A Smart System for Determining the Regional Origin of Shang and Zhou Dynasty Animal Mask Patterns
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-04-10
- Publication Date
- 2026-08-14
AI Technical Summary
[0003]传统类型学分析依赖人工经验判定兽面纹出土地域，需要研究人员深度掌握兽面纹相关信息，不仅主观性强，而且效率低，无法批量处理样本
[0030]本申请的智能判定系统设置了包含兽面纹识别模型和兽面纹数据库的AI识别与判定模块，用户仅需上传未知兽面纹后，通过兽面纹识别模型能准确判断出对应的出土地域和对应概率，实现快速交互的目的，降低兽面纹地域溯源的研究门槛，解决了传统类型学分析依赖人工经验判定兽面纹出土地域，造成主观性强和效率低的问题。另外，兽面纹识别模型通过基于resnet50的地域分类模型构建方法创建生成，并同步生成兽面纹数据库，兽面纹数据库提供基础信息、纹饰特征、铸造特征和出土地标签的关联管理，支撑识别模型的训练，以及向用户展示。而基于resnet50的地域分类模型构建方法通过人工标注兽面纹信息和分型分式标准，以ResNet50视觉特征提取模型为基础，使验证通过后的兽面纹识别模型能高效提取兽面纹角、目、鼻/扉棱等局部细微特征与全局结构特征，提高识别准确度，满足多形态纹饰特征学习要求，且兽面纹识别模型可以输出0-1区间概率值，进而可直接输出「出土地域+对应置信度」，快速且明确的让用户了解未知兽面纹的出土地域。
Smart Images
Figure CN122574818A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of cultural relic digitization technology, and in particular to an intelligent system for determining the geographical origin of Shang and Zhou dynasty animal mask patterns. Background Technology
[0002] The animal mask pattern, also known as the taotie pattern, is one of the most representative decorative patterns on bronzes from the Shang and Zhou dynasties in China. It is a mysterious pattern based on the frontal image of an animal (mainly its head), which has been highly artistic, abstract, and symmetrical.
[0003] Traditional typological analysis relies on human experience to determine the geographical origin of animal-face patterns. This requires researchers to have a deep understanding of the relevant information about animal-face patterns. It is not only highly subjective but also inefficient and cannot process samples in batches. Summary of the Invention
[0004] In view of the above-mentioned defects or deficiencies in the existing technology, it is desirable to provide an intelligent system for determining the geographical origin of Shang and Zhou dynasty animal mask patterns.
[0005] This invention provides an intelligent system for determining the geographical origin of Shang and Zhou dynasty animal mask patterns, comprising:
[0006] The data processing unit includes a preprocessing module and an AI recognition and judgment module;
[0007] User units include operation modules and display modules;
[0008] The preprocessing module is used to process unknown animal face patterns and obtain standardized images;
[0009] The AI recognition and judgment module includes an animal face pattern recognition model and an animal face pattern database. The animal face pattern recognition model and the animal face pattern database are created and generated using a regional classification model construction method based on ResNet50. The animal face pattern database includes animal face pattern images and corresponding animal face pattern information. The animal face pattern information includes basic information, decorative features, casting features, and excavation site labels. The animal face pattern recognition model is used to recognize standardized images and output judgment results. The judgment results include the excavation site region of unknown animal face patterns and the corresponding probability.
[0010] The operation module is used by the operator to upload unknown beast-face patterns and send them to the preprocessing module;
[0011] The display module is used to show the judgment results to the user and display the categorizable animal face pattern information.
