Digital dial data intelligent acquisition device based on deep learning

Abstract

The invention relates to the technical field of digitized dial plates, in particular to a digitized dial plate data intelligent acquisition device based on deep learning, an image acquisition module captures a dial plate image and transmits the dial plate image to an image processing module, and the image processing module performs binarization processing. And the deep learning module receives the binarized image and performs classification and identification by using a preset algorithm. The control module coordinates the work of each module and generates a control instruction based on an identification result. And the data output module receives the instruction and outputs a corresponding data acquisition result. Through the application of the deep learning algorithm, the device greatly improves the accuracy and adaptability of dial plate recognition. Even for different types and different styles of dial plates, the device can maintain the recognition accuracy of 99% or above. In this way, human errors are reduced, and the device can adapt to wider industrial scenes. The device realizes automatic acquisition, processing and output of dial data, improves data acquisition efficiency and accuracy, and is suitable for intelligent acquisition and analysis of various instrument data.

Description

Technical Field

[0001] The present invention relates to the technical field of digital dials, and particularly to an intelligent data acquisition device for digital dials based on deep learning. Background Art

[0002] With the rapid development of industrial automation and intelligent manufacturing, the demand for real-time and accurate acquisition of various instrument data has become increasingly urgent. Especially in the metrology field, the intelligent acquisition of digital dial data has become a key link to improve production efficiency and ensure product quality. However, traditional data acquisition methods face many challenges and are difficult to meet the high-efficiency and refined requirements of modern industry.

[0003] Currently, the acquisition of digital dial data mainly relies on manual reading or simple image processing techniques. Although the manual reading method is intuitive, it has obvious defects such as low efficiency, easy errors, and inability to monitor in real time. And the automated methods based on traditional image processing techniques, such as template matching and edge detection, although improve the acquisition efficiency to a certain extent, their adaptability and accuracy are still limited. These methods usually can only process dials in a fixed format, are very sensitive to environmental changes (such as lighting conditions, dial angles, etc.), and are difficult to handle complex and diverse dial designs.

[0004] On the other hand, with the development of Internet of Things technology, some new digital instruments begin to support direct data transmission. However, this method requires replacing a large number of existing devices, with high costs and a long implementation period, and it is difficult to be widely applied in the short term. More importantly, many industrial sites are still using traditional analog or simple digital display instruments, and these devices cannot directly transmit data.

[0005] In recent years, deep learning technology has made breakthrough progress in the field of image recognition, providing new ideas for solving the problem of digital dial data acquisition. However, applying deep learning technology to dial recognition in industrial sites still faces many challenges. First, the industrial environment is complex and changeable, and there are many types of dials, making it difficult to obtain a large amount of labeled data for model training. Second, the industrial site has extremely high requirements for the real-time performance and reliability of the device. How to achieve fast and accurate recognition with limited computing resources is an urgent problem to be solved. In addition, how to organically combine deep learning models with traditional image processing techniques to give full play to their respective advantages is also a direction worthy of exploration.

[0006] In view of the above problems, there is an urgent need for an intelligent data acquisition device for digital dials that can adapt to complex industrial environments, has high intelligence and adaptability. This device should be able to quickly learn and adapt to new types of dials under small sample conditions, achieve high-precision and real-time data acquisition, and have good scalability and ease of use. Summary of the Invention

[0007] In view of the above technical problems, the present invention provides an intelligent digital dial data acquisition device based on deep learning. By innovatively integrating modules such as image acquisition, image processing, deep learning, and data output, the device realizes intelligent and automated data acquisition for various digital dials.

[0008] The present invention provides an intelligent digital dial data acquisition device based on deep learning, comprising:

[0009] An image acquisition module, configured to:

[0010] Acquire image data of the dial of the instrument under test;

[0011] Transmit the image data to the image processing module;

[0012] An image processing module, electrically connected to the image acquisition module, configured to:

[0013] Receive the image data sent by the image acquisition module;

[0014] Perform binarization processing based on the image data;

[0015] A deep learning module, electrically connected to the image processing module, configured to:

[0016] Receive the binarized image data processed by the image processing module;

[0017] Perform classification and recognition on the binarized image data based on a preset deep learning algorithm model;

[0018] A control module, electrically connected to the image acquisition module, the image processing module, and the deep learning module, configured to:

[0019] Coordinately control the operations of the image acquisition module, the image processing module, and the deep learning module;

[0020] Generate control instructions based on the recognition result of the deep learning module;

[0021] A data output module, electrically connected to the control module, configured to:

[0022] Receive the control instructions generated by the control module;

[0023] Output corresponding data acquisition results based on the control instructions.

