firstly, the installation and the deployment are convenient, disassembly, assembly and transformation are not needed, the intelligent meter reading function of the traditional meter can be realized rapidly, and time and labor are saved; secondly, the measurement accuracy and reliability of the original mechanical meter are maintained, and meanwhile meter reading errors can be corrected through a machine vision AI model, so that the measurement accuracy is ensured; thirdly, compared with the replacement of the intelligent meter, the pipeline does not need to be closed, and normal use of a user is not affected; secondly, the equipment and the installation labor cost are low; fourthly, the network communication system has the functions of the Internet of things such as WIFI, NB-IoT and the like, and can meet the requirements of network communication in various application scenes; fifthly, the intelligent meter reading system has the functions of parameter configuration, OTA firmware upgrading function, third party integration API and the like, edge machine vision meter reading module hardware of the Internet of things is designed and developed, edge side deployment is completed by combining a meter reading model, the intelligent meter reading function of traditional analog and digital meters is realized, the problems that time and labor are wasted, cost is too high and metering accuracy is difficult to guarantee in the traditional manual meter reading and intelligent meter changing method are solved, meanwhile, the intelligent level of overall meter reading management is improved, and an edge end controller and image acquisition recommendation selection: ESP32-CAM, ESP8266-ArducAM, etc.; the data storage recommends selection of SD cards, micro SD cards, and the like.

Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.

Claims (10)

1. The intelligent edge machine vision meter reading module of the Internet of things is characterized by comprising a hardware management unit, a software management unit and a structure management unit, wherein the hardware management unit, the software management unit and the structure management unit all keep data real-time intercommunication and sharing through the Internet of things;

the hardware management unit is used for collecting and storing normal images, and supplementing light through the LEDs in the image collecting process;

the software management unit is used for preprocessing the acquired images in an alignment, fine adjustment and cutting mode, classifying the images through a deep convolutional neural network CNN, compressing a trained digital dial reading identification model through a TensorFlow Lite, compiling in combination with a main control program firmware, and deploying the whole to edge equipment for meter reading;

the structure management unit is used for customizing and processing the adapting fixture and the module shell according to the appearance structure of the meter to be identified.

2. The internet of things intelligent edge machine vision meter reading module according to claim 1, wherein: the system comprises a hardware management unit, a software management unit and a structure management unit, wherein the output end of the hardware management unit is in communication connection with the input end of the software management unit, and the hardware management unit and the software management unit are both in bidirectional communication connection with the structure management unit.

3. The internet of things intelligent edge machine vision meter reading module according to claim 1, wherein: the hardware management unit comprises a control and communication unit, a first image acquisition unit and a data storage unit, wherein the first image acquisition unit and the data storage unit are both in bidirectional communication connection with the control and communication unit, and the output end of the first image acquisition unit is in communication connection with the input end of the data storage unit.

4. The internet of things intelligent edge machine vision meter reading module according to claim 3, wherein:

the control and communication unit is used for managing the first image acquisition unit and the data storage unit to work normally, and is also used for adopting an Internet of things solution;

the first image acquisition unit is used for supplementing light through the LEDs in the image acquisition process;

the data storage unit is used for receiving the acquired data sent by the first image acquisition unit, and is also used for storing data and service programs by adopting an SD card.

5. The internet of things intelligent edge machine vision meter reading module according to claim 1, wherein: the software management unit comprises a second image acquisition unit, an image preprocessing unit, a digital dial reading identification unit, an analog dial reading identification unit, a main control and communication unit, wherein the output end of the second image acquisition unit is in communication connection with the input end of the image preprocessing unit, and the second image acquisition unit, the image preprocessing unit, the digital dial reading identification unit and the analog dial reading identification unit are in bidirectional communication connection with the main control and communication unit.

6. The internet of things intelligent edge machine vision meter reading module according to claim 5, wherein:

the second image acquisition unit is used for receiving the acquisition data sent by the hardware management unit, and is also used for sending the acquisition data to the inside of the image preprocessing unit for preprocessing;

the image preprocessing unit is used for receiving the acquired data sent by the second image acquisition unit, and the preprocessing mode of the image preprocessing unit comprises alignment, fine adjustment and cutting;

alignment: aligning the input dial image, identifying a subsequent model, and rotating the input image by a fixed angle by the image preprocessing unit in the aligning process;

fine tuning: fine tuning is to place the pre-aligned image in a relatively well-defined position to align the image with a fixed ROI for final segmentation;

cutting: the trimmed image is in the framed position, and the ROI is segmented and stored as a single picture;

the digital dial reading identification unit is used for classifying images through a deep convolutional neural network CNN, compressing a trained digital dial reading identification model through a TensorFlow Lite, compiling by combining with a main control program firmware, and finally integrally deploying to edge equipment;

the analog dial reading identification unit is used for identifying an analog dial target area, reading analog pointer display and converting the analog pointer display into a digital value, adopting a neuron training target value, compressing a trained analog dial reading identification model through a TensorFlow Lite, compiling by combining with a main control program firmware, and finally arranging the analog dial reading identification model to an edge;

the main control and communication unit is used for four processes including reading synthesis, communication program, file management and configuration management;

reading synthesis: synthesizing a final meter reading according to the output result of the digital dial reading identification model, the output result of the analog dial reading identification model and other information recognized by OCR;

communication program: the timing period release of meter reading and original photographed images is completed through the MQTT protocol of the Internet of things, and meanwhile, inquiry and control commands of a server side can be received;

file management: the system has the functions of a file server, supports the function of network disconnection data storage and supports the functions of OTA upgrading and REST API;

configuration management: the working parameters and the network communication parameters of the module are configured directly through the built-in Web interface.

7. The internet of things intelligent edge machine vision meter reading module according to claim 1, wherein: the structure management unit comprises a clamp customizing unit and a module shell customizing unit, and the clamp customizing unit and the module shell customizing unit are in bidirectional communication connection with the structure management unit.

8. The internet of things intelligent edge machine vision meter reading module of claim 7, wherein:

the fixture customizing unit is used for customizing and processing the adaptive fixture according to the appearance structure of the meter to be identified;

the module shell customizing unit is used for customizing and processing the adaptation module shell according to the appearance structure of the meter to be identified.

9. The method for realizing the intelligent edge machine vision meter reading module of the internet of things is characterized by comprising the following steps of: the method comprises the following specific steps:

s1, firstly, generating the edge machine vision meter reading module, and acquiring a large amount of dial image data by using edge hardware equipment;

s2, modeling dial image data based on a machine vision model, and accurately classifying images by using a convolutional neural network CNN to obtain data in the dial;

s3, compressing the training model by using a TensorFlow Lite, and finally integrating the training model into the firmware of the edge equipment, so that the rapid intelligent meter reading function of the edge side is realized.

10. The method for implementing the internet of things intelligent edge machine vision meter reading module according to claim 9, wherein the method comprises the following steps: the convolutional neural network CNN in step S2 includes a plurality of Conv2D layers, maxPooling and flame layers.

Images