[0012] According to the technical solution provided in the embodiments of this application, the method for constructing a regional classification model based on ResNet50 specifically includes the following steps:
[0013] S100, Data Acquisition and Production: Acquire images of animal face patterns and manually annotate animal face pattern information;
[0014] S200, Typological Processing: Based on the decorative features, the animal face pattern image is segmented, and classification standards are established;
[0015] S300. Constructing a database: Associate the animal face pattern images and animal face pattern information, store them in a structured database, and form an animal face pattern database;
[0016] S400, Image Processing: Perform image denoising, geometric calibration, edge extraction, and size normalization on the animal face pattern image to form a standardized image dataset;
[0017] S500. Dataset partitioning: Combine and associate the standardized image dataset with the corresponding animal face pattern information, and divide it into training set, validation set, and test set. The training set is used for model parameter training and updating, the validation set is used for hyperparameter tuning during training, and the test set is used to evaluate the final performance of the model.
[0018] S600, Model Training: Based on the ResNet50 visual feature extraction model, freeze all pre-trained parameters of the model backbone network, replace the last fully connected layer of the model with the classifier head adapted for land classification, set training parameters, use the training set as trainable parameters for iterative training, calculate the error through the loss function and backpropagate to update the trainable parameters, and generate the model to be verified.
[0019] S700, Model Performance Evaluation: Use the test set to validate the trained model to be validated. If the validation is successful, output the model to be validated as the animal face pattern recognition model and save the model weight and label mapping file; otherwise, return to step S500, adjust the dataset partition ratio and / or training parameters, and retrain.
[0020] According to the technical solution provided in the embodiments of this application, the training parameters include the number of training rounds, the learning rate, and the batch size.
[0021] According to the technical solution provided in the embodiments of this application, in step S600, the forward propagation formula of the ResNet50 visual feature extraction model is as follows: x_{l+1}=x_l+F(x_l,W_l)
[0022] Where x_l is the input feature of the l-th layer network, F(x_l,W_l) is the residual function (composed of convolutional layers, activation functions, etc.), and x_{l+1} is the output feature of the (l+1)-th layer network.
[0023] According to the technical solution provided in the embodiments of this application, in step S600, the loss function adopted is the cross-entropy loss function, and the formula is as follows:
[0024] Where N is the batch sample size, and C is the number of land excavation label categories. The true label for sample i (one-hot encoding). Predict the probability that sample i belongs to class c for the model.
[0025] According to the technical solution provided in the embodiments of this application, the verification parameters include the judgment response time, the judgment accuracy, and the small sample recognition accuracy.
[0026] According to the technical solution provided in the embodiments of this application, the preprocessing module includes image denoising, geometric calibration, edge extraction and size normalization to form a standardized image.
[0027] According to the technical solution provided in the embodiments of this application, the operation module includes a mini-program, a mobile application, a website application, and a computer application.
[0028] According to the technical solution provided in the embodiments of this application, the display module is also used to display animal face pattern images and corresponding animal face pattern information, and to classify and display them according to basic information, pattern features, casting features and excavation site labels.
[0029] Compared with the prior art, the beneficial effects of the present invention are:
[0030] The intelligent determination system of this application includes an AI recognition and determination module comprising an animal-face pattern recognition model and an animal-face pattern database. Users only need to upload an unknown animal-face pattern, and the animal-face pattern recognition model can accurately determine the corresponding excavation region and probability, achieving rapid interaction and lowering the research threshold for tracing the regional origin of animal-face patterns. This solves the problems of high subjectivity and low efficiency caused by traditional typological analysis relying on manual experience to determine the excavation region of animal-face patterns. Furthermore, the animal-face pattern recognition model is created and generated using a ResNet50-based regional classification model construction method, and simultaneously generates an animal-face pattern database. The animal-face pattern database provides associated management of basic information, decorative features, casting features, and excavation region labels, supporting the training of the recognition model and its display to users. The regional classification model construction method based on ResNet50 uses manually labeled animal face pattern information and classification standards. Based on the ResNet50 visual feature extraction model, the validated animal face pattern recognition model can efficiently extract local subtle features such as horns, eyes, nose / front ridges and global structural features of animal face patterns, improve recognition accuracy, meet the requirements of multi-morphological pattern feature learning, and the animal face pattern recognition model can output probability values in the range of 0-1, and then directly output "originating region + corresponding confidence level", so as to quickly and clearly let users understand the originating region of unknown animal face patterns.