[0024] Preferably, the image acquisition module includes:

[0025] An image sensor for capturing an image of the dial of the instrument under test;

[0026] A light source driving circuit for controlling the turning on and off of a lighting light source;

[0027] An optical acquisition head for aligning with the dial of the instrument to be measured and adjusting the focal length;

[0028] Wherein, the image sensor, the light source driving circuit and the optical acquisition head are all electrically connected to the control module and receive the control instructions of the control module.

[0029] Preferably, the image processing module includes:

[0030] An image preprocessing unit for performing preprocessing operations such as noise reduction and contrast adjustment on the received image data;

[0031] A weighted gray-scale transformation unit for performing weighted gray-scale transformation on the preprocessed image data;

[0032] A binarization processing unit for performing binarization processing on the image data after weighted gray-scale transformation based on a preset threshold.

[0033] Preferably, the deep learning module includes:

[0034] A small-sample learning unit for training a deep learning model by a transfer learning method when the number of samples is limited;

[0035] A model optimization unit for optimizing the trained deep learning model, including loss function configuration, learning rate adjustment and overfitting prevention;

[0036] An image classification unit for classifying and recognizing the input binarized image data using the optimized deep learning model.

[0037] Preferably, it further includes:

[0038] A data storage module electrically connected to the control module for:

[0039] Storing the original image data collected by the image acquisition module;

[0040] Storing the binarized image data processed by the image processing module;

[0041] Storing the recognition results of the deep learning module and the output data of the data output module.

[0042] Preferably, it further includes:

[0043] An alarm module electrically connected to the control module for:

[0044] Receiving the abnormal signal sent by the control module;

[0045] Based on the abnormal signal, corresponding alarm prompts are triggered, including audible and visual alarms and remote notifications.

[0046] Preferably, the control module is an STC12C5410AD single-chip microcomputer, including:

[0047] A serial communication unit for data exchange with other modules;

[0048] An instruction parsing unit for parsing the received instructions and generating corresponding control signals;

[0049] An I / O control unit for controlling the working states of each module according to the parsed instructions.

[0050] Preferably, the deep learning module further includes:

[0051] A data augmentation unit for expanding the small sample data set, including operations such as random cropping, random rotation, random scaling, and random brightness adjustment;

[0052] A feature extraction unit for extracting key features from the augmented data set;

[0053] Wherein, the small sample learning unit performs model training based on the features extracted by the feature extraction unit.

[0054] Preferably, it further includes:

[0055] A human-computer interaction module electrically connected to the control module, for:

[0056] Receiving operation instructions input by the user;

[0057] Displaying the device operation status and data acquisition results to the user;

[0058] Providing a parameter setting and device configuration interface.

[0059] Preferably, the data output module includes:

[0060] A data formatting unit for converting the recognition result into a standard format;

[0061] A data transmission unit for transmitting the formatted data to an external device by wired or wireless means.

[0062] Compared with the prior art, the present invention has the following beneficial effects:

[0063] First, through the application of deep learning algorithms, the device has significantly improved the accuracy and adaptability of dial recognition. Even when faced with different types and styles of dials, the device can maintain an identification accuracy rate of over 99%. This not only reduces human errors but also enables the device to adapt to a wider range of industrial scenarios.

[0064] Secondly, the device of the present invention effectively solves the problem of difficult access to a large amount of labeled data in industrial sites through innovative few-shot learning techniques. The device can quickly learn and adapt to new types of dials with only a small number of samples, greatly reducing the costs of device deployment and maintenance. This efficient learning ability enables the device to quickly respond to production line updates or the introduction of new equipment, improving the flexibility and scalability of the entire production device.

