Application method for classification and identification of non-abandoned images

Abstract

The invention discloses an application method for classification and recognition of non-abandoned images, and the method comprises the steps: S1, obtaining an initial non-abandoned image dataset which comprises Shenembroidery and related non-abandoned images collected from a museum, a public dataset or a web crawler, guaranteeing the image to be clear and complete, carrying out the format standardization of the image, uniformly adjusting the image to be 224 * 224 in resolution, and storing the image in a jpg format, through data enhancement methods such as horizontal overturning, rotation, brightness and contrast adjustment and the like, the image diversity is increased, and about 15 enhanced versions are generated from each original image. The method comprises the steps of data set preparation and enhancement, model improvement, transfer learning, training process, software design and test and the like. According to the method, spatial pyramid pooling and transfer learning are introduced, so that the recognition accuracy of the model on non-abandoned images is effectively improved, and a test result shows that the recognition accuracy of the improved model reaches 98.65%. The Shen embroidery image recognition software designed in the invention also shows high recognition precision, and has wide popularization and application values.

Description

Technical Field

[0001] The present invention relates to the technical field of image classification and recognition, and in particular to an application method for classification and recognition of intangible cultural heritage images. Background Art

[0002] With the rapid development of information technology and digital technology, image recognition technology has become a hot topic in research and application in many fields. Especially in the field of cultural heritage protection, the application of image recognition technology provides new ways and means for the protection, inheritance and promotion of intangible cultural heritage (abbreviated as "intangible cultural heritage"). As a treasure of human civilization, the protection and inheritance of intangible cultural heritage is of great significance.

[0003] Shen embroidery, as an intangible cultural heritage of arts and crafts with unique artistic style and regional characteristics, is deeply loved by people for its exquisite embroidery skills and profound cultural connotations. However, the traditional identification and authentication methods of Shen embroidery mainly rely on the manual judgment of experts. This method is not only inefficient, but also easily affected by subjective factors, and cannot meet the needs of large-scale and efficient intangible cultural heritage image recognition.

[0004] In recent years, the rise of deep learning technology has brought revolutionary breakthroughs in the field of image recognition. Among them, MobileNet V3, as a lightweight and efficient convolutional neural network model, has performed well in tasks such as image classification and object detection. However, there are still many challenges in directly applying the existing MobileNet V3 model to the recognition of intangible cultural heritage images of arts and crafts such as Shen embroidery, such as insufficient model generalization ability and inaccurate feature extraction of specific intangible cultural heritage images.

[0005] To address the above issues, we have introduced an application method for intangible cultural heritage image classification and recognition. Summary of the invention

[0006] The invention discloses an application method for classification and recognition of intangible cultural heritage images, aiming to solve the technical problems in the background technology.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] An application method for classification and recognition of intangible cultural heritage images, comprising:

[0009] S1. Obtain the initial intangible cultural heritage image dataset, including Shen embroidery and related intangible cultural heritage images collected from museums, public datasets or through web crawlers, ensure that the images are clear and complete, standardize the image format, uniformly adjust the resolution to 224×224 and save it in jpg format, increase image diversity through data enhancement methods such as horizontal flipping, rotation, brightness and contrast adjustment, generate about 15 enhanced versions for each original image, and finally expand the dataset to 18,960 images to increase the amount of training data;

[0010] The step S1 can be refined as follows:

[0011] S11, obtaining an initial intangible cultural heritage image dataset, wherein the dataset includes Shen embroidery and other related intangible cultural heritage image data collected from museums, public datasets or through web crawler technology, to ensure that the images have high fidelity and integrity;

[0012] S12, standardize the format of the initial intangible cultural heritage image data, adjust all images to a uniform resolution of 224×224, and save them in a uniform jpg format to ensure that the subsequent training model can smoothly accept the data;

[0013] S13, expanding the data set by using data enhancement technology, wherein the data enhancement processing includes operations such as horizontal flipping, vertical flipping, image rotation, brightness adjustment, and contrast adjustment of the image to simulate different image shooting conditions and improve the diversity of the data set;

[0014] S14. After data augmentation, each original image generates up to 15 different variant images through augmentation technology, and finally the dataset is expanded to 18,960 images to significantly increase the amount of training data.

