Screen reading identification method and system for alpha and beta surface contamination instrument

Abstract

The invention discloses an alpha and beta surface contamination instrument screen reading identification method and system, and the method comprises the steps: obtaining an alpha and beta surface contamination instrument screen image as a sample data set, and marking the sample data set; establishing a target detection model for detecting digits, multipliers' * 1000 'and units in the image, and training the model; identifying the screen image by using the target detection model to obtain frame position information of numbers and units; and carrying out image processing and numerical value identification on the identified image, and outputting a screen reading. The method is suitable for various models, automatic identification of screen readings is realized through a deep learning model, the labor cost is reduced, and the construction efficiency is improved. The device has the advantages of improving the verification accuracy, reducing the cost and adapting to various working environments.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, and in particular to a method and system for recognizing screen indications of α and β surface contamination meters. Background Art

[0002] As an important tool for monitoring and evaluating the radiation levels of alpha and beta particles in the working environment, α and β surface contamination meters play an irreplaceable role in the fields of nuclear industry, medical treatment, scientific research, etc. The accurate reading of the screen reading is crucial for automated verification devices and directly affects the assessment of environmental safety and personnel health.

[0003] The traditional manual reading method has problems such as low efficiency and prone to errors. Especially in high radiation environments, manual reading is not only inefficient, but also poses a threat to the health of operators. Therefore, automated image recognition technology has become an effective means to improve reading efficiency and accuracy. Although some image recognition technologies have been applied to the recognition of screen indications, these technologies often require complex preprocessing and feature extraction steps, and the recognition accuracy for α and β surface contamination meters of different brands is not high. Therefore, it is of great practical significance to develop a new method for recognizing the screen indications of α and β surface contamination meters to improve recognition efficiency and accuracy and reduce human errors.

[0004] Existing image recognition technologies mostly rely on traditional image processing algorithms, which are often difficult to adapt to the diversity and complexity of the screen readings of α and β surface contamination meters of different brands and models. In addition, these methods are insufficient in processing speed and accuracy, and cannot meet the needs of industrial automation and intelligent development. Summary of the invention

[0005] The present invention provides a method and system for identifying the screen indications of α and β surface contamination meters. The method significantly improves the recognition efficiency and accuracy of the indications through automated image recognition technology, while reducing human errors. The present invention involves obtaining a screen image sample data set, establishing and training a target detection model, as well as image processing and numerical recognition, and finally outputting the screen indications. Using a deep learning convolutional neural network model, the present invention can adapt to α and β surface contamination meters of different brands and models, achieve high-accuracy recognition, and is suitable for real-time monitoring and recognition in industrial automation production lines.

[0006] A method for identifying the screen indications of an α or β surface contamination meter comprises the following steps:

[0007] Obtain the screen images of α and β surface contamination instruments as sample data sets and annotate them;

[0008] Build and train an object detection model for detecting numbers, multipliers “×1000”, and units in images.

[0009] Using the target detection model to identify the screen image, and obtain the border position information of the numbers and units;

[0010] The recognized image is processed and numerically recognized, and the displayed value is output on the screen.

[0011] Furthermore, the target detection model is a convolutional neural network model based on deep learning.

[0012] Furthermore, the image processing and value recognition steps further include:

[0013] De-noise and enhance the recognized image;

[0014] Extract the value of the number, the multiplier "×1000" and the unit;

[0015] Convert the extracted numerical value into screen display and output it.

[0016] Furthermore, the annotations of the sample data set include the position of the numbers, the value, the multiplier "×1000" and the color or shape of the unit.

[0017] Furthermore, the training of the target detection model includes the following steps:

[0018] Input the labeled sample dataset into the model as training data;

[0019] Calculate the loss function of the model;

[0020] Update the model's weights via backpropagation.

[0021] An α, β surface contamination meter screen indication recognition system, comprising:

[0022] Data acquisition module, used to obtain the screen images of α and β surface contamination instruments;

[0023] Object detection models to detect numbers, multipliers “×1000”, and units in images;

[0024] An image processing module is used to perform image processing and numerical recognition on the recognized image;

[0025] Output module, used to output screen indications.