[0031] It should be understood that the description in the Summary of the Invention is not intended to limit the key or essential features of the embodiments of the present invention, nor is it intended to restrict the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description
[0032] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:
[0033] Figure 1 A structural block diagram of an intelligent system for determining the geographical origin of Shang and Zhou dynasty animal mask patterns provided in this application embodiment;
[0034] Figure 2 A flowchart illustrating the steps involved in constructing a regional classification model based on ResNet50. Detailed Implementation
[0035] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.
[0036] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.
[0037] Please refer to Figures 1-2 The present invention provides an intelligent system for determining the geographical origin of Shang and Zhou dynasty animal mask patterns, comprising:
[0038] The data processing unit includes a preprocessing module and an AI recognition and judgment module;
[0039] User units include operation modules and display modules;
[0040] The preprocessing module is used to process unknown animal face patterns to obtain a standardized image. The preprocessing module includes image denoising, geometric calibration, edge extraction, and size normalization to form a standardized image. The purpose is to ensure image clarity, ensure recognition accuracy, and avoid the reduction in recognition accuracy caused by image noise and blurring.
[0041] The AI recognition and judgment module includes an animal face pattern recognition model and an animal face pattern database. The animal face pattern recognition model and the animal face pattern database are created and generated using a regional classification model construction method based on ResNet50. The animal face pattern database includes animal face pattern images and corresponding animal face pattern information. The animal face pattern information includes basic information, decorative features, casting features, and excavation site labels. The animal face pattern recognition model is used to recognize standardized images and output judgment results. The judgment results include the excavation site region of unknown animal face patterns and the corresponding probability.
[0042] The operation module is used by the operator to upload unknown animal face patterns and send them to the preprocessing module; the operation module includes a mini-program, mobile application, website and computer application, which improves the convenience of uploading animal face pattern images and facilitates mobile-based determination of the place of origin.
[0043] The display module is used to show the user the judgment results and display the categorizable animal face pattern information. The display module is also used to display animal face pattern images and corresponding animal face pattern information, and to classify and display them according to basic information, pattern features, casting features and excavation site labels.
[0044] The basic information includes the era, size, and holding institution; decorative features include horns, eyes, nose / flanks, mouth, teeth, ears, and body (independent / forked tail / connected / decomposed); casting features include the technique, simplified / unsimplified design, background / auxiliary decorations, and casting quality; and the excavation site label includes provincial / municipal regional dimensions. When displaying the information by category, users can select the information they need, such as the same era, the same mouth, and the same excavation site label, making it easy for users to intuitively search for animal mask images and corresponding other information under the same category.
[0045] When using the intelligent judgment system:
[0046] Users can use the operation module to select one of the following options: a mini-program, mobile application, website, or desktop application. The image data containing the animal face pattern is then transmitted to the preprocessing module. After noise reduction, a clear, standardized image is obtained. This image is then recognized by the animal face pattern recognition model in the AI recognition and judgment module. The model outputs a judgment result by identifying the decorative features in the standardized image and, based on the different decorative features from different excavation sites, directly outputting the excavation site region and corresponding probability of the unknown animal face pattern. Furthermore, the display module can simultaneously show other animal face pattern information from similar images, allowing users to more clearly understand the association information of the unknown animal face pattern.
[0047] In some embodiments, the method for constructing a region classification model based on ResNet50 specifically includes the following steps:
[0048] S100, Data Acquisition and Production: Acquire images of animal face patterns and manually annotate animal face pattern information;
[0049] The animal mask imagery includes both actual images and line drawings, enabling unified determination of the provenance location for various image types and improving the system's compatibility and practicality. Animal mask information includes basic information, decorative features, casting characteristics, and provenance labels. Basic information includes the date, size, and the institution holding the artifact. Decorative features include horns, eyes, nose / flanks, mouth, teeth, ears, and body (independent / forked tail / connected / decomposed). Casting characteristics include the casting process, whether simplified or not, background / auxiliary decorations, and casting quality. Provenance labels include provincial / municipal regional dimensions.