[0015] S2. MobileNetV3 is selected as the basic architecture, and its lightweight and efficient performance is used to meet the needs of intangible cultural heritage image recognition. The spatial pyramid pooling (SPP) module is introduced to enhance the recognition of image features of different scales. The model parameters are reduced through deep separable convolution while maintaining computational efficiency to adapt to the limited resources of mobile devices. The optimized model structure is used for efficient classification and recognition of intangible cultural heritage image features.

[0016] The step S2 can be refined as follows:

[0017] S21. Select MobileNetV3 as the basic architecture of the recognition network, and use its lightweight advantages and high computing efficiency on mobile devices to meet the needs of intangible cultural heritage image recognition and ensure the efficient operation of the system;

[0018] S22. Based on MobileNetV3, the Spatial Pyramid Pooling (SPP) module is introduced to enhance the network's ability to recognize features of images of different scales and improve the recognition effect of multi-level and multi-texture non-legacy images.

[0019] S23. Optimize the design of convolutional layers in the network, reduce the number of model parameters by introducing depthwise separable convolution, while ensuring the computational efficiency of the model to adapt to the situation where mobile device resources are limited;

[0020] S24. Use the optimized model structure to classify and identify the initially extracted intangible cultural heritage image features to ensure the model's efficient extraction and classification of intangible cultural heritage image features.

[0021] S3. Use transfer learning, take the MobileNetV3 pre-trained model trained on large data sets (such as ImageNet) as the basis, reduce the initial training time, fine-tune the top layer of the pre-trained model, retrain on the features of intangible cultural heritage images to adapt to the features of intangible cultural heritage images such as Shen embroidery, import the Shen embroidery image dataset for training, improve recognition accuracy and model robustness, reduce the risk of overfitting, and speed up model convergence;

[0022] The step S3 can be refined as follows:

[0023] S31. Use the transfer learning method to use the MobileNetV3 pre-trained model trained on large public datasets (such as ImageNet) as the basic model to reduce the training time of the model in the initial stage;

[0024] S32, unfreeze and fine-tune the top layer of the pre-trained model, and retrain the network weights according to the specific characteristics of the intangible cultural heritage images to ensure that the model can better adapt to the feature representation of intangible cultural heritage images such as Shen embroidery;

[0025] S33, importing the Shen embroidery image dataset into the model for transfer learning training, using the generalized features learned by the pre-training model to improve the accuracy and robustness of intangible cultural heritage images in recognition tasks;

[0026] S34. Accelerate the convergence of the model through transfer learning, reduce the risk of model overfitting, and further train and verify the intangible cultural heritage images after fine-tuning the model to improve the overall performance of the model.

[0027] S4. Use the enhanced intangible cultural heritage image dataset to divide the training set and validation set into 9:1. The training set is used for model training, and the validation set is used to monitor model performance and adjust parameters. Perform multiple training iterations through the optimized MobileNetV3 model, adjust hyperparameters such as learning rate and batch size to ensure efficient model convergence and avoid gradient problems. Finally, use the test set to evaluate the model to ensure its accuracy and robustness in intangible cultural heritage image recognition.

[0028] The step S4 can be refined as follows:

[0029] S41, using the enhanced intangible cultural heritage image dataset, dividing the training set and the validation set into a ratio of 9:1, the training set is used for model training, and the validation set is used to monitor the performance of the model during the training process and adjust the model parameters;

[0030] S42. Use the optimized MobileNetV3 improved model, perform model training on the training set, and adjust the network parameters through multiple training iterations to find the best parameter combination;

[0031] S43. During model training, adjust hyperparameters such as learning rate and batch size to avoid gradient vanishing or exploding problems and ensure that the model can converge efficiently.

[0032] S44. After training, a separate test set is used to conduct a final evaluation of the model to ensure its accuracy and robustness in the intangible cultural heritage image recognition task.

[0033] S5. Shen Embroidery Intangible Heritage Image Recognition Software developed using PyQt5 library has a human-computer interaction interface, supports users to import images in jpg, jpeg, and png formats, and automatically adjusts to 256×256 size. The software calculates the pixel ratio of image recognition, displays the recognition results, and divides them into 5 levels, where level 1 is a recognition rate of 95%-100%, and the lowest is level 5 (less than 80%);

[0034] The step S5 can be refined as follows:

[0035] S51. Develop a Shen embroidery intangible cultural heritage image recognition software system based on Python language. The system uses the PyQt5 library to design a human-computer interaction interface to achieve a user-friendly operation experience.

