Method and system for automatically identifying section image of carbon fiber scanning electron microscope

Abstract

The invention relates to the technical field of image recognition, in particular to a carbon fiber scanning electron microscope cross section image automatic recognition method and system.The carbon fiber scanning electron microscope cross section image automatic recognition method comprises the steps that a carbon fiber cross section image collected by a scanning electron microscope is obtained, uniformly dividing the carbon fiber cross section image into a plurality of sub-images; inputting each sub-graph into a carbon fiber cross section recognition model for feature extraction and classification judgment, wherein the carbon fiber cross section recognition model is a neural network model based on the combination of an improved residual neural network and a transfer learning mechanism; and obtaining a classification result of each sub-image, and determining a final classification result of the carbon fiber cross-section image based on a majority voting mechanism to obtain a carbon fiber quality detection result. The efficiency and the standardization level of carbon fiber quality detection can be remarkably improved.

Description

Technical Field

[0001] This application relates to the field of image recognition technology, and in particular to an automatic recognition method and system for cross-sectional images of carbon fiber scanning electron microscopes. Background Technology

[0002] With the widespread application of carbon fiber composites in aerospace, high-end manufacturing, and other fields, the accurate identification and quality control of their microstructure have become crucial. The microscopic characteristics of carbon fiber cross-sections, such as geometry, pore distribution, and edge sharpness, directly affect their macroscopic properties, including tensile strength, thermal conductivity, and fatigue life. Scanning electron microscopy (SEM), with its high resolution and excellent imaging capabilities, has become the standard method for inspecting the cross-sectional structure of carbon fibers. However, SEM images typically involve large amounts of data, complex textures, and significant grayscale gradients. Coupled with variable imaging conditions, this presents significant challenges to traditional manual identification methods. Manual analysis is not only inefficient and costly, but also susceptible to significant subjective errors due to the operator's experience level, making it unsuitable for standardized processing of large batches of samples.

[0003] In the past, researchers have mostly used shallow machine learning methods based on manual feature extraction, such as edge extraction combined with classification models like SVM and KNN, to identify the quality of carbon fiber morphology. These methods suffer from poor stability when dealing with blurred contours, uneven contrast, or defective areas in the image, and lack the ability to learn from complex textures, failing to meet the accuracy requirements for microscale recognition. Therefore, in recent years, deep learning models such as Convolutional Neural Networks (CNNs) have been introduced to achieve end-to-end image classification. Typical networks such as AlexNet, VGGNet, and ResNet-18 have achieved good results in natural image classification, but their application in material microscopic image processing still faces significant bottlenecks. On the one hand, carbon fiber SEM images exhibit unique characteristics such as low contrast, weak edges, and high similarity, making it difficult for conventional networks to fully extract microscopic differences. On the other hand, image quality fluctuations caused by different imaging magnifications, lighting conditions, and sample preparation processes make models prone to overfitting or decreased generalization performance. Summary of the Invention

[0004] This application provides an automatic identification method and system for carbon fiber scanning electron microscope cross-sectional images, which overcomes the shortcomings of existing carbon fiber cross-sectional image identification methods such as low accuracy, poor robustness, and reliance on human experience. It features high precision, high adaptability, and high degree of automation, and can significantly improve the efficiency and standardization of carbon fiber quality inspection.

[0005] In a first aspect, this application provides an automatic identification method for cross-sectional images of carbon fiber scanning electron microscopes, including: A cross-sectional image of carbon fiber acquired by a scanning electron microscope is obtained, and the cross-sectional image of carbon fiber is uniformly divided into multiple sub-images; Each of the sub-images is input into the carbon fiber cross-section recognition model for feature extraction and classification. The carbon fiber cross-section recognition model is a neural network model based on a combination of an improved residual neural network and a transfer learning mechanism. The classification results of each sub-image are obtained, and the final classification result of the carbon fiber cross-section image is determined based on the majority voting mechanism to obtain the carbon fiber quality inspection result.