[0050] S200, Typological Processing: Based on the decorative features, the animal face pattern image is segmented, and classification standards are established;
[0051] This system synthesizes the archaeological typological classification of animal-face patterns by archaeologists Ma Chengyuan, Hayashi Minao, and Zhu Fenghan. It is primarily used for user demonstrations and also serves as an academic aid in the study of Shang and Zhou dynasty decorative patterns. Taking the shape of the horns as an example, it includes animal-face patterns with outward-curling horns, inward-curling horns, and giraffe horns. When the user selects horns as the decorative feature, the patterns are displayed independently according to their respective categories (outward-curling horns, inward-curling horns, and giraffe horns, etc.). This facilitates a clear understanding of the corresponding features' shapes and related information. Other decorative features are similarly classified; specific classification standards are not elaborated here.
[0052] S300. Constructing a database: Associate the animal face pattern images and animal face pattern information, store them in a structured database, and form an animal face pattern database;
[0053] In other words, in the animal mask pattern database, each animal mask image will have corresponding basic information, decorative features, casting features, and excavation site labels. Therefore, when calling specific decorative features, other information can be retrieved simultaneously, ensuring data integrity and relevance, and making it easier for users to query and understand certain new information. Optionally, the structured database can be one of PostgreSQL, MongoDB, or MySQL, facilitating direct retrieval of the corresponding data during subsequent training.
[0054] S400, Image Processing: Perform image denoising, geometric calibration, edge extraction, and size normalization on the animal face pattern image to form a standardized image dataset;
[0055] The image denoising can be performed using Gaussian filtering, median filtering, or bilateral filtering; the geometric calibration can be performed using perspective transformation or affine transformation to correct tilt and distortion; the edge extraction can be performed using the Canny operator or the Sobel operator; the size normalization process uses bilinear interpolation to uniformly scale the image to 224×224 pixels to meet the input requirements of deep learning models.
[0056] S500, Dataset partitioning: Combine the standardized image dataset with the corresponding animal face pattern information and divide it into a training set, a validation set, and a test set. The training set is used for training and updating model parameters, the validation set is used for hyperparameter tuning during training, and the test set is used to evaluate the final performance of the model.
[0057] S600, Model Training: Based on the ResNet50 visual feature extraction model, freeze all pre-trained parameters of the model's backbone network, replace the last fully connected layer of the model with a classifier head adapted for land classification, set training parameters, and iteratively train using the training set as trainable parameters. Calculate the error through the loss function and backpropagate to update the trainable parameters to generate a model to be validated. The training parameters include the number of training epochs, learning rate, and batch size. Initially, the number of training epochs is set to 12, the learning rate to 1e-4, and the batch size to 8.
[0058] S700, Model Performance Evaluation: Use the test set to validate the trained model to be validated. If the validation is successful, output the model to be validated as the animal face pattern recognition model and save the model weight and label mapping file; otherwise, return to step S500, adjust the dataset partition ratio and / or training parameters, and retrain.
[0059] Among them, the model weights can be loaded into another model structure of the same type, so that repeated training is not required. The label mapping file is a conversion of digital language. That is, the actual output of the animal face pattern recognition model is numbers. Loading the label mapping file can output the corresponding actual capabilities. The combination of the two makes it convenient to redeploy the animal face pattern recognition model.
[0060] The validation parameters include decision response time, decision accuracy, and small sample recognition accuracy.
[0061] Judgment response time: From the time the user uploads the unknown beast face pattern to the display of the judgment result, the response time should be ≤100ms to ensure that the user receives the correct feedback quickly.