[0036] S52. The software has image input function, allowing users to import local files. Supported image formats include jpg, jpeg, png, etc. The software will automatically convert the input image to the standard size of 256×256;

[0037] S53, realizing the image recognition function, the system calculates the ratio of the total pixel value of the recognized category to the pixel value of the entire image, gives the accuracy of image recognition, and displays the specific recognition result on the interface;

[0038] S54. The software is divided into five recognition levels according to the recognition accuracy, namely level 1 (recognition rate 95%-100%), level 2 (90%-95%), level 3 (85%-90%), level 4 (80%-85%), and level 5 (less than 80%).

[0039] S6. Build a test library containing 100 Shen embroidery images and 20 non-Shen embroidery images, test the recognition accuracy and robustness of the software under different input conditions, analyze the recognition results through confusion matrix, evaluate the accuracy, precision and recall of the model, ensure that it can accurately distinguish Shen embroidery from non-Shen embroidery images, verify the compatibility and stability of the software under different resolutions and lighting conditions, adjust the model parameters according to the test results, ensure that the recognition accuracy reaches more than 98%, and optimize the interface and user experience;

[0040] The step S6 can be refined as follows:

[0041] S61, constructing a test image library including positive examples and negative examples, wherein the test image library includes 100 Shen embroidery images and 20 non-Shen embroidery images, and testing the recognition accuracy and robustness of the software under different image input conditions;

[0042] S62. During the test, the confusion matrix is ​​used to analyze the recognition results, and the recognition accuracy, precision and recall of the model are checked to ensure that the model can accurately distinguish between Shen embroidery images and non-Shen embroidery images;

[0043] S63. Verify the compatibility and stability of the software for input images of different resolutions and different lighting conditions to ensure the effectiveness and accuracy of the system in practical applications;

[0044] S64. Adjust model parameters according to test results to ensure recognition accuracy above 98%, optimize the user experience of the software interface, and improve interactive performance.

[0045] S7. Apply the improved MobileNetV3 model to the recognition of other arts and crafts intangible cultural heritage images, verify the generalization ability and adaptability of the model, expand the data set of the Shen embroidery recognition system, add more intangible cultural heritage images, improve the recognition range and application value, help the digital protection of intangible cultural heritage, optimize the mobile application version, achieve cross-platform support, and allow users to perform intangible cultural heritage image recognition at any time on smartphones, tablets and other devices;

[0046] The step S7 can be refined as follows:

[0047] S71. Expand and apply the improved image recognition model based on MobileNetV3 to the recognition of other arts and crafts intangible cultural heritage images to further verify the generalization ability and adaptability of the model;

[0048] S72. Expand the data set of the Shen Embroidery Recognition System, add more types of intangible cultural heritage image resources, improve the recognition scope and application value of the system, and further promote the digital protection of intangible cultural heritage;

[0049] S73. Optimize the mobile application version of Shen Embroidery Image Recognition Software to achieve cross-platform support. Users can perform intangible cultural heritage image recognition operations anytime and anywhere through smart phones, tablet computers and other devices.

[0050] The application method of the intangible cultural heritage image classification and recognition provided by the present invention has the following advantages:

[0051] The present invention covers the steps of data set preparation and enhancement, model improvement, transfer learning, training process, software design and testing. This method effectively improves the recognition accuracy of the model for intangible cultural heritage images by introducing spatial pyramid pooling and transfer learning. The test results show that the recognition accuracy of the improved model reaches 98.65%. The Shen embroidery image recognition software designed in this paper also shows high recognition accuracy and has wide promotion and application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is a schematic diagram of the overall process of an application method for classification and recognition of intangible cultural heritage images proposed in the present invention.

[0053] Figure 2 This is a flow chart of step S1 of an application method for classification and recognition of intangible cultural heritage images proposed in the present invention.

[0054] Figure 3 This is a flow chart of step S2 of an application method for classification and recognition of intangible cultural heritage images proposed in the present invention.

[0055] Figure 4 This is a flow chart of step S3 of an application method for classification and recognition of intangible cultural heritage images proposed in the present invention.

