Hysteresis loop data extraction method, device and prediction system based on image recognition

Abstract

The present invention discloses a method, device and prediction system for extracting hysteresis loop data based on image recognition, and relates to the fields of material science and electronic engineering technology. The present invention comprises: receiving hysteresis loop image data, and preprocessing the image data; using the DB algorithm model that has been pre-trained on the paddleOCR platform to locate the coordinate axis text in the hysteresis loop image; using the CRNN algorithm model that has been pre-trained on the paddleOCR platform to identify the located coordinate axis text content, and extract the coordinate axis data; binarizing the image data, and determining the border position information of the hysteresis loop by calculating the sum of the image rows and columns. The present invention can directly extract key data from a hysteresis loop image with noise, and use these data to accurately predict the hysteresis loop, significantly reducing the dependence on expensive experimental equipment, and improving the efficiency and accuracy of data processing.

Description

Technical Field

[0001] The present invention relates to the field of material science and electronic engineering technology, and in particular to a hysteresis loop data extraction method, device and prediction system based on image recognition. Background Art

[0002] In recent years, with the rapid development of image processing technology, especially driven by computer vision and machine learning, image recognition technology has become a research hotspot. As a key tool for data analysis and material performance research, image recognition technology is an important basis for realizing the automation of magnetic material research and one of the key factors to improve efficiency and accuracy.

[0003] In many research fields, especially in materials science, the performance analysis of magnetic materials often relies on precise experimental equipment, such as instruments for generating hysteresis loops. These equipment are not only expensive, but also have strict requirements on experimental conditions. Traditional hysteresis performance analysis methods rely on expensive measuring instruments in the laboratory to obtain hysteresis loop data through precise voltage and current parameter measurements. This method is not only expensive in terms of experimental costs, but also difficult to popularize under limited conditions. For this reason, we propose a hysteresis loop data extraction method, device and prediction system based on image recognition. . Summary of the invention

[0004] The purpose of the present invention is to provide a hysteresis loop data extraction method, device and prediction system based on image recognition, which can directly extract key data from a hysteresis loop image with noise, and use these data to perform accurate hysteresis loop prediction, significantly reducing dependence on expensive experimental equipment and improving data processing efficiency and accuracy.

[0005] According to a first aspect of the present invention, in order to achieve the above-mentioned purpose, the present invention provides the following technical solution: a method for extracting hysteresis loop data based on image recognition, specifically comprising the following steps:

[0006] Receiving hysteresis loop image data and preprocessing the image data;

[0007] The pre-trained DB algorithm model is used to locate the coordinate axis text in the hysteresis loop image;

[0008] The pre-trained CRNN algorithm model is used to identify the text content of the located coordinate axis and extract the coordinate axis data;

[0009] Binarization is performed on the image data, and the border position information of the hysteresis loop is determined by calculating the sum of the image rows and columns;

[0010] By combining the coordinate axis data and frame position information, the complete hysteresis loop data is extracted.

[0011] Furthermore, the complete hysteresis loop data includes upper branch data and lower branch data.

[0012] Furthermore, the pre-trained DB algorithm model is used to locate the coordinate axis text in the hysteresis loop image. The specific method is as follows:

[0013] (1) Install the ch_PP-OCRv4_det text detection model from the paddleOCR platform and deploy it to the python environment;

[0014] (2) Read the image to be recognized in Python and call the ch_PP-OCRv4_det model for text detection.

[0015] Furthermore, the pre-trained CRNN algorithm model is used to recognize the text content of the located coordinate axis and extract the coordinate axis data. The specific method is as follows:

[0016] (1) Install the ch_PP-OCRv4_rec text recognition model from the paddleOCR platform and deploy it to the python environment;

[0017] (2) Read the image to be recognized in Python and call the ch_PP-OCRv4_rec model for text recognition.

