Material data curve identification and fitting method based on deep learning

Abstract

The invention provides a material data curve identification and fitting method based on deep learning, and relates to the technical field of material data extraction, and the method comprises the steps: obtaining an image file containing a plurality of material data curves; separating the material data curve from the image background by adopting an image segmentation technology, and extracting feature points of the material data curve through a convolutional neural network; and taking the feature points as control points, performing parametric fitting on the curve through polynomial or spline interpolation, and generating a smooth continuous curve. The method further comprises the following steps: detecting the position and boundary of a coordinate axis in the image file through an image processing technology; the OCR technology is used to identify the scale label on the coordinate axis, and the deep learning model is combined to analyze the numerical value and unit of the scale. According to the method, through innovative technologies such as multi-modal feature fusion, self-adaptive preprocessing and intelligent data labeling, precise recognition of complex curves, multi-curve separation and association, intelligent recognition of non-standard coordinate axes and efficient real-time processing are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of material data extraction, and particularly to a method for identifying and fitting material data curves based on deep learning. Background Art

[0002] As the core driving force of the modern industrial system and frontier technology innovation, materials science continues to play a fundamental supporting role in strategic industries such as aerospace, new energy, and high-end equipment manufacturing. In recent years, with the breakthrough development of new research paradigms such as the Materials Genome Initiative and high-throughput experiments, the explosive growth of advanced materials represented by high-performance alloys and intelligent composite materials has led to an exponential increase in the complexity of their performance characterization parameters and service behavior data. On the one hand, this has promoted the expansion of material application scenarios, and on the other hand, it has also exposed the serious lag of the traditional data management system - especially in the field of dynamic curve data related to the performance evolution of materials throughout their life cycle, and the existing technical means are difficult to meet the deep mining requirements of data value for industrial upgrading.

[0003] As the core carrier of material performance research, dynamic curve data (such as stress-strain curves, fatigue life curves, phase transformation kinetics curves, etc.) carries key information on the correlation between the evolution of material microstructure and macroscopic properties. However, the full-process management of current curve data faces multi-dimensional challenges: in the data acquisition link, the lack of compatibility of multi-source heterogeneous devices leads to the lack of data standardization; in the analysis and application level, the feature extraction of massive unstructured data relies on manual experience, and the processing efficiency and repeatability need to be improved urgently; more critically, the lack of intelligent data fusion tools for cross-scale correlation analysis of material properties under complex working conditions directly restricts the shortening of the material R & D cycle and the process optimization process. According to industry research, nearly 78% of material research institutions invest more than 40% of their R & D resources in the data governance link, but the effective utilization rate of data assets is still less than 30%. Summary of the Invention

[0004] Based on the above industrial pain points, the present invention proposes a method for identifying and fitting material data curves based on deep learning, aiming to build an intelligent data engine for the entire life cycle of materials. By developing multi-modal data fusion algorithms, high-precision curve feature analysis models, and knowledge graph-based association reasoning methods, it realizes the standardized reconstruction and value multiplication of curve material performance data, providing core data infrastructure support for accelerating the industrialization process of new materials.

[0005] Specifically, the present invention proposes a method for identifying and fitting material data curves based on deep learning, including the following processes: Obtain an image file containing multiple material data curves, where the curves are used to show the performance changes of materials; The material data curve is separated from the image background by using image segmentation technology, and the feature points of the material data curve are extracted by a convolutional neural network; Taking the feature points as control points, the curve is parametrically fitted by polynomial or spline interpolation to generate a smooth continuous curve.

[0006] As a further description of the present invention, after obtaining an image file containing multiple material data curves, the method further includes: Converting the color image to a grayscale image and then performing binarization processing to highlight the curve features; Removing noise by Gaussian filtering and enhancing the curve edges using an edge detection algorithm; Automatically adjusting the contrast and brightness according to the image quality to optimize the overall visual effect of the image.

[0007] As a further description of the present invention, the extracting the feature points of the material data curve by a convolutional neural network specifically includes: Constructing a training set and a test set, both of which contain a large number of material data curves with labeled feature points; Training a convolutional neural network model with the training set so that the trained convolutional neural network model can extract the coordinate values of the feature points of the material data curve; evaluating the model performance of the convolutional neural network model with the test set; Inputting the material data curve to be extracted into the trained convolutional neural network model, and outputting the coordinate values of the feature points of the material data curve.