[0056] Figure 5 This is a flow chart of step S4 of an application method for classification and recognition of intangible cultural heritage images proposed in the present invention.

[0057] Figure 6 This is a flow chart of step S5 of an application method for classification and recognition of intangible cultural heritage images proposed in the present invention.

[0058] Figure 7 This is a flow chart of step S6 of an application method for classification and recognition of intangible cultural heritage images proposed in the present invention.

[0059] Figure 8 This is a flow chart of step S7 of an application method for classification and recognition of intangible cultural heritage images proposed in the present invention.

[0060] Fig. 9 This is a software function diagram of an application method for intangible cultural heritage image classification and recognition proposed by the present invention. DETAILED DESCRIPTION

[0061] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application usually described and marked in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0062] The present invention discloses an application method for classification and recognition of intangible cultural heritage images.

[0063] As shown in the accompanying drawings, an application method for classification and recognition of intangible cultural heritage images includes:

[0064] S1. Obtain the initial intangible cultural heritage image dataset, including Shen embroidery and related intangible cultural heritage images collected from museums, public datasets or through web crawlers, ensure that the images are clear and complete, standardize the image format, uniformly adjust the resolution to 224×224 and save it in jpg format, increase image diversity through data enhancement methods such as horizontal flipping, rotation, brightness and contrast adjustment, generate about 15 enhanced versions for each original image, and finally expand the dataset to 18,960 images to increase the amount of training data;

[0065] The step S1 can be refined as follows:

[0066] S11, obtaining an initial intangible cultural heritage image dataset, wherein the dataset includes Shen embroidery and other related intangible cultural heritage image data collected from museums, public datasets or through web crawler technology, to ensure that the images have high fidelity and integrity;

[0067] S12, standardize the format of the initial intangible cultural heritage image data, adjust all images to a uniform resolution of 224×224, and save them in a uniform jpg format to ensure that the subsequent training model can smoothly accept the data;

[0068] S13, expanding the data set by using data enhancement technology, wherein the data enhancement processing includes operations such as horizontal flipping, vertical flipping, image rotation, brightness adjustment, and contrast adjustment of the image to simulate different image shooting conditions and improve the diversity of the data set;

[0069] S14. After data augmentation, each original image generates up to 15 different variant images through augmentation technology, and finally the dataset is expanded to 18,960 images to significantly increase the amount of training data.

[0070] S2. MobileNetV3 is selected as the basic architecture, and its lightweight and efficient performance is used to meet the needs of intangible cultural heritage image recognition. The spatial pyramid pooling (SPP) module is introduced to enhance the recognition of image features of different scales. The model parameters are reduced through deep separable convolution while maintaining computational efficiency to adapt to the limited resources of mobile devices. The optimized model structure is used for efficient classification and recognition of intangible cultural heritage image features.

[0071] The step S2 can be refined as follows:

[0072] S21. Select MobileNetV3 as the basic architecture of the recognition network, and use its lightweight advantages and high computing efficiency on mobile devices to meet the needs of intangible cultural heritage image recognition and ensure the efficient operation of the system;

[0073] S22. Based on MobileNetV3, the Spatial Pyramid Pooling (SPP) module is introduced to enhance the network's ability to recognize features of images of different scales and improve the recognition effect of multi-level and multi-texture non-legacy images.

[0074] S23. Optimize the design of convolutional layers in the network, reduce the number of model parameters by introducing depthwise separable convolution, while ensuring the computational efficiency of the model to adapt to the situation where mobile device resources are limited;

[0075] Convolutional neural networks achieve recognition and classification tasks by learning features in images, so a large number of image datasets are required for training. However, intangible cultural heritage images usually lack obvious texture features such as shape and color, and are unlabeled data, making them difficult to process directly. In addition, the cost of collecting and labeling these datasets is very high [Fan Xiangpeng, Xu Yan, Zhou Jianping, et al. Grape leaf disease detection system based on transfer learning and improved CNN [J]. Transactions of the Chinese Society of Agricultural Engineering, 2021, 37(6): 151-159.]], which further increases the difficulty of model training. Therefore, this paper uses transfer learning methods to help the network learn a richer and more generalized intangible cultural heritage feature representation, thereby reducing the risk of overfitting of the network model and further improving the recognition accuracy and robustness of the model.