[0006] Furthermore, the step of acquiring the carbon fiber cross-sectional image obtained by scanning electron microscopy, and after uniformly dividing the carbon fiber cross-sectional image into multiple sub-images, further includes: Each of the sub-images is subjected to preprocessing and data augmentation operations. The preprocessing includes grayscale conversion, contrast enhancement, and noise reduction filtering. The data augmentation operations include rotation, flipping, scaling, and translation.

[0007] Furthermore, the improved residual neural network is built based on ResNet-50.

[0008] Furthermore, the improved residual neural network freezes all convolutional layers and only fine-tunes the fully connected layers.

[0009] Furthermore, the improved residual neural network includes a residual module that introduces an attention mechanism, namely an SE channel attention module.

[0010] Furthermore, the transfer learning mechanism uses a pre-trained model as the base weights and only updates the parameters of the final fully connected layer.

[0011] Furthermore, the improved residual neural network training process uses the Adam optimizer, with an initial learning rate of 0.001, a batch size of 32, 100 training rounds, and a cross-entropy loss function.

[0012] Secondly, this application provides an automatic recognition system for cross-sectional images of carbon fiber scanning electron microscopes, comprising: The image acquisition module is used to acquire cross-sectional images of carbon fibers acquired by a scanning electron microscope, and to uniformly divide the cross-sectional images of carbon fibers into multiple sub-images; The model inference module is used to input each of the sub-images into the carbon fiber cross-section recognition model for feature extraction and classification. The carbon fiber cross-section recognition model is a neural network model based on a combination of an improved residual neural network and a transfer learning mechanism. The result fusion module is used to obtain the classification results of each of the sub-images and determine the final classification result of the carbon fiber cross-section image based on the majority voting mechanism, so as to obtain the carbon fiber quality inspection result.

[0013] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the automatic recognition method for cross-sectional images of carbon fiber scanning electron microscopes as described above.

[0014] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the automatic recognition method for cross-sectional images of carbon fiber scanning electron microscopes as described above.

[0015] The above-mentioned technical solution of this application has the following advantages: The automatic identification method for carbon fiber scanning electron microscope cross-sectional images provided in the first aspect of this application acquires carbon fiber cross-sectional images collected by scanning electron microscopes, uniformly divides the carbon fiber cross-sectional images into multiple sub-images, and inputs each sub-image into a carbon fiber cross-sectional identification model for feature extraction and classification. The carbon fiber cross-sectional identification model is a neural network model based on a combination of an improved residual neural network and a transfer learning mechanism. The classification results of each sub-image are obtained, and the final classification result of the carbon fiber cross-sectional image is determined based on a majority voting mechanism to obtain the carbon fiber quality inspection result. It has the characteristics of high precision, high adaptability and high degree of automation, and can significantly improve the efficiency and standardization level of carbon fiber quality inspection.

[0016] It is understood that the beneficial effects of the second, third and fourth aspects mentioned above can be found in the relevant descriptions in the first aspect above, and will not be repeated here. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the specific embodiments of this application or the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 A flowchart of the automatic recognition method for cross-sectional images of carbon fiber scanning electron microscope provided in this application; Figure 2 The data preprocessing and dataset construction flowchart provided for this application; Figure 3Accuracy variation curves and confusion matrix diagrams of the model training process provided in this application; Figure 4 The training accuracy curve and confusion matrix of the transfer learning model provided in this application on the validation set; Figure 5 The classification and inference results of the model provided in this application in carbon fiber SEM images at different magnifications; Figure 6 A structural diagram of the automatic recognition system for cross-sectional images of carbon fiber scanning electron microscopes provided in this application; Figure 7 A structural diagram of the electronic device provided in this application. Detailed Implementation

[0019] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of this application with unnecessary detail.

[0020] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0021] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0022] References to "one embodiment" or "some embodiments" in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized. "A plurality" means "two or more."