[0018] Furthermore, the DB algorithm model and CRNN algorithm model correspond to the ch_PP-OCRv4_det and ch_PP-OCRv4_det models respectively. The paddleOCR platform has completed the corresponding pre-training for the objects it recognizes and can be used directly. The specific method of pre-training is as follows:

[0019] (1) Use several hysteresis loop images to pre-train the DB open source model provided by the paddleOCR platform;

[0020] (2) Several hysteresis loop images are used to pre-train the resent34-based CRNN model provided by the paddleOCR platform.

[0021] Furthermore, the image data is binarized, and the border position information of the hysteresis loop is determined by calculating the sum of the image rows and columns, as follows:

[0022] (1) After binarizing the image data, a grayscale image is obtained, and the grid is removed from the gray image;

[0023] (2) Sum each row and column in the image. The row and column with the largest sum is the boundary of the image.

[0024] According to a second aspect of the present invention, the present invention provides a hysteresis loop data extraction device based on image recognition, comprising:

[0025] A receiving module, used for receiving hysteresis loop image data and preprocessing the image data;

[0026] The first image recognition module is used to locate the coordinate axis text in the hysteresis loop image using a pre-trained DB algorithm model;

[0027] The second image recognition module is used to recognize the text content of the located coordinate axis using the pre-trained CRNN algorithm model to extract the coordinate axis data;

[0028] An image processing module is used to perform binarization processing on the image data and determine the border position information of the hysteresis loop by calculating the sum of the image rows and columns;

[0029] The extraction output module is used to combine the coordinate axis data and the frame position information to extract the complete hysteresis loop data.

[0030] According to a third aspect of the present invention, the present invention provides a prediction system, which uses the hysteresis loop data extracted by the above-mentioned hysteresis loop data extraction method based on image recognition as input to perform prediction analysis.

[0031] Furthermore, the system includes an improved preisach model, specifically:

[0032] Bu * (H) = 2 × Intg (S6) - Intg (S1) + B rev (H)

[0033] Here Intg(S*) represents the double integral value on the triangular integral area of ​​the Preisach model change process. rev (H) represents the reversible component of magnetic induction intensity. Bu * (H) represents the predicted magnetic induction intensity value.

[0034] According to a fourth aspect of the present invention, the present invention provides a prediction method, using the above-mentioned prediction system, the specific steps are as follows:

[0035] (1) According to the hysteresis loop data, the upper and lower branches are fitted using the tanh function by the least squares method. The fitting formula is:

[0036] f(x)=F×tanh(Ax+B)+D

[0037] F, A, B, and D are the degree of freedom parameters for fitting the upper and lower branches of the hysteresis upper loop.

[0038] (2) The f(x) obtained after fitting is put into the improved Prisach model, and the improved Prisach model can be used for predictive analysis.

[0039] The present invention has at least the following beneficial effects:

[0040] (1) Compared with the traditional method that relies on expensive experimental equipment, the present invention significantly reduces the dependence on experimental equipment and reduces the research cost; by directly extracting key data from the hysteresis loop image with noise, the efficiency and accuracy of data processing are improved, so that high-quality magnetic material research can still be carried out in an environment with limited resources;

[0041] (2) The present invention uses the improved Preisach model for predictive analysis, which can accurately predict the shape and characteristics of the hysteresis loop under different external field strength conditions, providing a new tool for the performance evaluation and optimization of magnetic materials. It is particularly suitable for scenarios where experimental data need to be analyzed but only images are available. It greatly improves the accessibility and flexibility of hysteresis loop analysis and helps promote the progress of scientific research and industrial applications.

[0042] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a flow chart of the data extraction method and prediction method of the present invention;

[0044] Figure 2 Schematic diagram of the integral region of the calculation formula Intg(S*) of the present invention;

[0045] Figure 3 It is a schematic diagram of the hysteresis loop measured in the first embodiment of the present invention after the coordinate axis data is extracted;

[0046] Figure 4 The image of the hysteresis loop image of the present invention is a binarized image;

[0047] Figure 5 The present invention uses the fitting result of tanh function;

[0048] Figure 6 Magnetic hysteresis loops at different field strengths measured in the laboratory of the present invention;

[0049] Figure 7 The present invention simulates and compares the results. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present disclosure to clearly and completely describe the technical solutions in the embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of the present disclosure, rather than all the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present disclosure.