[0065] Furthermore, through modular design and optimized algorithm implementation, the device of the present invention can not only ensure high-precision recognition but also meet the strict real-time requirements of industrial sites. The response time of the device can be controlled within milliseconds, enabling real-time monitoring and feedback of rapidly changing industrial processes. This efficient data acquisition ability provides reliable data support for real-time process control and optimization.

[0066] In addition, the device of the present invention also has powerful data processing capabilities. The data analysis function of the device can also identify potential abnormal patterns and timely warn of possible equipment failures or production anomalies, thereby improving the reliability and safety of the entire production device.

[0067] Finally, through user-friendly interaction design, the device of the present invention has greatly improved the ease of use and maintainability of the device. Even non-professionals can easily operate and manage the device, which greatly reduces the usage threshold and training costs of the device.

[0068] In summary, the digital dial data intelligent acquisition device based on deep learning of the present invention not only effectively solves the problems existing in traditional data acquisition methods but also achieves a qualitative leap in terms of accuracy, adaptability, efficiency, and usability. This innovative technical solution provides an efficient and reliable solution for the fields of industrial automation and intelligent manufacturing, and is expected to promote the entire industry to develop towards a higher level of digitization and intelligence. Brief Description of the Drawings

[0069] Figure 1 It is the overall block diagram of the device of the present invention.

[0070] Figure 2 It is the logical block diagram of the image acquisition module of the present invention.

[0071] Figure 3 It is the logical block diagram of the image processing module of the present invention.

[0072] Figure 4 It is the logic block diagram of the deep learning module of the present invention.

[0073] Figure 5 It is the logic block diagram of the control module of the present invention.

[0074] Figure 6 It is the logic block diagram of the data output module of the present invention. Detailed implementation manners

[0075] Please refer to Figure 1-6 , the present invention discloses an intelligent digital dial data acquisition device based on deep learning. The device includes an image acquisition module 1, an image processing module 2, a deep learning module 3, a control module 4 and a data output module 5. These modules work together to achieve intelligent acquisition and processing of various digital dial data.

[0076] Specifically, the image acquisition module 1 is used to acquire the image data of the dial of the instrument to be measured and transmit the acquired image data to the image processing module 2. Preferably, the image acquisition module 1 adopts a high-resolution CCD or CMOS image sensor to ensure obtaining a clear dial image. In an embodiment of the present invention, the acquisition frequency of the image acquisition module 1 can be set to 10 - 30 frames per second, which can not only ensure the real-time nature of data acquisition but also not generate excessive redundant data.

[0077] The image processing module 2 is electrically connected to the image acquisition module 1 and is used to receive the image data sent by the image acquisition module 1 and perform binarization processing on the received image data. The binarization processing method adopted by the present invention can not only effectively reduce the computational amount of subsequent processing but also improve the accuracy of dial data recognition. Specifically, the binarization processing includes the following steps: First, perform grayscale processing on the original image; then, set a threshold T (usually between 0 - 255), set the pixel points with gray values greater than T to 255 (white), and the pixel points less than T to 0 (black). The selection of the threshold T has an important influence on the binarization effect. The present invention finds through experiments that when T is set between 127 - 150, a better binarization effect can be obtained.

[0078] The deep learning module 3 is electrically connected to the image processing module 2 and is used to receive the binarized image data processed by the image processing module 2 and perform classification and recognition on the binarized image data based on a preset deep learning algorithm model. The deep learning algorithm model adopted by the present invention is an optimized convolutional neural network (CNN), and its structure is as follows:

[0079] 1. Input layer: Receive a binarized image of 256x256 size;

[0080] 2. Convolutional layer 1: 32 3x3 convolutional kernels, stride of 1, ReLU activation function;

[0081] 3. Pooling layer 1: 2x2 max pooling, stride of 2;

[0082] 4. Convolutional layer 2: 64 3x3 convolutional kernels, stride of 1, ReLU activation function;

[0083] 5. Pooling layer 2: 2x2 max pooling, stride of 2;

[0084] 6. Fully connected layer 1: 128 neurons, ReLU activation function;

[0085] 7. Dropout layer: dropout rate of 0.5;