[0026] Storage module.

[0027] Furthermore, the data acquisition module includes settings for the number of cameras and image buffers, disconnection and reconnection time, image width and height, pixel format, frame rate, exposure mode, and exposure time.

[0028] Furthermore, the target detection model is lightweight to adapt to different computing environments.

[0029] Furthermore, the storage module includes a computer-readable storage medium, and the storage medium stores a computer-executable program.

[0030] Furthermore, the computer executable program further includes a logic flow for implementing image source setting, high-precision matching, position correction, screen indication symbol recognition, formatting processing and data sending.

[0031] Beneficial Effects

[0032] The patent advantages of this invention include high-efficiency identification, high accuracy, reduced human error, strong adaptability, real-time monitoring, reduced work risk and cost savings. Through the application of deep learning technology, this method can achieve high-accuracy identification in α and β surface contamination meters of different brands and models, which has significant benefits for improving the monitoring efficiency and quality control level of industrial automation production lines.

[0033] In the component description, the morphological features, connection features and functions of each component of the present invention are elaborated in detail to ensure the completeness of the description. At the same time, through the description of appropriate associations and alternatives, the scope of the claims is expanded so that the present invention is not limited to specific components or technologies. For example, although a convolutional neural network model is used, any deep learning model that can achieve high-accuracy recognition can be substituted. In addition, the image acquisition module can be not only a camera, but also any device that can capture screen images. The image processing module can also use different image processing technologies as long as the recognition accuracy can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A schematic diagram of the screen identification area of ​​the α and β surface contamination meter provided by the present invention;

[0035] Figure 2 A target detection model training flow chart of the α and β surface contamination meter screen indication recognition method provided by the present invention;

[0036] Figure 3 A flow chart of the method for identifying the screen indications of α and β surface contamination meters provided by the present invention;

[0037] Figure 4 This is a system framework diagram of the method for identifying the screen indications of α and β surface contamination meters provided by the present invention. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solution and beneficial effects of the present invention more clear, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described in this specification are only for explaining the present invention, not for limiting the present invention.

[0039] Embodiment 1:

[0040] In an embodiment of the present invention, a method and system for identifying the screen readings of an α or β surface contamination meter include the following steps:

[0041] Acquire the screen images of the α and β surface contamination meters as sample data sets, and mark the numbers, multipliers “×1000” and unit areas in the sample data sets;

[0042] Establishing an object detection model for detecting numbers, multipliers "×1000" and units in an image, and training the object detection model using the sample dataset;

[0043] The target detection model is used to identify the screen image of the α and β surface contamination meters to obtain the frame position information of the number, the multiplier "×1000" and the unit;

[0044] The recognized image is processed and numerically recognized, and the displayed value is output on the screen.

[0045] In a preferred embodiment, the target detection model is a convolutional neural network (CNN) model based on deep learning. CNN models are widely used due to their excellent performance in image recognition and classification tasks. The following are the key steps and formulas for building and training CNN models:

[0046] Network structure design: Design the number of layers of the CNN model and the parameters of each layer, including convolutional layers, pooling layers, fully connected layers, etc.

[0047] Convolutional layer: The convolutional layer is the core of CNN and is used to extract image features. The convolution operation can be expressed as:

[0048] (I*K)(x,y)=∑ a m＝-a ∑ b n＝--b I(x+m,y+n).K(m,n)

[0049] Where I is the input image, K is the convolution kernel, and a and b are the sizes of the convolution kernel.

[0050] Activation function: An activation function is usually applied after the convolution layer, such as ReLU (Rectified Linear Unit), and its formula is:

[0051] f(x)=max(0,x)

[0052] Pooling layer: The pooling layer is used to reduce the dimension of the feature map and reduce the amount of calculation. The formula for the maximum pooling operation is:

[0053] P(x,y)=max m,n I(x+m,y+n)

[0054] Among them, P is the feature map after pooling.