[0051] Embodiment 1:

[0052] See also Figure 1-7 The present invention provides a technical solution: a method for extracting hysteresis loop data based on image recognition, comprising the following steps:

[0053] S1. Receive hysteresis loop image data and pre-process the image data;

[0054] S2. Use the pre-trained DB algorithm model to locate the coordinate axis text in the hysteresis loop image, as follows:

[0055] (S2.1) Install the ch_PP-OCRv4_det text detection model from the paddleOCR platform and deploy it to the python environment;

[0056] (S2.2) Read the image to be recognized in Python and call the ch_PP-OCRv4_det model for text detection;

[0057] like Figure 3 The model shown will automatically position the text content of Chinese and English characters in the diagram;

[0058] It should be noted that the DB algorithm model is obtained by directly downloading the software package through the paddleOCR platform. The technical solution of this embodiment uses segmentation-based scene text detection, that is, converting the probability map (heat map) generated by the segmentation method into a bounding box and a text area, which will include a binarization post-processing process. This embodiment adopts the DB algorithm model. The traditional network structure uses a fixed threshold to perform binarization detection on the segmented heat map, while the DB algorithm model embeds the binarization operation into the segmentation network for combined optimization, and generates a threshold map corresponding to the heat map. The final binarization operation is generated by combining the two.

[0059] Furthermore, if the model has a higher accuracy requirement, the model can be retrained. Use a large-scale data set with numbers, English, and characters, including 50 million English precision-labeled data sets, 8,000 English precision-labeled data sets, and 8,000 character precision-labeled data sets (marking the position and text content of all text boxes), to pre-train the DB algorithm model, obtain a pre-trained model, and then use multiple hysteresis loop images for secondary training to obtain a pre-trained DB algorithm model.

[0060] S3. Use the pre-trained CRNN algorithm model to identify the text content of the located coordinate axis and extract the coordinate axis data, as follows:

[0061] (S3.1) Install the ch_PP-OCRv4_rec text recognition model from the paddleOCR platform and deploy it to the python environment;

[0062] (S3.2) Read the image to be recognized in Python and call the ch_PP-OCRv4_rec model for text recognition;

[0063] like Figure 3 The content at the bottom shows the text content in the box recognized by the model.

[0064] It should be noted that the technical solution of this embodiment adopts CRNN as a text sequence recognition model, and does not need to perform character segmentation on the text, thus avoiding the problem of cutting errors caused by a large amount of manual parameter adjustment and interference from the background of printed fonts. The CRNN network structure is clear. First, CNN extracts image features, and then LSTM further extracts sequence features from image convolution features. Finally, CTC loss is introduced to solve the problem of character alignment failure during training, providing an end2end text recognition algorithm.

[0065] Furthermore, if the model has a higher accuracy requirement, the model can be retrained. Most of the training data of the CRNN pre-trained model are English, numbers, characters, etc. Combined with the performance analysis scenario of magnetic materials, 3,000 hysteresis loop images are manually annotated with letters, characters, and numbers. In addition, 1,000 Chinese text data are added to pre-train the resnet34 CRNN algorithm model to improve the recognition accuracy of hysteresis loop images.

[0066] S4. Binarize the image data and determine the border position information of the hysteresis loop by calculating the sum of the image rows and columns, as follows:

[0067] (S4.1) binarizing the image data to obtain a grayscale image, and removing the grid from the grayscale image;

[0068] (S4.2) summing each row and column in the image, and the row and column with the largest summation result are the boundaries of the image;

[0069] S5. Combine the coordinate axis data and the frame position information to extract the complete hysteresis loop data.