[0008] As a further description of the present invention, when labeling the feature points, semi-supervised learning is performed with a small amount of labeled data and a large amount of unlabeled data to automatically label the unlabeled data.

[0009] As a further description of the present invention, the method further includes: modeling the geometric and topological relationships of the curves through a graph neural network (GNN) structure, and realizing the efficient discrimination of multiple curves by using message passing and hierarchical feature learning.

[0010] As a further description of the present invention, after obtaining an image file containing multiple material data curves, the method further includes: Detecting the position and boundary of the coordinate axes in the image file by image processing technology; Using a deep learning model GNN to parse the scale values of the coordinate axes, using multi-task learning or staged processing to parse the units of the coordinate axes, combining text detection, numerical regression and semantic understanding technologies, through the collaborative modeling of object detection, OCR and semantic understanding, and combining multi-task learning with physical rule post-processing to parse the scale values and their units of the coordinate axes, adapting to different scales and layout scenarios.

[0011] Compared with the prior art, the present invention has the following beneficial technical effects: The present invention can effectively process complex shapes such as multi-peak curves, non-linear curves, and serrated curves. By adopting image segmentation and curve feature extraction, combined with image semantic understanding, the recognition accuracy and fitting precision of complex curves are significantly improved.

[0012] For the complex scenario of intertwined multi-curves, the present invention proposes a curve separation and association algorithm based on graph neural network (GNN), which can automatically identify and distinguish multiple curves in the image, and at the same time establish the logical association between the curves to avoid confusion and incorrect association, so as to achieve high-precision multi-curve extraction.

[0013] The present invention can automatically identify and process complex situations such as logarithmic coordinate axes and non-linear scales through semantic parsing of the coordinate axes and scales by a deep learning model, and at the same time supports custom coordinate axis annotation, greatly improving the adaptability and flexibility of the system.

[0014] Other features and advantages of the present technical solution will be described in the subsequent specification, and part of them will become obvious from the specification or be understood by implementing the present technical solution. The purpose and other advantages of the present technical solution can be achieved and obtained through the structures specifically pointed out in the written specification and the drawings.

[0015] The technical solution of the present technical solution will be further described in detail below through the drawings and embodiments. Brief Description of the Drawings

[0016] The drawings are used to provide a further understanding of the present technical solution, and constitute a part of the specification. They are used together with the embodiments of the present technical solution to explain the present technical solution, and do not constitute a limitation to the present technical solution. In the drawings: Figure 1 It is a flow chart of a method for identifying and fitting material data curves based on deep learning provided by the present invention. Detailed Description of the Embodiments

[0017] The following describes the preferred embodiments of the present technical solution with reference to the drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present technical solution, and are not used to limit the present technical solution.

[0018] As the core carrier of material property research, dynamic curve data (such as stress-strain curves, fatigue life curves, phase transformation kinetics curves, etc.) carries key information on the correlation between the evolution of material microstructure and macroscopic properties. However, there are the following technical problems in the extraction and recognition of curve data: Complex curve shapes: The curves in the document may have complex shapes, such as multi-peak curves, non-linear curves, serrated curves, etc. When dealing with these complex shapes in the prior art, problems such as inaccurate recognition or large fitting errors may occur.

[0019] Curve and background fusion: When the curve is mixed with other image elements (such as text, grid lines, noise, etc.), the difficulty of curve extraction will increase significantly. When separating the curve from the background in the prior art, key information may be misjudged or missed.

[0020] Multi-curve recognition: When there are multiple curves in the image and they are intertwined, it may be difficult for the prior art to accurately distinguish and extract each curve, and confusion or incorrect association is likely to occur.

[0021] Embodiment 1

[0022] The present invention proposes a method for identifying and fitting material data curves based on deep learning, which includes the following steps: Step 1: Obtain an image file containing multiple material data curves, where the curves are used to show the performance changes of the material.

[0023] Specifically, the image file can be, for example, a PDF file or a picture file.