[0055] Fully connected layer: At the end of CNN, the fully connected layer maps the extracted features to the output space. The output of the fully connected layer can be expressed as:

[0056] O=W.A+b

[0057] Where O is the output, W is the weight matrix, A is the activation value of the previous layer, and b is the bias.

[0058] Loss function: During training, the loss function is used to evaluate the difference between the model prediction and the true label. Commonly used loss functions include cross entropy loss:

[0059]

[0060] in, is the true label, is the predicted probability.

[0061] Back propagation: By calculating the gradient of the loss function with respect to the network parameters, the back propagation algorithm is used to update the weights and biases to minimize the loss function. The gradient update formula is:

[0062] W=W-η(dL / dW)

[0063] Among them, η is the learning rate.

[0064] Optimization algorithms: such as SGD (Stochastic Gradient Descent), Adam, or RMSprop, which are used to optimize model parameters during training.

[0065] Model evaluation: Use the validation set to evaluate model performance. Common indicators include accuracy, precision, recall, and F1 score.

[0066] Through these steps and formulas, a deep learning-based convolutional neural network model can be constructed for object detection tasks. In a preferred embodiment, the model will be trained with a large amount of annotated data to identify and locate specific objects or characters in the image.

[0067] In a preferred embodiment, the target detection model file generation comprises the following steps:

[0068] Acquisition of sample data set for α and β surface contamination meter identification: photos of α and β surface contamination meters with indications in different situations are taken by camera. The photos contain different indication sizes, different screen brightness, and different screen indication numerical symbols.

[0069] Sample image marking: select the position of the α and β surface contamination meter screen display, including numbers, unit symbols, and scientific notation symbols, and then manually confirm and correct the incorrect symbols.

[0070] Image import: Import these images into the dataset, ensuring image diversity, including different display sizes, brightness, and symbols.

[0071] Image annotation: Annotate the numbers, multipliers “×1000”, and units in the image to provide supervision information for model training.

[0072] Create a model: On the training platform, select the algorithm type as "Character training", name the dataset, select the sample photos to be trained, and set automatic photo format conversion.

[0073] Model training: Select the labeled sample images, set the version number and image resolution, set the model capabilities to be "lightweight", then automatically test the model, and finally generate the target detection model file.

[0074] Export model training files: After training is complete, export the model files for deployment and use on other systems or devices.

[0075] In a preferred embodiment, the sample identification logic flow includes the following steps:

[0076] Image source settings: Set the input image source, select local image or associated camera, adjust exposure and gain. You can also set the parameters of the associated camera, such as: image cache quantity, disconnection reconnection time, image width and height, pixel format, frame rate and exposure mode and exposure time. Of course, you can also set the trigger source and trigger delay time.

[0077] High-precision matching: This item is mainly used to adjust the ROI recognition area of ​​the target image. You can manually draw the recognition ROI recognition area to ensure that the screen display is in this area. You can also set the operating parameters, such as: minimum matching score, minimum number of matches, matching polarity, recognition angle range, and scale range type. At the same time, you can also set the color and font size of the recognition result characters in the result display.

[0078] Position correction: This item is mainly used to set the correction position determination method and position orientation. The methods include point and coordinates, and the position orientation can be adjusted from angle, X-direction scale, Y-direction scale and benchmark.

[0079] Identification of screen indication units: According to the display of the screen indications of different models of α and β surface contamination meters, different unit forms and the presence or absence of "×1000" values ​​are considered. The three-aspect recognition logic process of the screen indications of α and β surface contamination meters is considered in parallel. First, for most cases that do not contain "×1000", first select the numerical training file specifically for the target detection model file generation, set character filtering and minimum confidence, and set the color and font size of the recognition result characters. After the numerical value is recognized, it will be converted through the script 1 program to convert the recognized characters into floating point types. Then proceed with the unit recognition process, also according to the settings of the indication recognition, but no script program writing is required. If the screen indication contains "×1000", the "×1000" symbol processing is also required. First, the script 2 program will judge the recognized characters. If it meets the requirements, it will be recognized from the "×1000" recognition area, and then the characters will be converted into floating point types through the script 2 program.