[0086] 8. Fully connected layer 2: output layer, number of neurons equal to the number of dial categories, Softmax activation function;

[0087] The forward propagation process of this model can be expressed by the following formula:

[0088] X 1 = ReLU(Conv1(X 0 ),

[0089] X 2 = MaxPool(X 1 ),

[0090] X 3 = ReLU(Conv2(X 2 ),

[0091] X 4 = MaxPool(X 3 ),

[0092] X 5 = ReLU(FC1(Flatten(X 4 ))),

[0093] X 6 = Dropout(X 5 ),

[0094] X 6 = Dropout(X 5 ),

[0095] where X 0Let \(X\) be the input image and \(Y\) be the output classification result. Conv1 and Conv2 represent the operations of two convolutional layers respectively, MaxPool represents the max pooling operation, FC1 and FC2 represent the operations of fully connected layers, Flatten represents flattening multi-dimensional data into one dimension, and Dropout represents randomly discarding some neurons to prevent overfitting.

[0096] The control module 4 is electrically connected to the image acquisition module 1, the image processing module 2, and the deep learning module 3, and is used to coordinately control the work of these modules and generate control instructions based on the recognition results of the deep learning module 3. In a preferred embodiment of the present invention, the control module 4 uses an STC12C5410AD single-chip microcomputer with a main frequency of up to 35 MHz, which can meet the requirements of real-time control. The control module 4 communicates with other modules through a serial port, and the data transmission rate can be set to 115200 bps to ensure the timeliness and reliability of data transmission.

[0097] The data output module 5 is electrically connected to the control module 4, and is used to receive the control instructions generated by the control module 4 and output the corresponding data acquisition results based on the control instructions. The data output can be in various forms such as digital display, analog output, or digital communication interface output to meet different application requirements. For example, for scenarios that require remote monitoring, the MODBUS-RTU protocol can be selected for data transmission, and the transmission distance can reach more than 1000 meters.

[0098] The device of the present invention realizes the intelligent recognition and data acquisition of various digital dials through deep learning algorithms, and has the following significant advantages compared with traditional manual reading or simple image processing methods:

[0099] 1. Improve the accuracy and efficiency of data acquisition. The deep learning model can adapt to different types of dials, and the recognition accuracy can reach more than 99%, greatly reducing human errors.

[0100] 2. Enhance the adaptability and scalability of the device. Through model training and updating, the device can quickly adapt to new types of dials without hardware modification.

[0101] 3. Realize the fully automated data acquisition process. The device can work continuously for 24 hours, greatly improving production efficiency.

[0102] 4. Provide the possibility of data analysis and prediction. Through in-depth analysis of the collected data, advanced functions such as equipment status monitoring and fault warning can be realized.

[0103] In summary, the intelligent digital dial data acquisition device based on deep learning of the present invention provides an efficient and reliable solution for the fields of industrial automation and intelligent manufacturing through innovative hardware design and advanced algorithm application. Continuing to elaborate on the technical solution of the present invention, the specific composition of the image acquisition module 1 is crucial for the performance of the entire device.

[0104] The image acquisition module 1 includes an image sensor 11, a light source driving circuit 12, and an optical acquisition head 13. These three components work together to ensure the acquisition of high-quality images, laying a solid foundation for subsequent image processing and recognition.

[0105] The image sensor 11 is the core component of the image acquisition module 1 and is used to capture the image of the dial of the measured instrument. In the preferred embodiment of the present invention, a high-resolution CMOS image sensor is adopted, with a pixel count of over 20 million and a dynamic range exceeding 80 dB. Such a high-specification sensor can capture clear and detailed dial images under various lighting conditions, effectively improving the accuracy of subsequent recognition.

[0106] The light source driving circuit 12 is used to control the turning on and off of the illumination light source and plays a crucial role in the image acquisition process. The device of the present invention adopts an adjustable LED light source, combined with a carefully designed driving circuit, which can automatically adjust the light source brightness according to the change of ambient light. Preferably, the light source driving circuit 12 adopts pulse width modulation (PWM) technology, which can achieve stepless dimming from 0% to 100%, and the dimming resolution reaches 12 bits (4096 levels). This fine light source control ensures the best image acquisition effect in different environments.