[0062] Accuracy of land discovery determination: The overall accuracy of determination within the test set is ≥85%, and the accuracy of determination of core land discovery areas is ≥90%. If the accuracy of determination is too low, it will cause errors in candidate identification and reduce feasibility.
[0063] Small sample recognition accuracy: For small sample scenarios with less than 30 samples in a single region, the regional feature recognition and judgment accuracy is ≥90%. Small sample has a small number of samples and is prone to local optima. Therefore, it is necessary to meet the judgment accuracy requirements to ensure recognition accuracy.
[0064] In some embodiments, in step S600, the forward propagation formula of the ResNet50 visual feature extraction model is as follows: x_{l+1}=x_l+F(x_l,W_l)
[0065] Where x_l is the input feature of the l-th layer network, F(x_l,W_l) is the residual function (composed of convolutional layers, activation functions, etc.), and x_{l+1} is the output feature of the (l+1)-th layer network.
[0066] The model's forward propagation formula is adapted to the needs of ornamentation recognition. Relying on the ResNet50 residual structure, it can efficiently extract local subtle features such as horns, eyes, nose / frontal ridges, and global structural features of animal face patterns, meeting the requirements for learning features of multiple ornamentation forms. The formula's operation logic is concise, relying on residual connections to solve the gradient vanishing problem in deep networks, resulting in fast training convergence and low inference latency, adapting to the performance requirements of real-time recognition on web pages. The feature weights are highly interpretable, with W_l corresponding to the ornamentation feature learning weights. The model can be matched with the typological classification standards of animal face patterns through feature visualization, facilitating the verification of the model's rationality in archaeological scenarios. After using this formula, the model has good image compatibility, supporting multiple types of image inputs such as animal face pattern images and line drawings, and can automatically adapt to images of different resolutions, reducing preprocessing complexity.
[0067] In some embodiments, in step S600, the loss function is the cross-entropy loss function, as shown in the following formula:
[0068] Where N is the batch sample size, and C is the number of land excavation label categories. The true label for sample i (one-hot encoding). Predict the probability that sample i belongs to class c for the model.
[0069] The cross-entropy loss function is specifically designed for multi-classification tasks, directly handling the determination of land origin labels across multiple regions. It effectively distinguishes similar features of animal-face patterns from different regions, improving the accuracy of core region determination. It optimizes the difference between predicted probabilities and true labels, with the model outputting probability values in the 0-1 range, directly outputting "land origin region + corresponding confidence level," matching the system's data output business logic. The gradient characteristics are adapted for lightweight fine-tuning, with large gradients and fast convergence when the predicted probability is low, and small gradients and overfitting prevention when the probability is high, which is highly compatible with the "frozen backbone, training only the classification head" training scheme. Batch mean calculation improves robustness, avoiding the influence of extreme values of a single sample on the training direction, adapting to scenarios with small land origin samples, and enhancing the model's generalization ability. The computation has no additional overhead, only including basic summation, logarithmic and multiplication operations, and does not increase the time consumption for training and inference, forming a synergy with the efficient inference of ResNet50.
[0070] In the description of this specification, the terms "connection," "installation," and "fixing," etc., should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.
[0071] In the description of this specification, the terms "one embodiment," "some embodiments," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.
[0072] The above are merely preferred embodiments of this application and are not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.
Claims
1. An intelligent system for determining the geographical origin of Shang and Zhou dynasty animal mask patterns, characterized in that, include: The data processing unit includes a preprocessing module and an AI recognition and judgment module; User units include operation modules and display modules; The preprocessing module is used to process unknown animal face patterns and obtain standardized images; The AI recognition and judgment module includes an animal face pattern recognition model and an animal face pattern database. The animal face pattern recognition model and the animal face pattern database are created and generated using a regional classification model construction method based on ResNet50. The animal face pattern database includes animal face pattern images and corresponding animal face pattern information. The animal face pattern information includes basic information, decorative features, casting features, and excavation site labels. The animal face pattern recognition model is used to recognize standardized images and output judgment results. The judgment results include the excavation site region of unknown animal face patterns and the corresponding probability. The operation module is used by the operator to upload unknown beast-face patterns and send them to the preprocessing module; The display module is used to show the judgment results to the user and display the categorizable animal face pattern information.