[0023] In existing research, some works have attempted to classify material images using deep networks such as ResNet-50, achieving a certain level of accuracy. However, these models are mostly pre-trained on natural image datasets such as ImageNet, lacking specific adaptation to the characteristics of material images. Furthermore, the network structure and training mechanisms have not been optimized for the microstructural features of carbon fibers. In practical applications, batch SEM image classification often involves judging complex information from numerous image regions; a single model is easily affected by local defects in the image, leading to recognition errors. Moreover, models lacking rapid adaptability cannot meet the deployment requirements of industrial applications given the diversity of image quality.

[0024] In summary, existing technologies for intelligent recognition of carbon fiber SEM images face several unresolved issues, including insufficient network structure for extracting fine features, weak model generalization and adaptability, lack of standardized processing procedures, and impractical system deployability. These problems severely limit the widespread application of image recognition models in practical carbon fiber quality inspection.

[0025] To address the aforementioned technical challenges, this application proposes an automatic identification method for cross-sectional images of carbon fiber scanning electron microscopes. This method combines a structure-optimized deep neural network with a transfer learning mechanism and constructs a corresponding automated detection system to improve the accuracy, stability, and application adaptability of the identification. This technical approach overcomes the limitations of traditional methods, such as reliance on large samples, low identification accuracy, and poor model adaptability, and possesses significant engineering practical value and potential for widespread application.

[0026] The specific embodiments of this application will be described in further detail below with reference to the accompanying drawings and examples. The following examples are used to illustrate this application, but are not intended to limit the scope of this application.

[0027] This application provides a method for automatic recognition of cross-sectional images of carbon fiber scanning electron microscope, which specifically includes the following steps: Obtain cross-sectional images of carbon fibers acquired by scanning electron microscopy (SEM) and divide the cross-sectional images of carbon fibers into multiple sub-image regions; Each sub-image undergoes preprocessing, including grayscale conversion, contrast enhancement, and noise reduction filtering; specifically, this includes using the CLAHE algorithm for local contrast enhancement. Data augmentation operations are performed on each subgraph, including rotation, flipping, scaling, or translation; specifically, random rotation within ±15 degrees, as well as horizontal mirror flipping and scaling, are performed to enhance the model's ability to identify fiber cross-sections in different directions. The improved ResNet-50 neural network is used for feature extraction and classification, where all convolutional layers are frozen and only the fully connected layers are fine-tuned. The improved ResNet-50 neural network includes a residual module that introduces an attention mechanism to enhance the ability to focus on detailed structures. The attention mechanism is an SE channel attention module, which is used to improve the adaptive ability of feature channel weights. The transfer learning process uses a pre-trained model as the base weights and only updates the parameters of the final fully connected layer; the classification model uses the cross-entropy loss function as the objective function, the optimization method is the Adam optimizer, the learning rate is 0.001, and the batch size is 32. Obtain the classification results of each sub-image, classify the entire image based on the majority voting mechanism, and output the carbon fiber quality inspection results.

[0028] This application focuses on the challenge of intelligent inspection of the microstructure quality of carbon fiber composite materials. It proposes and constructs a carbon fiber cross-section recognition model based on a combination of an improved residual neural network (ResNet-50) and a transfer learning mechanism, used for automated identification of the quality of carbon fiber cross-sections in scanning electron microscope (SEM) images. Compared with traditional manual evaluation methods, this method has significant advantages such as high recognition efficiency, low subjective error, and strong batch processing capability.

[0029] The innovation of this application is reflected in three aspects: 1. Network architecture innovation: Convolutional kernel optimization and local attention mechanism are introduced on the basis of standard ResNet-50 to improve the model's ability to capture micro-details (such as holes and flat cross-sections); 2. Transfer learning mechanism embedding: By freezing the feature extraction layer and fine-tuning only the fully connected classifier, the model's adaptability to different image qualities and imaging conditions at different magnifications is enhanced; 3. Image intelligent partitioning and majority voting fusion strategy: The entire SEM image is divided into multiple sub-images, which are identified independently and then the final results are statistically fused to improve the stability and accuracy of the overall judgment.