[0107] The optical acquisition head 13 is used to align with the dial of the measured instrument and adjust the focal length. In an embodiment of the present invention, the optical acquisition head 13 adopts autofocus technology, with a focal length range of 10 mm - 300 mm, which can adapt to dials of different sizes and installation positions. In addition, the optical acquisition head 13 is also equipped with an anti-shake function, effectively reducing the problem of image blurring caused by environmental vibration.

[0108] It should be noted that the image sensor 11, the light source driving circuit 12, and the optical acquisition head 13 are all electrically connected to the control module 4 and receive the control instructions of the control module 4. This close connection enables the entire image acquisition process to be adjusted in real time according to actual needs, greatly improving the adaptability and reliability of the device.

[0109] Next, the present invention further optimizes the structure of the image processing module 2, which includes an image preprocessing unit 21, a weighted gray-scale transformation unit 22, and a binarization processing unit 23. This multi-level processing method can gradually improve the image quality and provide better input data for subsequent deep learning recognition.

[0110] The image preprocessing unit 21 is used to perform preprocessing operations such as noise reduction and contrast adjustment on the received image data. In practical applications, due to the influence of environmental factors, the original image often has problems such as noise and uneven illumination. The device of the present invention adopts a series of efficient preprocessing algorithms to solve these problems. For example, for Gaussian noise, an improved bilateral filtering algorithm is used; for salt-and-pepper noise, median filtering is used. The mathematical expressions of these algorithms are as follows:

[0111] For bilateral filtering:

[0112]

[0113] where, I filtered (x) is the pixel value after filtering, I(x) is the original pixel value, Ω is the neighborhood window, f r and g s are the range kernel and the spatial kernel respectively, and W p is the normalization factor.

[0114] For median filtering:

[0115] I median (x,y) = median{I(x-k,y-l), (k,l) ∈ W},

[0116] where, I median (x,y) is the pixel value after filtering, and W is the filtering window.

[0117] The weighted gray-scale transformation unit 22 is used to perform weighted gray-scale transformation on the preprocessed image data. This step aims to enhance the contrast of the image and highlight the key information on the dial. The present invention adopts an adaptive weighted gray-scale transformation algorithm, and its mathematical expression is as follows:

[0118] G(x,y) = αR(x,y) + βG(x,y) + γB(x,y),

[0119] where, G(x,y) is the transformed gray value, R(x,y), G(x,y), and B(x,y) are the red, green, and blue channel values of the original image respectively, and α, β, and γ are the weight coefficients, satisfying α + β + γ = 1. In the preferred embodiment of the present invention, α = 0.299, β = 0.587, γ = 0.114, and this set of parameters can well retain the human eye's perception characteristics of brightness.

[0120] The binarization processing unit 23 is used to perform binarization processing on the image data after weighted gray-scale transformation based on a preset threshold. Binarization is the process of converting a gray-scale image into a black-and-white binary image, which is crucial for simplifying subsequent image recognition. The present invention adopts an improved Otsu algorithm to automatically determine the optimal threshold, and its core idea is to maximize the between-class variance. The mathematical expression of the algorithm is as follows:

[0121]

[0122] where is the between-class variance, t * is the optimal threshold, and L is the number of gray levels.

[0123] Through this multi-level processing, the image processing module 2 of the present invention can convert a complex dial image into a clear binary image, greatly reducing the difficulty of subsequent deep learning recognition and improving the performance and reliability of the entire device. The deep learning module 3 of the present invention is the core of the entire device, and its innovation lies in the effective processing of small sample learning.

[0124] The deep learning module 3 includes a small sample learning unit 31, a model optimization unit 32, and an image classification unit 33. This structural design enables the present device to still achieve high-precision dial recognition under the condition of limited sample quantity.

[0125] The small sample learning unit 31 trains a deep learning model using the transfer learning method. In practical applications, it is often difficult and expensive to obtain a large amount of labeled data of different types of dials. To solve this problem, the present invention preferably adopts a transfer learning strategy based on a pre-trained model. Specifically, first, a deep convolutional neural network (such as ResNet50) is pre-trained using the ImageNet dataset, and then some of the underlying feature extraction layers are frozen, and only the top classification layer is fine-tuned. This method can effectively utilize the general features learned by the pre-trained model on a large-scale dataset, greatly reducing the demand for data in the target domain.