[0076] Transfer learning means that by fully training the network on a large data set, the network learns a large number of features required for image classification and recognition, and then applies the learned features to new learning tasks. The trained model can achieve good recognition and classification effects after simple training, and can assist the network in generalization and accelerate convergence. During transfer learning, the top few layers of the pre-trained model can be unfrozen by fine-tuning the model, so that the trained features are closer to the current task.

[0077] S24. Use the optimized model structure to classify and identify the initially extracted intangible cultural heritage image features to ensure the model's efficient extraction and classification of intangible cultural heritage image features.

[0078] The network structure of MobileNet V3 is shown in the following table, which mainly includes four parts: basic convolution module, inverted residual module, squeeze-and-excitation module, and hard-swish activation function. There are 11 bottleneck layers, one standard convolution layer, and two point-by-point convolution layers in the network. Among them, the squeeze-and-excitation (SE) module and h-swish (HS) activation function are introduced to improve the accuracy of the model. The last two layers of point-by-point convolution do not use batch normalization (Batch Norm), and the NBN logo is used in the MobileNetV3 structure diagram. In the table, Relu (RE) is a common activation function, and different activation functions are used in different convolution layers. The following table shows the network structure of MobileNet V3:

[0079]

[0080] S3. Use transfer learning, take the MobileNetV3 pre-trained model trained on large data sets (such as ImageNet) as the basis, reduce the initial training time, fine-tune the top layer of the pre-trained model, retrain on the features of intangible cultural heritage images to adapt to the features of intangible cultural heritage images such as Shen embroidery, import the Shen embroidery image dataset for training, improve recognition accuracy and model robustness, reduce the risk of overfitting, and speed up model convergence;

[0081] The step S3 can be refined as follows:

[0082] S31. Use the transfer learning method to use the MobileNetV3 pre-trained model trained on large public datasets (such as ImageNet) as the basic model to reduce the training time of the model in the initial stage;

[0083] S32, unfreeze and fine-tune the top layer of the pre-trained model, and retrain the network weights according to the specific characteristics of the intangible cultural heritage images to ensure that the model can better adapt to the feature representation of intangible cultural heritage images such as Shen embroidery;

[0084] S33, importing the Shen embroidery image dataset into the model for transfer learning training, using the generalized features learned by the pre-training model to improve the accuracy and robustness of intangible cultural heritage images in recognition tasks;

[0085] S34. Accelerate the convergence of the model through transfer learning, reduce the risk of model overfitting, and further train and verify the intangible cultural heritage images after fine-tuning the model to improve the overall performance of the model.

[0086] S4. Use the enhanced intangible cultural heritage image dataset to divide the training set and validation set into 9:1. The training set is used for model training, and the validation set is used to monitor model performance and adjust parameters. Perform multiple training iterations through the optimized MobileNetV3 model, adjust hyperparameters such as learning rate and batch size to ensure efficient model convergence and avoid gradient problems. Finally, use the test set to evaluate the model to ensure its accuracy and robustness in intangible cultural heritage image recognition.

[0087] The evaluation index of this model is accuracy, and the formula is as follows:

[0088]

[0089] The step S4 can be refined as follows:

[0090] S41, using the enhanced intangible cultural heritage image dataset, dividing the training set and the validation set into a ratio of 9:1, the training set is used for model training, and the validation set is used to monitor the performance of the model during the training process and adjust the model parameters;

[0091] S42. Use the optimized MobileNetV3 improved model, perform model training on the training set, and adjust the network parameters through multiple training iterations to find the best parameter combination;

[0092] S43. During model training, adjust hyperparameters such as learning rate and batch size to avoid gradient vanishing or exploding problems and ensure that the model can converge efficiently.

[0093] S44. After training, a separate test set is used to conduct a final evaluation of the model to ensure its accuracy and robustness in the intangible cultural heritage image recognition task.

[0094] S5. Shen Embroidery Intangible Heritage Image Recognition Software developed using PyQt5 library has a human-computer interaction interface, supports users to import images in jpg, jpeg, and png formats, and automatically adjusts to 256×256 size. The software calculates the pixel ratio of image recognition, displays the recognition results, and divides them into 5 levels, where level 1 is a recognition rate of 95%-100%, and the lowest is level 5 (less than 80%);

[0095] The step S5 can be refined as follows:

[0096] S51. Develop a Shen embroidery intangible cultural heritage image recognition software system based on Python language. The system uses the PyQt5 library to design a human-computer interaction interface to achieve a user-friendly operation experience.