[0030] The automatic recognition method for carbon fiber scanning electron microscope cross-sectional images provided in this application first performs multi-step preprocessing on the original SEM image, including grayscale normalization, contrast enhancement, and noise suppression, to enhance the image's detail and edge structure clarity, thereby improving the effectiveness of subsequent feature extraction. Subsequently, data augmentation techniques (such as rotation, flipping, and translation) are used to expand the original training sample space, addressing the problem of limited sample sizes for actual material images and providing more generalization training opportunities for deep neural networks.

[0031] Regarding the neural network architecture, this application selects ResNet-50 as the basic architecture and optimizes its residual modules. Specifically, this includes adjusting the convolutional kernel size, embedding a channel attention mechanism, and adopting activation functions and normalization strategies that are more adapted to the distribution characteristics of material images. This structural design significantly improves the model's ability to extract complex textures, weak edges, and irregular shapes, thereby achieving high-sensitivity recognition of the morphological details of carbon fiber cross-sections. Through the residual connection mechanism, the network effectively avoids problems such as gradient vanishing, allowing deeper layers of the network to maintain training stability.

[0032] To further enhance the model's adaptability to multi-source images, this application introduces a transfer learning mechanism. Specifically, the convolutional layers pre-trained on the initial training dataset are frozen, and only the fully connected classification layers are fine-tuned. This allows the model to quickly adapt to new image domain features without retraining all parameters. This strategy not only significantly reduces training costs and computational resource consumption but also effectively prevents overfitting risks caused by insufficient training data. This method is particularly suitable for industrial applications where image quality is significantly affected by factors such as imaging equipment and sample preparation methods.

[0033] Considering that a single SEM image may contain multiple carbon fiber cross-sectional regions, direct overall classification is often susceptible to interference from local noise or structural overlap. Therefore, this application designs an image partitioning recognition mechanism: the original image is divided into several sub-image regions, each input into the model for independent classification, and finally, the entire image is comprehensively classified through a majority voting mechanism. This strategy fully utilizes the redundancy of local features, effectively improving the robustness and anti-interference ability of the overall recognition.

[0034] A large-scale dataset containing over 13,000 carbon fiber SEM images was constructed, and the model's stability and generalization ability were validated in batch data from practical engineering applications. Preliminary experimental results show a classification accuracy of 97.13% and an F1-score of 97.03%. While maintaining high accuracy, it achieves an inference speed of hundreds of images per second, significantly improving the automation level of carbon fiber quality inspection. This application has been embedded in a prototype system platform, possessing a solid foundation for engineering applications. This application will have significant practical implications for promoting quality standardization in the carbon fiber industry, reducing manual inspection costs, and accelerating product delivery efficiency, providing a technological demonstration for the field of intelligent recognition of material microscopic images.

[0035] Based on the above identification process, this application further constructs an intelligent detection platform integrating image preprocessing, model inference, result fusion, and visualization feedback. Users can upload SEM images through the interface, and the system will automatically complete image segmentation, batch processing, and quality judgment output, which is suitable for quality inspection in actual carbon fiber production lines or laboratories.

[0036] The following is a description through specific embodiments.

[0037] Example The automatic carbon fiber cross-section image recognition method provided in this embodiment is applicable to the classification task of scanning electron microscope (SEM) images, aiming to achieve automatic identification and standardized detection of carbon fiber quality grades. Its overall system structure is as follows: Figure 1 As shown, it includes an image acquisition and segmentation module, an image preprocessing module, a deep feature extraction module, a classification decision module, and a visualization output module. Figure 1 This diagram presents the overall flowchart of a carbon fiber cross-section image classification model, illustrating the entire process from raw image acquisition, image preprocessing, feature extraction, classification decision to result output. The diagram includes steps such as image grayscale conversion, convolution processing, activation, and pooling, demonstrating the application structure of deep learning models in SEM image processing.