[0070] Next, the technical solution of the present invention is described in detail with reference to specific embodiments:

[0071] like Figure 3 As shown, the DB algorithm is used to locate the content of the image containing characters as shown in the red box. The order of positioning is from top to bottom and from left to right. The positioning result is the coordinates of the four corners of each red box. Then the CRNN algorithm is used to identify the content in the red box as shown in the lower part of the figure. It should be noted that the DB and CRNN inspection models here are both pre-trained models. The software package can be downloaded directly from the paddleOCR platform and can be used directly. The specific method of pre-training is as follows:

[0072] (S5.1) Using several hysteresis loop images to pre-train the DB open source model provided by the paddleOCR platform;

[0073] (S5.2) Use several hysteresis loop images to pre-train the resent34-based CRNN model provided by the paddleOCR platform.

[0074] according to Figure 3 The text data in the image obtained by the regular expression method can be used to distinguish letters and numbers, and then the horizontal and vertical coordinate data can be distinguished by the difference in the order of magnitude of the magnetic induction intensity and the magnetic field intensity, thereby completing the recognition of the coordinate axis data.

[0075] After binarizing the experimental image, a grayscale image will be obtained. Each color block in the image is represented by a value 0-255 (0 represents black and 255 represents white). All values ​​greater than 150 are set to 0, and the grid lines can be removed to obtain Figure 4 .

[0076] right Figure 4 By summing up each row and column, it is easy to determine that the row and column with the largest sum is the boundary of the image. Once the boundary position is determined, the image data can be extracted in combination with the previously identified coordinate axis data. For example, starting from the bottom boundary of the y-axis, the first non-zero data is found from bottom to top, which is the data point of the hysteresis loop. Then, according to the corresponding proportion of the coordinates of the position, the data point is assigned a value through the identified coordinate axis data. In this way, all the lower branch data of the hysteresis loop can be found, and the same applies to the upper branch data.

[0077] Embodiment 2:

[0078] The present invention provides a hysteresis loop data extraction device based on image recognition, comprising:

[0079] A receiving module, used for receiving hysteresis loop image data;

[0080] The first image recognition module is used to locate the coordinate axis text in the hysteresis loop image using a pre-trained DB algorithm model;

[0081] The second image recognition module is used to recognize the text content of the located coordinate axis using the pre-trained CRNN algorithm model to extract the coordinate axis data;

[0082] An image processing module is used to perform binarization processing on the image data and determine the border position information of the hysteresis loop by calculating the sum of the image rows and columns;

[0083] The extraction output module is used to combine the coordinate axis data and the frame position information to extract the complete hysteresis loop data.

[0084] Specifically, the above-mentioned receiving module, the first image recognition module, the second image recognition module, the image processing module and the extraction and output module can be embedded in a computer processing system. The computer calls the above-mentioned modules to complete the task of equivalent modeling according to the above-mentioned hysteresis loop data extraction method based on image recognition; the above-mentioned receiving module, the first image recognition module, the second image recognition module, the image processing module and the extraction and output module can perform operations according to the specific steps given in the above-mentioned hysteresis loop data extraction method based on image recognition.

[0085] It should be noted that it should be understood that the division of the various modules of the above system is only the division of logical functions. In actual implementation, they can be fully or partially integrated into one physical entity, or they can be physically separated, and these modules can all be implemented in the form of software calling through processing elements; they can also be all implemented in the form of hardware; some modules can also be implemented in the form of software calling through processing elements, and some modules can be implemented in the form of hardware. For example, the receiving module can be a separately established processing element, or it can be integrated in a certain chip of the above device. In addition, it can also be stored in the memory of the above device in the form of program code, and a certain processing element of the above device calls and executes the functions of the above signal processing module. The implementation of other modules is similar. In addition, these modules can be fully or partially integrated together, or they can be implemented independently. The processing element described here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each module above can be completed by the hardware integrated logic circuit in the processor element or the instructions in the form of software.