[0097] S52. The software has image input function, allowing users to import local files. Supported image formats include jpg, jpeg, png, etc. The software will automatically convert the input image to the standard size of 256×256;

[0098] S53, realizing the image recognition function, the system calculates the ratio of the total pixel value of the recognized category to the pixel value of the entire image, gives the accuracy of image recognition, and displays the specific recognition result on the interface;

[0099] S54. The software is divided into five recognition levels according to the recognition accuracy, namely level 1 (recognition rate 95%-100%), level 2 (90%-95%), level 3 (85%-90%), level 4 (80%-85%), and level 5 (less than 80%).

[0100] S6. Build a test library containing 100 Shen embroidery images and 20 non-Shen embroidery images, test the recognition accuracy and robustness of the software under different input conditions, analyze the recognition results through confusion matrix, evaluate the accuracy, precision and recall of the model, ensure that it can accurately distinguish Shen embroidery from non-Shen embroidery images, verify the compatibility and stability of the software under different resolutions and lighting conditions, adjust the model parameters according to the test results, ensure that the recognition accuracy reaches more than 98%, and optimize the interface and user experience;

[0101] The step S6 can be refined as follows:

[0102] S61, constructing a test image library including positive examples and negative examples, wherein the test image library includes 100 Shen embroidery images and 20 non-Shen embroidery images, and testing the recognition accuracy and robustness of the software under different image input conditions;

[0103] S62. During the test, the confusion matrix is ​​used to analyze the recognition results, and the recognition accuracy, precision and recall of the model are checked to ensure that the model can accurately distinguish between Shen embroidery images and non-Shen embroidery images;

[0104] S63. Verify the compatibility and stability of the software for input images of different resolutions and different lighting conditions to ensure the effectiveness and accuracy of the system in practical applications;

[0105] S64. Adjust model parameters according to test results to ensure recognition accuracy above 98%, optimize the user experience of the software interface, and improve interactive performance.

[0106] S7. Apply the improved MobileNetV3 model to the recognition of other arts and crafts intangible cultural heritage images, verify the generalization ability and adaptability of the model, expand the data set of the Shen embroidery recognition system, add more intangible cultural heritage images, improve the recognition range and application value, help the digital protection of intangible cultural heritage, optimize the mobile application version, achieve cross-platform support, and allow users to perform intangible cultural heritage image recognition at any time on smartphones, tablets and other devices;

[0107] The step S7 can be refined as follows:

[0108] S71. Expand and apply the improved image recognition model based on MobileNetV3 to the recognition of other arts and crafts intangible cultural heritage images to further verify the generalization ability and adaptability of the model;

[0109] S72. Expand the data set of the Shen Embroidery Recognition System, add more types of intangible cultural heritage image resources, improve the recognition scope and application value of the system, and further promote the digital protection of intangible cultural heritage;

[0110] S73. Optimize the mobile application version of Shen Embroidery Image Recognition Software to achieve cross-platform support. Users can perform intangible cultural heritage image recognition operations anytime and anywhere through smart phones, tablet computers and other devices.

[0111] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. The replacement may be a replacement of a part of the structure, device, method step, or a complete technical solution. Any equivalent replacement or change according to the technical solution and the inventive concept of the present invention shall be covered within the protection scope of the present invention.