[0080] Result calculation formatting: The numerical identification symbol, unit identification symbol and "×1000" symbol are combined and expressed in characters.

[0081] Send data: Send the above identification result code to the specified location of the host computer to complete the identification of the α and β surface contamination meter automatic calibration device.

[0082] like Figure 1 The following is a schematic diagram of the screen identification area of ​​α and β surface contamination meters. For α and β surface contamination meters of different brands and models, the screen indication display area includes a digital display area, a multiplier "×1000" display area, and a unit display area. The identification results of these three areas constitute the indication.

[0083] like Figure 2 As shown, the target detection model training flow chart of the α and β surface contamination meter screen indication recognition method includes the following steps S201-S206.

[0084] Step S201, in this step, the user needs to initialize a new data set, which will be used for subsequent image recognition model training. Creating a data set includes defining the name and description of the data set and determining the storage location of the data set. Use a camera to take photos of the α and β surface contamination meter screen indications so that the sample photos cover the background indications, α plane source indications, and β plane source indications of the calibration. The amount of data reaches about 100.

[0085] Step S202, this step involves importing image files into the created dataset. The user can select a single or multiple image files, which will be used to train the model to recognize specific characters or objects.

[0086] Step S203, after the image is imported, the user needs to annotate the target characters or objects in each image. This includes marking the region of interest (ROI) on the image and assigning labels to these regions, such as numbers, letters or other symbols. The numbers, multipliers "×1000" and unit areas in the image are mainly selected, and then compared with the initial algorithm recognition results, the recognition errors are manually corrected, and then the recognition algorithm is improved.

[0087] In step S204, based on the labeled image data, the user will create a model for image recognition, which may involve selecting an appropriate machine learning or deep learning algorithm and defining the architecture and parameters of the model.

[0088] Step S205, after the model is created, the user will use the imported and annotated data set to train the model. During the training process, the model will learn to recognize the annotated objects in the image. This step may require adjusting the model parameters to optimize the recognition accuracy.

[0089] Step S206: export the trained model as a file for subsequent deployment or further optimization. The exported file can be used in the recognition logic flow configuration to implement the image recognition function.

[0090] like Figure 3 As shown, the method for identifying the screen indications of α and β surface contamination meters includes the following steps S301-S307.

[0091] Step S301, the image source module realizes efficient management and processing of image sequences through core functions such as SN initial value setting, output Mono8 function, splicing enable, trigger clearing mechanism and character trigger filtering. The user can set the initial sequence number for the first image, and the system automatically increments it for easy tracking. At the same time, the system supports the synchronous output of color images and grayscale images, as well as splicing images in the Y direction and setting the cropping area. The trigger clearing mechanism ensures the accuracy of the image cache, and the character trigger filtering allows the start of the image processing process to be controlled through external communication, providing a flexible and efficient solution for application scenarios that require precise control of image processing.

[0092] Step S302, the high-precision feature matching module realizes accurate recognition and matching of image features through a series of fine template configuration parameters. Template configuration includes the setting of matching points, which is used to create a position reference. Users can click "Select Model Matching Center" and set the matching center point by themselves. The scale mode allows switching between automatic and manual to adapt to different recognition needs. The speed scale parameter controls the extraction sparsity of feature scale and edge points, affecting the matching speed. The larger the value, the larger the feature scale, the sparser the edge points, and the faster the speed, but the default recommendation is between 1 and 20. The feature scale parameter determines the fineness of the feature particles. It takes an integer value and is not greater than the coarse scale. The finest value is 1. After adjustment, it may affect the contour points. The threshold mode can also be switched between automatic and manual, and the contrast threshold affects the screening of feature points. The larger the value, the more feature points are eliminated. The default range of the contrast threshold is 1 to 255. The flexible setting of these parameters ensures high precision and high efficiency of feature matching, which is suitable for various image recognition tasks.