[0086] For example, the above modules may be one or more integrated circuits configured to implement the above methods, such as one or more application specific integrated circuits (ASIC), or one or more digital singnal processors (DSP), or one or more field programmable gate arrays (FPGA). For another example, when a module above is implemented in the form of a processing element scheduling program code, the processing element may be a general-purpose processor, such as a central processing unit (CPU) or other processor that can call program code. For another example, these modules may be integrated together and implemented in the form of a system-on-a-chip (SOC).

[0087] Embodiment three:

[0088] The present invention provides a prediction method, using the above prediction system, the specific steps are as follows:

[0089] like Figure 5 As shown, according to the identified hysteresis loop data, the upper and lower branches are fitted with the tanh function through the least squares method. The fitting formula here is:

[0090] f(x)=F×tanh(Ax+B)+D (1)

[0091] The f(x) obtained after fitting can be put into the improved Prisach model for prediction.

[0092] Embodiment 4:

[0093] The present invention provides a prediction system, which comprises an improved Preisach model.

[0094] The traditional Preisach model prediction is based on the formula (here taking the rising branch of the hysteresis loop as an example):

[0095] Bu * (H)=2×Intg(S6)-Intg(S1) (2)

[0096] Intg(S1)=0.5×(F(-Hs)-F(Hs)) 2 (3)

[0097] Intg(S6)=0.5×(f(Hs)+f(-H))+F(Hs)×F(-H) (4)

[0098]

[0099] Here Intg(S*) represents the double integral value on the triangular integral area of ​​the change process of the Preisach model. The specific integral area is as follows Figure 2 As shown. F(H) is the intermediate formula connecting Intg(S*) and f(H), and Brev(H) represents the reversible component of magnetic induction intensity.

[0100] Bu*(H) is the explicit expression of the rising branch under a specific Hs (Hs represents the maximum magnetic field intensity), and f(H) is the fitting formula obtained by fitting the limiting hysteresis loop of the material. However, a significant error will occur as the predicted magnetic field intensity decreases. The main reason is that the segmentation of the reversible component of the hysteresis loop is not considered.

[0101] Therefore, the improved Preisach model proposed in this embodiment can take into account the reversible component of the hysteresis loop. Due to the properties of the Preisach model, Figure 5 The differential permeability under two maximum magnetic field intensities is proposed in Equation (7) and the two unknown parameters in Equation (6) can be solved. By optimizing f(x) in Equation (8), the improved Preisach model in Equation (9) can be obtained:

[0102] B rev (x) = K1 × arctan (K2 × x) (6)

[0103]

[0104] f(x)=F×tanh(Ax+B)+D-K1×arctan(K2×x) (8)

[0105] Bu * (H) = 2 × Intg (S6) - Intg (S1) + B rev (H) (9)

[0106] Formula (6) uses the arctan function to represent the reversible component of the hysteresis loop. K1 and K2 are degree of freedom parameters, which can be expressed by Figure 6 The two position parameters shown are calculated by analytical equations. rev is the differential permeability of the reversible component corresponding to formula (6). Formula (8) is the new fitting function f(x), which is to subtract the reversible component from formula (1) and add the reversible component to formula (9). Since the integral area change of the Preisach model is only related to the irreversible component, this operation will undoubtedly greatly improve the accuracy of fitting and prediction.

[0107] Finally, a smaller magnetic field strength is selected to compare the experimental data, the traditional model and the invented model. Figure 7 As shown, it can be seen that the new model has higher accuracy than the old model.

[0108] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the term "include", "comprise" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or equipment. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances. When an element is referred to as "assembled to", "installed to", "fixed to" or "set to" another element, it can be directly on another element or there can also be a centered element. When an element is considered to be "connected" to another element, it can be directly connected to another element or there may be a centered element at the same time. The terms "vertical", "horizontal", "up", "down", "left", "right" and similar expressions used herein are only for illustrative purposes and do not represent the only implementation method.