Claims

1. An application method for classification and recognition of intangible cultural heritage images, characterized in that: include: S1. Obtain the initial intangible cultural heritage image dataset, including Shen embroidery and related intangible cultural heritage images collected from museums, public datasets or through web crawlers, ensure that the images are clear and complete, standardize the image format, uniformly adjust the resolution to 224×224 and save it in jpg format, increase image diversity through data enhancement methods such as horizontal flipping, rotation, brightness and contrast adjustment, generate about 15 enhanced versions for each original image, and finally expand the dataset to 18,960 images to increase the amount of training data; S2. MobileNetV3 is selected as the basic architecture, and its lightweight and efficient performance is used to meet the needs of intangible cultural heritage image recognition. The spatial pyramid pooling (SPP) module is introduced to enhance the recognition of image features of different scales. The model parameters are reduced through deep separable convolution while maintaining computational efficiency to adapt to the limited resources of mobile devices. The optimized model structure is used for efficient classification and recognition of intangible cultural heritage image features. S3. Use transfer learning, take the MobileNetV3 pre-trained model trained on large data sets (such as ImageNet) as the basis, reduce the initial training time, fine-tune the top layer of the pre-trained model, retrain on the features of intangible cultural heritage images to adapt to the features of intangible cultural heritage images such as Shen embroidery, import the Shen embroidery image dataset for training, improve recognition accuracy and model robustness, reduce the risk of overfitting, and speed up model convergence; S4. Use the enhanced intangible cultural heritage image dataset to divide the training set and validation set into 9:

1. The training set is used for model training, and the validation set is used to monitor model performance and adjust parameters. Perform multiple training iterations through the optimized MobileNetV3 model, adjust hyperparameters such as learning rate and batch size to ensure efficient model convergence and avoid gradient problems. Finally, use the test set to evaluate the model to ensure its accuracy and robustness in intangible cultural heritage image recognition. S5. Shen Embroidery Intangible Heritage Image Recognition Software developed using PyQt5 library has a human-computer interaction interface, supports users to import images in jpg, jpeg, and png formats, and automatically adjusts to 256×256 size. The software calculates the pixel ratio of image recognition, displays the recognition results, and divides them into 5 levels, where level 1 is a recognition rate of 95%-100%, and the lowest is level 5 (less than 80%); S6. Build a test library containing 100 Shen embroidery images and 20 non-Shen embroidery images, test the recognition accuracy and robustness of the software under different input conditions, analyze the recognition results through confusion matrix, evaluate the accuracy, precision and recall of the model, ensure that it can accurately distinguish Shen embroidery from non-Shen embroidery images, verify the compatibility and stability of the software under different resolutions and lighting conditions, adjust the model parameters according to the test results, ensure that the recognition accuracy reaches more than 98%, and optimize the interface and user experience; S7. Apply the improved MobileNetV3 model to the recognition of other arts and crafts intangible cultural heritage images, verify the generalization ability and adaptability of the model, expand the data set of the Shen embroidery recognition system, add more intangible cultural heritage images, improve the recognition range and application value, assist in the digital protection of intangible cultural heritage, optimize the mobile application version, achieve cross-platform support, and allow users to perform intangible cultural heritage image recognition at any time on smartphones, tablets and other devices.

2. The application method of intangible cultural heritage image classification and recognition according to claim 1 is characterized in that: The step S1 can be refined as follows: S11, obtaining an initial intangible cultural heritage image dataset, wherein the dataset includes Shen embroidery and other related intangible cultural heritage image data collected from museums, public datasets or through web crawler technology, to ensure that the images have high fidelity and integrity; S12, standardize the format of the initial intangible cultural heritage image data, adjust all images to a uniform resolution of 224×224, and save them in a uniform jpg format to ensure that the subsequent training model can smoothly accept the data; S13, expanding the data set by using data enhancement technology, wherein the data enhancement processing includes operations such as horizontal flipping, vertical flipping, image rotation, brightness adjustment, and contrast adjustment of the image to simulate different image shooting conditions and improve the diversity of the data set; S14. After data augmentation, each original image generates up to 15 different variant images through augmentation technology, and finally the dataset is expanded to 18,960 images to significantly increase the amount of training data.

3. The application method of intangible cultural heritage image classification and recognition according to claim 1 is characterized in that: The step S2 can be refined as follows: S21. Select MobileNetV3 as the basic architecture of the recognition network, and use its lightweight advantages and high computing efficiency on mobile devices to meet the needs of intangible cultural heritage image recognition and ensure the efficient operation of the system; S22. Based on MobileNetV3, the Spatial Pyramid Pooling (SPP) module is introduced to enhance the network's ability to recognize features of images of different scales and improve the recognition effect of multi-level and multi-texture non-legacy images. S23. Optimize the design of convolutional layers in the network, reduce the number of model parameters by introducing depthwise separable convolution, while ensuring the computational efficiency of the model to adapt to the situation where mobile device resources are limited; S24. Use the optimized model structure to classify and identify the initially extracted intangible cultural heritage image features to ensure the model's efficient extraction and classification of intangible cultural heritage image features.