[0038] First, images of the carbon fiber cross-section were obtained using a scanning electron microscope. Figure 2 (As shown). The original image size is a high-resolution full-frame image. The image was divided into 12 sub-images using the OpenCV image processing library. The division ratio is 4:3 (horizontal:vertical), ensuring that each sub-image contains a complete or partial carbon fiber cross-section.

[0039] Subsequently, the segmented images undergo image preprocessing, including grayscale conversion, contrast enhancement, noise suppression, and edge refinement, to improve the clarity of microstructures. Addressing common image issues such as blurred edges and low contrast, this embodiment employs linear stretching combined with CLAHE (Adaptive Histogram Equalization) for brightness and texture enhancement. To expand the training data space, this embodiment further applies data augmentation techniques, including rotation (within ±15°), translation (within 10 pixels), mirroring, and scaling, to improve the model's generalization ability. Figure 2 A flowchart illustrating the data preprocessing and dataset construction process demonstrates how the original carbon fiber SEM image is divided into multiple sub-images, as well as the organization of the training, validation, and test sets. The flowchart also indicates the classification path and storage structure for positive and negative samples.

[0040] In the deep feature extraction stage, this embodiment constructs a model structure based on the ResNet-50 deep residual neural network, such as... Figure 3As shown. To adapt to the texture characteristics and detail extraction requirements of carbon fiber SEM images, the standard ResNet residual module is structurally optimized, specifically including: using multi-scale convolutional kernels (1×1, 3×3 hybrid) for feature extraction; introducing channel attention mechanisms (such as the SE module) to improve the ability to identify key regions; replacing some ReLU activation functions with Swish to enhance nonlinear expression capabilities; and preserving low-level features in the residual path to enhance the model's sensitivity to detailed structures such as edges and pores.

[0041] This embodiment employs a transfer learning mechanism to train the model. Specifically, it uses ResNet-50 weights pre-trained with a large number of fiber cross-sections as the backbone network, freezes all convolutional layers, retains and fine-tunes only the last fully connected layer, and replaces its output unit count with 2 to achieve a binary classification task of "qualified" and "unqualified". The model training process uses the Adam optimizer, with an initial learning rate of 0.001, a batch size of 32, 100 training epochs, and cross-entropy loss as the loss function.

[0042] The training dataset contains 13,082 images: 8,424 for training, 2,406 for validation, and 2,252 for testing (approximately a 7:2:1 ratio). An additional 3,248 new images from different imaging conditions were prepared for transfer training evaluation to ensure the model's ability to adapt to multi-source images. During training, the model accuracy remained stable above 97%, and the confusion matrix (…) Figure 3 , Figure 4 The system shows that there are very few false positives.

[0043] Figure 3 The left and right graphs show the accuracy change curves and confusion matrix of the model during training. The left graph reflects the accuracy improvement trend of the ResNet-50 model during training, while the right graph shows the classification performance of the model on the validation set, including the comparison between the true and predicted labels. Figure 4 The training accuracy curve and confusion matrix of the model on the validation set after transfer learning reflect the changes in classification accuracy and generalization ability of the model after adapting to different imaging conditions, including the improvement in recognition accuracy and visual comparison of misclassified regions.

[0044] In the classification decision-making stage, a subgraph-level discrimination + majority voting mechanism is used to determine the entire graph. For example... Figure 5 As shown, each original SEM image is segmented into several sub-images, which are then input into a neural network model for separate classification. Finally, the classification result of the entire image is determined according to the "majority rule". This mechanism significantly improves the robustness and fault tolerance of the overall discrimination, especially when there is local contamination, overlap, or blurring in the image. Figure 5The model's classification and inference performance on carbon fiber SEM images at different magnifications is demonstrated, including images at four magnification levels: 1000×, 2000×, 4000×, and 8000×. The gray areas in the figure represent misclassified regions where the model occurred at different image resolutions, highlighting the difficulty of detail recognition in high-magnification images.