[0024] Step 2: Preprocess the obtained image file. Specifically, the preprocessing includes the following processes: Grayscale and binarization: Convert the color image to a grayscale image and then perform binarization processing to highlight the curve features; Denosing and edge enhancement: Remove noise through Gaussian filtering and use an edge detection algorithm to enhance the curve edges; Adaptive enhancement: Automatically adjust the contrast and brightness according to the image quality to optimize the overall visual effect of the image.

[0025] The recognition of curves highly depends on the quality of the input image. If the image has problems such as noise, blurring, low resolution, etc., the recognition accuracy of the curves will drop significantly. Therefore, the present invention adopts the above image preprocessing process to sequentially perform grayscale and binarization, denosing and edge enhancement, and adaptive enhancement on the obtained image file, thereby effectively improving the image quality and enabling the subsequent curve recognition to proceed smoothly.

[0026] Step 3: Use image segmentation technology to separate the material data curves from the image background, and extract the feature points of the material data curves, such as extreme points, inflection points, etc., through a convolutional neural network (CNN).

[0027] Specifically, an adaptive curve segmentation technology can be used to separate the material data curves from the image background.

[0028] Specifically, the extraction of the feature points of the material data curve by the convolutional neural network includes: Construct a training set and a test set. Both the training set and the test set contain a large number of material data curves (such as stress-strain curves) with labeled feature points. The key feature points (such as peak points, inflection points, starting points, etc.) of each curve need to be pre-marked, usually in the form of coordinates.

[0029] Use the training set to train the convolutional neural network model so that the trained convolutional neural network model can extract the coordinate values of the feature points of the material data curve; use the test set to evaluate the performance of the convolutional neural network model.

[0030] Input the material data curve to be extracted into the trained convolutional neural network model, and output the coordinate values of the feature points of the material data curve.

[0031] Step 4: Use the feature points as control points, and perform parametric fitting on the curve through polynomial or spline interpolation to generate a smooth continuous curve.

[0032] Feature points are only discrete samples. Direct connection will result in a jagged broken line. In the present invention, parametric fitting of the curve through polynomial or spline interpolation can eliminate noise, fill in missing points, and generate a smooth curve that can be analyzed mathematically. After parameterization, it is convenient to calculate physical quantities such as derivatives, integrals, and curvatures, or for subsequent simulations and optimizations.

[0033] When labeling the feature points, perform semi-supervised learning through a small amount of labeled data and a large amount of unlabeled data to automatically label the unlabeled data.

[0034] Step 5: Model the geometric and topological relationships of the curves through the graph neural network (GNN) graph structure, and use message passing and hierarchical feature learning to achieve efficient differentiation of multiple curves, which is suitable for accurate analysis in dynamic and noisy environments, and establish logical associations between curves.

[0035] By introducing advanced deep learning methods and multi-modal feature fusion technologies, the present invention can effectively process complex shapes such as multi-peak curves, non-linear curves, and jagged curves. Using the adaptive curve segmentation and CNN feature extraction algorithm, combined with image semantic understanding, the recognition accuracy and fitting accuracy of complex curves are significantly improved.

[0036] For the complex scenario of multi-curve intersection, the present invention proposes a curve separation and association algorithm based on the graph neural network (GNN), which can automatically identify and distinguish multiple curves in the image, and at the same time establish logical associations between curves, avoiding confusion and incorrect associations, so as to achieve high-precision multi-curve extraction.

[0037] Embodiment 2

[0038] After obtaining an image file containing multiple material data curves in Embodiment 1, the method further includes: Step 1: Axis detection: Detect the position and boundaries of the axes in the image file through image processing techniques.

[0039] Step 2: Scale recognition: Use the deep learning model GNN to parse the scale values of the axes, and use multi-task learning or staged processing to parse the units of the axes. Combine text detection, numerical regression, and semantic understanding techniques. Through the collaborative modeling of object detection, OCR, and semantic understanding, and combine multi-task learning with physical rule post-processing to parse the scale values and their units of the axes, and adapt to different scales and layout scenarios.

[0040] Step 3: Non-standard axis processing: Support the recognition and processing of logarithmic axes (ordered axes) and non-linear scales (unordered axes).