4. The application method of intangible cultural heritage image classification and recognition according to claim 1 is characterized in that: The step S3 can be refined as follows: S31. Use the transfer learning method to use the MobileNetV3 pre-trained model trained on large public datasets (such as ImageNet) as the basic model to reduce the training time of the model in the initial stage; S32, unfreeze and fine-tune the top layer of the pre-trained model, and retrain the network weights according to the specific characteristics of the intangible cultural heritage images to ensure that the model can better adapt to the feature representation of intangible cultural heritage images such as Shen embroidery; S33, importing the Shen embroidery image dataset into the model for transfer learning training, using the generalized features learned by the pre-training model to improve the accuracy and robustness of intangible cultural heritage images in recognition tasks; S34. Accelerate the convergence of the model through transfer learning, reduce the risk of model overfitting, and further train and verify the intangible cultural heritage images after fine-tuning the model to improve the overall performance of the model.

5. The application method of intangible cultural heritage image classification and recognition according to claim 1 is characterized in that: The step S4 can be refined as follows: S41, using the enhanced intangible cultural heritage image dataset, dividing the training set and the validation set into a ratio of 9:1, the training set is used for model training, and the validation set is used to monitor the performance of the model during the training process and adjust the model parameters; S42. Use the optimized MobileNetV3 improved model, perform model training on the training set, and adjust the network parameters through multiple training iterations to find the best parameter combination; S43. During model training, adjust hyperparameters such as learning rate and batch size to avoid gradient vanishing or exploding problems and ensure that the model can converge efficiently. S44. After training, a separate test set is used to conduct a final evaluation of the model to ensure its accuracy and robustness in the intangible cultural heritage image recognition task.

6. The application method of intangible cultural heritage image classification and recognition according to claim 1 is characterized in that: The step S5 can be refined as follows: S51. Develop a Shen embroidery intangible cultural heritage image recognition software system based on Python language. The system uses the PyQt5 library to design a human-computer interaction interface to achieve a user-friendly operation experience. S52. The software has image input function, allowing users to import local files. Supported image formats include jpg, jpeg, png, etc. The software will automatically convert the input image to the standard size of 256×256; S53, realizing the image recognition function, the system calculates the ratio of the total pixel value of the recognized category to the pixel value of the entire image, gives the accuracy of image recognition, and displays the specific recognition result on the interface; S54. The software is divided into five recognition levels according to the recognition accuracy, namely level 1 (recognition rate 95%-100%), level 2 (90%-95%), level 3 (85%-90%), level 4 (80%-85%), and level 5 (less than 80%).

7. The application method of intangible cultural heritage image classification and recognition according to claim 1 is characterized in that: The step S6 can be refined as follows: S61, constructing a test image library including positive examples and negative examples, wherein the test image library includes 100 Shen embroidery images and 20 non-Shen embroidery images, and testing the recognition accuracy and robustness of the software under different image input conditions; S62. During the test, the confusion matrix is ​​used to analyze the recognition results, and the recognition accuracy, precision and recall of the model are checked to ensure that the model can accurately distinguish between Shen embroidery images and non-Shen embroidery images; S63. Verify the compatibility and stability of the software for input images of different resolutions and different lighting conditions to ensure the effectiveness and accuracy of the system in practical applications; S64. Adjust model parameters according to test results to ensure recognition accuracy above 98%, optimize the user experience of the software interface, and improve interactive performance.

8. The application method of intangible cultural heritage image classification and recognition according to claim 1 is characterized in that: The step S7 can be refined as follows: S71. Expand and apply the improved image recognition model based on MobileNetV3 to the recognition of other arts and crafts intangible cultural heritage images to further verify the generalization ability and adaptability of the model; S72. Expand the data set of the Shen Embroidery Recognition System, add more types of intangible cultural heritage image resources, improve the recognition scope and application value of the system, and further promote the digital protection of intangible cultural heritage; S73. Optimize the mobile application version of Shen Embroidery Image Recognition Software to achieve cross-platform support. Users can perform intangible cultural heritage image recognition operations anytime and anywhere through smart phones, tablet computers and other devices.

Images