[0045] To facilitate practical deployment and use, this embodiment also developed an embedded intelligent detection platform. The platform integrates modules for image uploading, automatic partitioning, batch recognition, and result visualization, and features real-time processing and report export capabilities. The neural network model is deployed within the embedded detection platform, which provides functional interfaces for image uploading, automatic recognition, and result visualization output. Users can upload SEM images through a graphical interface; the system automatically completes all image processing steps and returns classification results and probabilities, making it suitable for batch product testing in enterprises and automated analysis in research institutes.

[0046] Furthermore, to verify the model's generalization ability at different magnifications, this embodiment tested four image resolutions: 1000×, 2000×, 4000×, and 8000×. Figure 5 As shown, the model performs best at 2000× resolution; in high-magnification images (such as 8000×), due to increased texture complexity, the model misclassifies more areas, but overall it still possesses strong discrimination ability. This test demonstrates the adaptability of this application to image variations in real-world scenarios.

[0047] In summary, this embodiment fully reproduces the automatic carbon fiber identification process and system construction method proposed in this application, and has good feasibility, versatility and engineering value.

[0048] Corresponding to the automatic recognition method for cross-sectional images of carbon fiber scanning electron microscopes described above, this application also provides an automatic recognition system for cross-sectional images of carbon fiber scanning electron microscopes, such as... Figure 6 As shown, the automatic recognition system for cross-sectional images of carbon fiber scanning electron microscopes includes: The image acquisition module is used to acquire cross-sectional images of carbon fibers acquired by a scanning electron microscope, and to uniformly divide the cross-sectional images of carbon fibers into multiple sub-images; The model inference module is used to input each of the sub-images into the carbon fiber cross-section recognition model for feature extraction and classification. The carbon fiber cross-section recognition model is a neural network model based on a combination of an improved residual neural network and a transfer learning mechanism. The result fusion module is used to obtain the classification results of each of the sub-images and determine the final classification result of the carbon fiber cross-section image based on the majority voting mechanism, so as to obtain the carbon fiber quality inspection result.

[0049] Compared with existing technologies, this application has the following advantages: 1. Significantly improved recognition accuracy. The improved ResNet network structure, combined with image enhancement and transfer learning strategies, enables the model to achieve an accuracy of over 97% on the test dataset; 2. Enhanced model generalization ability. The transfer learning mechanism ensures that the model remains highly adaptable to varying image magnification, contrast, and imaging noise; 3. Reduced data and computational costs. By freezing the backbone network and optimizing only the classifier part, training time and resource requirements are significantly reduced; 4. Enhanced system practicality. The introduction of image sub-region voting fusion and embedded platform design ensures more stable results and enables automated batch recognition; 5. Strong scalability. This scheme can be widely extended to intelligent image recognition tasks of microstructures of other types of materials (such as ceramic fibers, glass fibers, etc.).

[0050] In summary, this application has successfully constructed an intelligent identification method and system suitable for industrial-grade carbon fiber quality inspection by combining advanced deep learning models, transfer learning strategies and multi-region fusion mechanisms. This method fills the gaps in traditional image processing methods in terms of recognition accuracy, adaptability and automation level, and has significant engineering application value and industrial promotion prospects.

[0051] This application also provides an electronic device, such as Figure 7 As shown, it includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the automatic recognition method for cross-sectional images of carbon fiber scanning electron microscopes provided in the first aspect.

[0052] In applications, electronic devices may include, but are not limited to, processors and memory. Figure 7 This is merely an example of an electronic device and does not constitute a limitation on the device. It may include more or fewer components than illustrated, or combinations of certain components, or different components, such as input / output devices, network access devices, etc. Input / output devices may include cameras, audio capture / playback devices, displays, etc. Network access devices may include network modules for wireless network communication with external devices.