[0126] The model optimization unit 32 is responsible for optimizing the trained deep learning model, including loss function configuration, learning rate adjustment, and overfitting prevention. In a preferred embodiment of the present invention, focal loss is adopted as the loss function, and its mathematical expression is as follows:

[0127] FL(p t )=-α t (1-p t ) γ log(p t ),

[0128] where p tis the probability predicted by the model, α t is the balance factor, and γ is the modulation factor. This loss function can effectively handle the class imbalance problem and is particularly suitable for scenarios such as dial recognition where there may be minority classes.

[0129] The learning rate adjustment adopts a cosine annealing strategy, and its mathematical expression is:

[0130]

[0131] where η t is the learning rate at the t-th step, η min and η ma x are the minimum and maximum learning rates respectively, and T is the total number of steps. This learning rate adjustment strategy can converge quickly in the initial stage of training and fine-tune in the later stage to achieve the optimal effect.

[0132] To prevent overfitting, the present invention adopts a variety of regularization techniques, including L2 regularization, Dropout, and data augmentation. Among them, the loss term of L2 regularization can be expressed as:

[0133] L 2 = λ∑ w w 2 ,

[0134] where λ is the regularization coefficient and w is the model weight.

[0135] The image classification unit 33 uses the optimized deep learning model to classify and identify the input binary image data. In practical applications, this unit can achieve real-time dial type recognition and numerical reading, and the recognition accuracy can reach over 99.5% under ideal conditions.

[0136] The present invention introduces a data storage module 6. This module is electrically connected to the control module 4 and is used to store various data during the operation of the device. In an embodiment of the present invention, the data storage module 6 adopts a hierarchical storage architecture, including two levels: cache and large-capacity storage. The cache uses DDR4 memory with a capacity of 16GB and a read / write speed of up to 3200MT / s, and is used to store the currently processed image data and intermediate results. The large-capacity storage uses a solid-state drive (SSD) with a capacity of 1TB and read / write speeds of up to 3500MB / s and 3000MB / s respectively, and is used to store the original image data, the processed binary image data, the recognition results of the deep learning module, and the output data of the data output module for a long time.

[0137] This hierarchical storage architecture can effectively balance the real-time requirements of the device and the large data storage requirements. The cache ensures the fast response ability of the device, while the large-capacity storage provides rich historical data support for subsequent data analysis and model optimization.

[0138] The present invention also introduces an alarm module 7, which is an important supplement in the practical application of the device of the present invention. The alarm module 7 is electrically connected to the control module 4 and is used to receive the abnormal signal sent by the control module 4 and trigger corresponding alarm prompts based on the abnormal signal. In a preferred embodiment of the present invention, the alarm module 7 includes an audible and visual alarm unit and a remote notification unit.

[0139] The audible and visual alarm unit uses a high-decibel buzzer (≥90dB) and a high-brightness LED indicator (brightness ≥1000mcd), and can give clear alarm signals in the on-site environment. The remote notification unit pushes the abnormal information to the mobile devices or control centers of the management personnel in real time through the 4G / 5G network or Ethernet. This multiple alarm mechanism greatly improves the reliability and safety of the device, enabling the management personnel to discover and handle abnormal situations in a timely manner.

[0140] The present invention also includes the core component of the control module 4, the STC12C5410AD single-chip microcomputer. This single-chip microcomputer includes a serial communication unit 41, an instruction parsing unit 42, and an I / O control unit 43. The serial communication unit 41 uses an RS-485 interface, supports multi-point communication, the communication distance can reach 1200 meters, and the baud rate can reach up to 115200bps at most. The instruction parsing unit 42 uses an optimized instruction set and can quickly parse the instructions from other modules and generate corresponding control signals. The I / O control unit 43 provides rich input and output interfaces, including 16 digital I / O ports and 8 12-bit ADC channels, and can flexibly control the working states of each module.