[0093] Step S303, the correction tool is used to assist in positioning and accurately correct the offset in the target motion. The tool uses the template matching results to establish a position offset benchmark through matching points and matching frame angles. Subsequently, based on the feature matching results, the relative position offset between the running point and the reference point is calculated to achieve the coordinate rotation and offset adjustment of the ROI (region of interest) detection frame. This process ensures that the ROI area can adapt to changes in image angles and pixels. During implementation, the reference point and the reference frame are determined by feature matching when creating the reference, while the running point and the running frame are determined when the target image features are matched. By comparing the reference point with the running point, the pixel offset of the image can be identified; by comparing the reference frame with the running frame, the angle offset can be identified. This comparison and adjustment mechanism enables the ROI area to accurately track changes in the image, thereby improving the accuracy and reliability of positioning.

[0094] In step S304, the image recognition system provides a set of basic parameters and operating parameters to achieve accurate character recognition and processing. The basic parameters require the user to clearly select the ROI area where the target character is located in the settings to ensure the accuracy of recognition, especially when the position of the graphic changes. It is recommended to use it with the character positioning function. The setting of the operating parameters allows the user to specify the model file path. The system provides a default model and supports loading user-defined trained model files. The solution save model function allows model data to be saved in the solution or process file, which is convenient for cross-device use without repeatedly entering the model file path. The character filtering function allows the user to set character filtering conditions for each ROI recognition box separately. The user can customize the type and number of recognized characters as needed, including numbers, uppercase and lowercase letters, special characters, spaces, etc., and can even define characters that are easily misread through the "Custom" option, but these characters must exist in the character library. The minimum confidence parameter is used to set the minimum score requirement for the positioning box to ensure the reliability of the recognition result. Through the setting of these parameters, the system can flexibly adapt to different recognition needs and improve the accuracy and efficiency of recognition. The recognized image is subjected to image processing and numerical recognition, and the screen indication is output; this includes denoising and enhancing the recognized image, extracting the numerical value of the number, the multiplier "×1000" and the unit, converting the extracted numerical value into the screen indication and outputting it.

[0095] Step S305, the formatting tool is used to integrate and convert the data into a string output so as to organize the data before communication output. The user can directly enter the character format in the input box of the formatting tool, or link the result output of the previous module. The tool provides basic parameter settings, such as input terminator, delimiter and array delimiter, as well as the result display area. Through these settings, the user can accurately control the formatted output of the data to ensure that the data is transmitted and displayed in the desired format. This flexibility enables the formatting tool to play a key role in data processing and communication output, and improves the efficiency and accuracy of data management.

[0096] Step S306, the data transmission module is designed to efficiently and accurately send the formatted recognition results to the system host computer. This module does not output data separately, but integrates the data into the data queue to ensure unified management and orderly transmission of data. In this way, communication errors caused by chaotic data output can be avoided and the reliability of data transmission can be improved. In addition, the module supports condition detection before data transmission, which further ensures the accuracy and integrity of the data. This design makes the data transmission process more stable, reduces the possibility of errors, and thus improves the operating efficiency and performance of the entire system.

[0097] Step S307, the image storage module provides a comprehensive set of parameter configuration options to achieve flexible image storage management. Users can select image sources according to their needs, including images directly from image sources or processed by other modules. The image storage function is turned off by default. When it is turned on, users can configure the storage parameters in detail and save them at the same time when outputting images. The trigger save function allows users to automatically save images under specific conditions, such as binding condition detection results, which can be set to save all or only when OK or NG. The debug save function saves images when specific conditions are met, which is convenient for problem tracking and performance optimization. The generate directory function enables users to automatically create storage folders based on dates, which is convenient for image organization and management. The synchronized storage option ensures the stability and consistency of the image storage process. The save rendering / original image function allows users to choose to save processed images or original images, which increases the flexibility of storage. Users also need to set the storage path. Note that the path length is limited to 200 characters. Exceeding the limit will result in a save failure. Finally, users can customize the naming rules of renderings / original images to meet different naming requirements. The combination of these functions provides users with a powerful image storage solution that is suitable for a variety of application scenarios.