[0041] Curve recognition requires accurate identification of the position, scale, and unit of the axes. However, in the prior art, when dealing with non-standard axes (such as logarithmic axes and non-linear scales), there may be problems of misidentification or insufficient accuracy. The present invention performs semantic parsing on the axes and scales through a deep learning model, can automatically identify and process complex situations such as logarithmic axes and non-linear scales, and at the same time supports custom axis annotation, greatly improving the adaptability and flexibility of the system.

[0042] To solve the dependence of deep learning methods on labeled data, the present invention introduces semi-supervised learning and transfer learning techniques. It can quickly adapt to new data types and scenarios through transfer learning based on a small amount of labeled data, and at the same time use semi-supervised learning to automatically label unlabeled data, significantly reducing the data annotation cost.

[0043] Specifically: Perform semi-supervised learning through a small amount of labeled data and a large amount of unlabeled data to automatically label unlabeled data; use transfer learning to quickly adapt to new data types and scenarios using a pre-trained model; when optimizing the model, verify the accuracy of the model through K-fold cross-validation and curve fitting, and optimize the model parameters.

[0044] In summary: The present invention proposes an efficient, accurate, and widely applicable curve recognition and processing method, aiming to solve the deficiencies of the prior art in complex curve recognition, image quality dependence, algorithm generalization ability, etc. The present invention realizes the accurate recognition of complex curves, the separation and association of multiple curves, the intelligent recognition of non-standard axes, and efficient real-time processing through innovative technologies such as multi-modal feature fusion, adaptive preprocessing, and intelligent data annotation.

[0045] Obviously, those skilled in the art can make various changes and modifications to this technical solution without departing from the spirit and scope of this technical solution. Thus, if these modifications and variations of this technical solution fall within the scope of the claims of this technical solution and their equivalent technologies, then this technical solution is also intended to include these changes and modifications.

Claims

1. A method for identifying and fitting material data curves based on deep learning, characterized in that, Including: Obtain an image file containing multiple material data curves, where the curves are used to show the performance changes of the material; Use image segmentation technology to separate the material data curves from the image background, and extract the feature points of the material data curves through a convolutional neural network; Using the feature points as control points, perform parametric fitting on the curve through polynomial or spline interpolation to generate a smooth continuous curve.

2. The method for identifying and fitting material data curves based on deep learning according to claim 1, wherein After obtaining the image file containing multiple material data curves, the method further includes: Convert the color image to a grayscale image, and then perform binarization processing to highlight the curve features; Remove noise through Gaussian filtering, and use an edge detection algorithm to enhance the curve edges; Automatically adjust the contrast and brightness according to the image quality to optimize the overall visual effect of the image.

3. The method for identifying and fitting a material data curve based on deep learning according to claim 1, wherein The extracting the feature points of the material data curves through the convolutional neural network specifically includes: Construct a training set and a test set, both of which contain a large number of material data curves with labeled feature points; Use the training set to train the convolutional neural network model so that the trained convolutional neural network model can extract the coordinate values of the feature points of the material data curves; use the test set to evaluate the model performance of the convolutional neural network model; Input the material data curve to be extracted into the trained convolutional neural network model, and output the coordinate values of the feature points of the material data curve.

4. The method for identifying and fitting material data curves based on deep learning according to claim 3, wherein, When performing the annotation of the feature points, perform semi-supervised learning through a small amount of labeled data and a large amount of unlabeled data to automatically label the unlabeled data.

5. The method for identifying and fitting material data curves based on deep learning according to claim 1, wherein The method further includes: modeling the geometric and topological relationships of the curves through a graph neural network graph structure, and using message passing and hierarchical feature learning to achieve efficient differentiation of multiple curves.

6. The method for identifying and fitting material data curves based on deep learning according to claim 1, characterized in that After obtaining the image file containing multiple material data curves, the method further includes: Detect the position and boundary of the coordinate axes in the image file through image processing technology; Use the deep learning model GNN to parse the scale values of the coordinate axes, use multi-task learning or staged processing to parse the units of the coordinate axes, combine text detection, numerical regression and semantic understanding technologies, through the collaborative modeling of object detection, OCR and semantic understanding, and combine multi-task learning and physical rule post-processing to parse the scale values and their units of the coordinate axes, and adapt to different scales and layout scenarios.

Images