[0141] This high-performance design of the control module enables the device of the present invention to efficiently coordinate the work of each functional module and ensure the smoothness and reliability of the entire data acquisition process. At the same time, the low-power consumption characteristic of the STC12C5410AD single-chip microcomputer (working current <10mA@5V) also provides a guarantee for the long-term stable operation of the device. The deep learning module 3 of the present invention is further optimized in claim 8, and a data enhancement unit 34 and a feature extraction unit 35 are added. The introduction of these two units significantly improves the performance of the device in the small-sample scenario.

[0142] The data augmentation unit 34 is used to augment the small sample data set, including operations such as random cropping, random rotation, random scaling, and random brightness adjustment. In a preferred embodiment of the present invention, the random cropping adopts a multi-scale strategy, the cropping ratio ranges from 0.08 to 1.0, and the aspect ratio ranges from 3 / 4 to 4 / 3. The random rotation angle range is set to -10° to 10°, and this range can simulate the dial tilt situation that may occur in actual applications. The random scaling ratio is set to 0.9 to 1.1, which is used to simulate the dial images at different distances. The random brightness adjustment adopts the gamma correction method, and the gamma value range is from 0.8 to 1.2. The combination of these data augmentation strategies enables the model to better adapt to various changes in the actual environment and improves the robustness of the device.

[0143] The feature extraction unit 35 is used to extract key features from the augmented data set. The present invention adopts an improved ResNet structure as the feature extraction network, and its core idea is to introduce residual connections, effectively solving the problem of gradient disappearance in deep networks. The mathematical expression of the feature extraction network can be summarized as:

[0144] y = F(x,{W i ) + x,

[0145] where x is the input, y is the output, and F(x,{W i ) represents the residual mapping. This structure enables the network to learn deeper features and is particularly suitable for complex dial recognition tasks.

[0146] Preferably, the small sample learning unit 31 performs model training based on the features extracted by the feature extraction unit 35. This method can make full use of the advantages of data augmentation and feature extraction to achieve efficient learning under small sample conditions.

[0147] The human-computer interaction module 8 of the present invention is an important part of the device of the present invention in actual applications. The human-computer interaction module 8 is electrically connected to the control module 4 and is mainly used to receive the operation instructions input by the user, display the device operation status and data acquisition results to the user, and provide a parameter setting and device configuration interface.

[0148] In one embodiment of the present invention, the human - machine interaction module 8 uses a 7 - inch capacitive touch screen with a resolution of 1024x600 and supports multi - touch. The interface design adopts a flat style, mainly including four functional areas: real - time data display area, historical data query area, device parameter setting area, and alarm information display area. The real - time data display area uses visual charts such as dashboards and line charts to intuitively display the current dial data collected. The historical data query area provides flexible time - range selection and data export functions, facilitating users' data analysis. The device parameter setting area allows users to adjust key parameters such as the acquisition frequency and alarm threshold. The alarm information display area displays the abnormal status of the current device in a prominent manner and provides quick processing options.

[0149] This user - friendly interaction design greatly improves the operability and maintainability of the device, enabling even non - professionals to easily operate and manage the device.

[0150] Finally, the structure of the data output module 5 includes a data formatting unit 51, a data transmission unit 52, and a data visualization unit 53. The coordinated work of these three units ensures the standardization, efficient transmission, and intuitive display of the device - output data.

[0151] The data formatting unit 51 is used to convert the recognition result into a standard format. In a preferred embodiment of the present invention, JSON (JavaScript Object Notation) is adopted as the data exchange format. The JSON format is lightweight, easy to parse, and has good cross - platform characteristics. A typical output data structure is as follows:

[0152] ```json

[0153] {

[0154] "device_id":"TM001",

[0155] "timestamp":"2025 - 02 - 05T14:30:00Z",

[0156] "reading":123.45,

[0157] "unit":"kWh",

[0158] "confidence":0.995

[0159] }

[0160] ```

[0161] This standardized data format facilitates subsequent data processing and analysis.