[0109] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

[0110] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present disclosure. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

Claims

1. A method for extracting hysteresis loop data based on image recognition, characterized in that: The specific steps include: Receiving hysteresis loop image data and processing the image data; The pre-trained DB algorithm model is used to locate the coordinate axis text in the hysteresis loop image; The pre-trained CRNN algorithm model is used to identify the text content of the located coordinate axis and extract the coordinate axis data; Binarization is performed on the image data, and the border position information of the hysteresis loop is determined by calculating the sum of the image rows and columns; By combining the coordinate axis data and frame position information, the complete hysteresis loop data is extracted.

2. The method for extracting hysteresis loop data based on image recognition according to claim 1, characterized in that: The hysteresis loop data includes upper branch data and lower branch data.

3. The method for extracting hysteresis loop data based on image recognition according to claim 2, characterized in that: The pre-trained DB algorithm model is used to locate the coordinate axis text in the hysteresis loop image. The specific method is as follows: (1) Install the ch_PP-OCRv4_det text detection model from the paddleOCR platform and deploy it to the python environment; (2) Read the image to be recognized in Python and call the ch_PP-OCRv4_det model for text detection.

4. The method for extracting hysteresis loop data based on image recognition according to claim 3, characterized in that: The pre-trained CRNN algorithm model is used to identify the text content of the located coordinate axis and extract the coordinate axis data. The specific method is as follows: (1) Install the ch_PP-OCRv4_rec text recognition model from the paddleOCR platform and deploy it to the python environment; (2) Read the image to be recognized in Python and call the ch_PP-OCRv4_rec model for text recognition.

5. The method for extracting hysteresis loop data based on image recognition according to claim 4, characterized in that: The DB algorithm model and CRNN algorithm model correspond to the ch_PP-OCRv4_det and ch_PP-OCRv4_rec models respectively. The paddleOCR platform has completed the corresponding pre-training for the objects it recognizes and can be used directly. The specific method of pre-training is as follows: (1) Use several hysteresis loop images to pre-train the DB open source model provided by the paddleOCR platform; (2) Several hysteresis loop images are used to pre-train the resent34-based CRNN model provided by the paddleOCR platform.

6. The method for extracting hysteresis loop data based on image recognition according to claim 5, characterized in that: The image data is binarized, and the border position information of the hysteresis loop is determined by calculating the sum of the image rows and columns, as follows: (1) After binarizing the image data, a grayscale image is obtained, and the grid is removed from the grayscale image; (2) Sum each row and column in the image. The row and column with the largest sum is the boundary of the image.

7. A hysteresis loop data extraction device based on image recognition, characterized in that: include: A receiving module, used for receiving hysteresis loop image data; The first image recognition module is used to locate the coordinate axis text in the hysteresis loop image using a pre-trained DB algorithm model; The second image recognition module is used to recognize the text content of the located coordinate axis using the pre-trained CRNN algorithm model to extract the coordinate axis data; An image processing module is used to perform binarization processing on the image data and determine the border position information of the hysteresis loop by calculating the sum of the image rows and columns; The extraction output module is used to combine the coordinate axis data and the frame position information to extract the complete hysteresis loop data.

8. A prediction system, characterized in that: It includes the hysteresis loop data extraction device based on image recognition as described in claim 7.

9. The prediction system according to claim 8, characterized in that The system also includes an improved Preisach model, specifically: Bu * (H)=2×Intg(S6)-Intg(S1)+B rev (H) Here Intg(S*) represents the double integral value on the triangular integral area of ​​the change process of the Preisach model, B rev (H) represents the reversible component of magnetic induction intensity, Bu * (H) represents the predicted magnetic induction intensity value.

10. A prediction method, using the prediction system as claimed in claim 8 or 9, characterized in that: The specific steps are as follows: (1) According to the hysteresis loop data, the upper and lower branches are fitted using the tanh function by the least squares method. The fitting formula is: f(x)=F×tanh(Ax+B)+D F, A, B, and D are the degree of freedom parameters of the upper and lower branches of the fitted hysteresis loop; (2) The f(x) obtained after fitting is put into the improved Prisach model, and the improved Prisach model can be used for predictive analysis.

Images