[0098] like Figure 4 As shown, the system framework diagram of the method for identifying the screen indication of the α and β surface contamination meters is shown. The method for identifying the screen indication of the surface contamination meter of the present invention realizes an automated and efficient identification process through a series of modular steps. These steps include data set acquisition, detection model establishment, frame position identification, center point correction and instrument image recognition, and each module has its specific function and role.

[0099] Dataset acquisition module 10: This module is responsible for collecting image data of the surface contamination meter screen. By using a camera to shoot the contamination meter screen under different conditions, images containing various reading sizes, brightness and numerical symbols are obtained. These images will be used for subsequent model training to ensure that the model can adapt to different practical application scenarios.

[0100] Object detection model building module 20: In this module, the object detection model is built using the acquired data set. An algorithm suitable for character recognition, such as a convolutional neural network in deep learning, is selected to train the model. During the training process, the model learns to recognize numbers, unit symbols, scientific notation symbols, etc. in the image.

[0101] High-precision matching module 30: This module uses the trained model to identify key information in the screen image, such as the border position of numbers and units. The model outputs the precise position information of these elements, providing a basis for subsequent image processing.

[0102] Position correction module 40: After identifying the border position, this module further processes the recognition result, including the correction of the center point. By adjusting the position and orientation of the recognition area, the accuracy of recognition is ensured, even when the screen display is incomplete or deviated.

[0103] Instrument image recognition module 50: Finally, this module formats the recognized value, unit and "×1000" symbol, generates a character code, and sends it to the specified location of the host computer. This step completes the automatic recognition of the surface contamination meter screen indication, achieving high accuracy and convenience of the automated verification device.

[0104] Therefore, this method improves the recognition efficiency and accuracy of the screen readings of the α and β surface contamination meters through automated image recognition technology, and reduces human errors.

Claims

1. A method and system for identifying the screen readings of α and β surface contamination meters, characterized in that: The following steps are involved: Obtain the screen images of α and β surface contamination instruments as sample data sets and annotate them; Build and train an object detection model for detecting numbers, multipliers "×1000", and units in images; Using the target detection model to identify the screen image, and obtain the border position information of the numbers and units; The recognized image is processed and numerically recognized, and the displayed value is output on the screen.

2. A method for identifying screen indications of α and β surface contamination meters according to claim 1, wherein the target detection model is a convolutional neural network model based on deep learning.

3. A method for identifying the screen readings of an α and β surface contamination meter according to claim 1, wherein the image processing and numerical value identification steps further include: De-noise and enhance the recognized image; Extract the value of the number, the multiplier "×1000" and the unit; Convert the extracted numerical value into screen display and output it.

4. A method for identifying the screen indications of an α or β surface contamination meter according to claim 1, wherein the annotations of the sample data set include the position, value, multiplier "×1000" and color or shape of the unit.

5. A method for identifying screen readings of α and β surface contamination meters according to claim 1, wherein the training of the target detection model comprises the following steps: Input the labeled sample dataset into the model as training data; Calculate the loss function of the model; Update the model's weights via backpropagation.

6. An α, β surface contamination meter screen indication recognition system, used in any one of claims 1-5. An α, β surface contamination meter screen indication recognition method and system, characterized in that: include: Data acquisition module, used to obtain the screen images of α and β surface contamination instruments; Object detection models to detect numbers, multipliers "×1000", and units in images; An image processing module is used to perform image processing and numerical recognition on the recognized image; Output module, used to output screen indications. Storage module.

7. An α, β surface contamination meter screen indication recognition system, characterized in that: The data acquisition module includes the settings of the number of cameras and image buffers, disconnection and reconnection time, image width and height, pixel format, frame rate, exposure mode and exposure time.

8. An α, β surface contamination meter screen indication recognition system, characterized in that: The target detection model is lightweight to adapt to different computing environments.

9. An α, β surface contamination meter screen indication recognition system, characterized in that: The storage module includes a computer-readable storage medium, and the storage medium stores a computer-executable program.

10. An α, β surface contamination meter screen indication recognition system, characterized in that: The computer executable program further includes a logic flow for implementing image source setting, high-precision matching, position correction, screen display symbol recognition, formatting processing and data sending.

Images