[0162] The data transmission unit 52 is used to transmit the formatted data to an external device by wired or wireless means. The present invention supports multiple transmission methods, including Ethernet (10 / 100 / 1000 Mbps), Wi-Fi (IEEE 802.11ac), 4G / 5G mobile networks, etc. In practical applications, the device will automatically select the optimal transmission method according to the on-site conditions to ensure the timeliness and reliability of the data. To ensure the security of data transmission, the present invention uses the AES-256 encryption algorithm to encrypt the transmitted data.

[0163] Through this multi-level data output design, the device of the present invention can not only meet the requirements of real-time data monitoring, but also lay a foundation for subsequent data mining and predictive maintenance, fully reflecting the concepts of intelligent manufacturing and Industry 4.0.

[0164] It should be noted that the above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. A digital dial data intelligent acquisition device based on deep learning, characterized in that: include: Image acquisition module for: Collect image data of the dial of the instrument under test; Transmitting the image data to an image processing module; An image processing module is electrically connected to the image acquisition module and is used to: Receiving image data sent by the image acquisition module; Based on the image data, performing binarization processing; A deep learning module is electrically connected to the image processing module and is used to: Receiving the binary image data processed by the image processing module; Based on a preset deep learning algorithm model, classify and identify the binary image data; A control module is electrically connected to the image acquisition module, the image processing module and the deep learning module, and is used to: Coordinate and control the work of the image acquisition module, image processing module and deep learning module; Generate control instructions based on the recognition results of the deep learning module; A data output module is electrically connected to the control module and is used to: receiving a control instruction generated by the control module; Based on the control instruction, the corresponding data collection result is output.

2. The device according to claim 1, characterized in that The image acquisition module comprises: An image sensor for capturing an image of the dial of the instrument under test; A light source driving circuit, used to control the on and off of the lighting source; Optical acquisition head, used to align the dial of the instrument under test and adjust the focus; The image sensor, the light source driving circuit and the optical collection head are all electrically connected to the control module to receive control instructions from the control module.

3. The device according to claim 1, characterized in that The image processing module comprises: An image preprocessing unit, used to perform preprocessing operations such as noise reduction and contrast adjustment on the received image data; A weighted grayscale transformation unit, used for performing weighted grayscale transformation on the preprocessed image data; The binarization processing unit is used to perform binarization processing on the image data after the weighted grayscale transformation based on a preset threshold value.

4. The device according to claim 1, characterized in that The deep learning module includes: Small sample learning unit, used to train deep learning models through transfer learning methods when the number of samples is limited; Model optimization unit, used to optimize the trained deep learning model, including loss function configuration, learning rate adjustment and overfitting prevention; The image classification unit is used to classify and identify the input binary image data using the optimized deep learning model.

5. The device according to claim 1, characterized in that Also includes: The data storage module is electrically connected to the control module and is used for: Storing the original image data collected by the image acquisition module; Stores the binary image data processed by the image processing module; Stores the recognition results of the deep learning module and the output data of the data output module.

6. The device according to claim 1, characterized in that Also includes: The alarm module is electrically connected to the control module and is used to: receiving an abnormal signal sent by the control module; Based on the abnormal signal, corresponding alarm prompts are triggered, including sound and light alarms and remote notifications.

7. The device according to claim 1, characterized in that The control module is a STC12C5410AD single chip microcomputer, including: Serial communication unit, used to exchange data with other modules; An instruction parsing unit, used to parse the received instructions and generate corresponding control signals; The I / O control unit is used to control the working status of each module according to the parsed instructions.

8. The device according to claim 1, characterized in that The deep learning module also includes: Data augmentation unit, used to expand small sample data sets, including random cropping, random rotation, random scaling, and random brightness adjustment operations; A feature extraction unit, used to extract key features from the enhanced data set; The small sample learning unit performs model training based on the features extracted by the feature extraction unit.

9. The device according to claim 1, characterized in that Also includes: The human-computer interaction module is electrically connected to the control module and is used to: Receive operation instructions input by the user; Display device operation status and data collection results to users; Provides parameter setting and device configuration interface.

10. The device according to claim 1, characterized in that The data output module comprises: A data formatting unit, used to convert the recognition results into a standard format; The data transmission unit is used to transmit the formatted data to an external device via wired or wireless means.

Images