2. The intelligent system for determining the geographical origin of Shang and Zhou dynasty animal mask patterns according to claim 1, characterized in that, The method for constructing a regional classification model based on ResNet50 specifically includes the following steps: S100, Data Acquisition and Production: Acquire images of animal face patterns and manually annotate animal face pattern information; S200, Typological Processing: Based on the decorative features, the animal face pattern image is segmented, and classification standards are established; S300. Constructing a database: Associate the animal face pattern images and animal face pattern information, store them in a structured database, and form an animal face pattern database; S400, Image Processing: Perform image denoising, geometric calibration, edge extraction, and size normalization on the animal face pattern image to form a standardized image dataset; S500. Dataset partitioning: Combine and associate the standardized image dataset with the corresponding animal face pattern information, and divide it into training set, validation set, and test set. The training set is used for model parameter training and updating, the validation set is used for hyperparameter tuning during training, and the test set is used to evaluate the final performance of the model. S600, Model Training: Based on the ResNet50 visual feature extraction model, freeze all pre-trained parameters of the model backbone network, replace the last fully connected layer of the model with the classifier head adapted for land classification, set training parameters, use the training set as trainable parameters for iterative training, calculate the error through the loss function and backpropagate to update the trainable parameters, and generate the model to be verified. S700, Model Performance Evaluation: Use the test set to validate the trained model to be validated. If the validation is successful, output the model to be validated as the animal face pattern recognition model and save the model weight and label mapping file; otherwise, return to step S500, adjust the dataset partition ratio and / or training parameters, and retrain.
3. The intelligent system for determining the geographical origin of Shang and Zhou dynasty animal mask patterns according to claim 2, characterized in that, The training parameters include the number of training epochs, the learning rate, and the batch size.
4. The intelligent system for determining the geographical origin of Shang and Zhou dynasty animal mask patterns according to claim 2, characterized in that, In step S600, the forward propagation formula of the ResNet50 visual feature extraction model is as follows: x_{l+1}=x_l+F(x_l,W_l) Where x_l is the input feature of the l-th layer network, F(x_l,W_l) is the residual function (composed of convolutional layers, activation functions, etc.), and x_{l+1} is the output feature of the (l+1)-th layer network.
5. The intelligent system for determining the geographical origin of Shang and Zhou dynasty animal mask patterns according to claim 2, characterized in that, In step S600, the loss function adopted is the cross-entropy loss function, and the formula is as follows: Where N is the batch sample size, and C is the number of land excavation label categories. The true label for sample i (one-hot encoding). Predict the probability that sample i belongs to class c for the model.
6. The intelligent system for determining the geographical origin of Shang and Zhou dynasty animal mask patterns according to claim 2, characterized in that, The verification parameters include decision response time, decision accuracy, and small sample recognition accuracy.
7. The intelligent system for determining the geographical origin of Shang and Zhou dynasty animal mask patterns according to claim 1, characterized in that, The preprocessing module includes image denoising, geometric calibration, edge extraction, and size normalization to form a standardized image.
8. The intelligent system for determining the geographical origin of Shang and Zhou dynasty animal mask patterns according to claim 1, characterized in that, The operation modules include mini-programs, mobile applications, websites, and desktop applications.
9. The intelligent system for determining the geographical origin of Shang and Zhou dynasty animal mask patterns according to claim 1, characterized in that, The display module is also used to display animal face pattern images and corresponding animal face pattern information, and to classify and display them according to basic information, pattern features, casting features and excavation site labels.