[0053] In applications, the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0054] In applications, memory can be an internal storage unit of an electronic device in some embodiments, such as a hard drive or RAM. In other embodiments, memory can be an external storage device of the electronic device, such as a plug-in hard drive, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card. Memory can also include both internal and external storage units of the electronic device. Memory is used to store operating systems, applications, bootloaders, data, and other programs, such as program code for computer programs. Memory can also be used to temporarily store data that has been output or will be output.

[0055] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps in the above-described method embodiments.

[0056] This application implements all or part of the processes in the methods of the above embodiments, which can be accomplished by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium can at least include: any entity or device capable of carrying the computer program code to an electronic device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, such as a USB flash drive, a portable hard drive, a magnetic disk, or an optical disk.

[0057] Those skilled in the art will recognize that the device and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0058] In the embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interface, or the device may be indirectly coupled or communicated, and may be electrical, mechanical, or other forms.

[0059] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for automatic recognition of cross-sectional images of carbon fiber scanning electron microscope, characterized in that, include: A cross-sectional image of carbon fiber acquired by a scanning electron microscope is obtained, and the cross-sectional image of carbon fiber is uniformly divided into multiple sub-images; Each of the sub-images is input into the carbon fiber cross-section recognition model for feature extraction and classification. The carbon fiber cross-section recognition model is a neural network model based on a combination of an improved residual neural network and a transfer learning mechanism. The classification results of each sub-image are obtained, and the final classification result of the carbon fiber cross-section image is determined based on the majority voting mechanism to obtain the carbon fiber quality inspection result.

2. The automatic recognition method for cross-sectional images of carbon fiber scanning electron microscopes as described in claim 1, characterized in that, The process of acquiring a carbon fiber cross-sectional image using a scanning electron microscope, and then uniformly dividing the carbon fiber cross-sectional image into multiple sub-images, further includes: Each of the sub-images is subjected to preprocessing and data augmentation operations. The preprocessing includes grayscale conversion, contrast enhancement, and noise reduction filtering. The data augmentation operations include rotation, flipping, scaling, and translation.

3. The automatic recognition method for cross-sectional images of carbon fiber scanning electron microscopes as described in claim 1, characterized in that, The improved residual neural network is built based on ResNet-50.

4. The automatic recognition method for cross-sectional images of carbon fiber scanning electron microscopes as described in claim 1, characterized in that, The improved residual neural network freezes all convolutional layers and only fine-tunes the fully connected layers.

5. The automatic recognition method for cross-sectional images of carbon fiber scanning electron microscopes as described in claim 1, characterized in that, The improved residual neural network includes a residual module that introduces an attention mechanism, namely an SE channel attention module.

6. The automatic recognition method for cross-sectional images of carbon fiber scanning electron microscopes as described in claim 1, characterized in that, The transfer learning mechanism uses a pre-trained model as the base weights and updates the parameters only for the final fully connected layer.

7. The automatic recognition method for cross-sectional images of carbon fiber scanning electron microscopes as described in claim 1, characterized in that, The improved residual neural network training process uses the Adam optimizer, with an initial learning rate of 0.001, a batch size of 32, 100 training rounds, and cross-entropy loss as the loss function.

8. An automatic recognition system for cross-sectional images of carbon fiber scanning electron microscopes, characterized in that, include: The image acquisition module is used to acquire cross-sectional images of carbon fibers acquired by a scanning electron microscope, and to uniformly divide the cross-sectional images of carbon fibers into multiple sub-images; The model inference module is used to input each of the sub-images into the carbon fiber cross-section recognition model for feature extraction and classification. The carbon fiber cross-section recognition model is a neural network model based on a combination of an improved residual neural network and a transfer learning mechanism. The result fusion module is used to obtain the classification results of each of the sub-images and determine the final classification result of the carbon fiber cross-section image based on the majority voting mechanism, so as to obtain the carbon fiber quality inspection result.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the automatic recognition method for cross-sectional images of carbon fiber scanning electron microscopes as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the automatic recognition method for cross-sectional images of carbon fiber scanning electron microscopes as described in any one of claims 1 to 7.

Images