A noise reduction method and related device for digital models of cultural relics

Abstract

The present application discloses a denoising method and related device for a digital model of cultural relics, which relates to the field of artificial intelligence technology. The method includes inputting original feature data into a preset denoising model; performing multiple iterative encodings through the preset denoising model to obtain low-dimensional feature data, performing geometric transformations on the low-dimensional feature data to obtain features to be reconstructed; decoding the features to be reconstructed to obtain reconstructed feature data of the digital model of cultural relics. The iterative encoding includes: performing a linear transformation on the input features of this iterative encoding; performing a projection transformation on the linear transformation results using the corresponding linear weight matrix and nonlinear weight matrix; normalizing the projection transformation results to obtain the activation output of this iterative encoding, and using the activation output of this iterative encoding as the input feature of the next iterative encoding; it can be seen that this scheme improves the accuracy and fidelity of the digital model of cultural relics by combining linear projection transformation and nonlinear projection transformation to project features during the encoding stage.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a noise reduction method and related device for a digital model of cultural relics. Background Art

[0002] Digital models of cultural relics are created by collecting information about a cultural relic's geometry, texture, color, and other features and, using 3D modeling software and techniques, converting this information into computer-processable 3D data models. With the rapid development of cultural relic preservation and digitization technologies, digital models have become a crucial tool for research, restoration, display, and archiving. They can accurately depict the shape, structure, and details of a cultural relic, providing crucial support for its preservation, restoration, display, and research.

[0003] However, due to the limitations of acquisition equipment accuracy, environmental interference, and data processing technology, noise is inevitable in digital models of cultural relics. This noise will significantly affect the accuracy and fidelity of digital models of cultural relics, thereby reducing the visual effect. Summary of the Invention

[0004] In view of the above problems, this application provides a method and related device for reducing noise of digital models of cultural relics to achieve the purpose of improving the accuracy and fidelity of digital models of cultural relics. The specific solution is as follows:

[0005] In a first aspect, the present application provides a method for reducing noise of a digital model of cultural relics, comprising:

[0006] Obtaining original feature data of the digital model of the cultural relic to be denoised;

[0007] Inputting the original feature data into a preset noise reduction model;

[0008] Low-dimensional feature data is obtained by performing multiple iterative encodings through the preset denoising model, and each iterative encoding includes: performing a linear transformation on the input features of this iterative encoding to obtain a linear transformation result; performing a projection transformation on the linear transformation result using a corresponding linear weight matrix and a nonlinear weight matrix to obtain a projection transformation result; normalizing the projection transformation result to obtain an activation output of this iterative encoding, and using the activation output of this iterative encoding as the input feature of the next iterative encoding; wherein the input feature of the first iterative encoding is the original feature data, and the activation output of the last iterative encoding is the low-dimensional feature data;

[0009] Performing geometric transformation on the low-dimensional feature data using the preset denoising model to obtain features to be reconstructed;

[0010] The features to be reconstructed are decoded by the preset noise reduction model to obtain reconstructed feature data of the digital model of the cultural relic.

[0011] In a possible implementation, obtaining original feature data of the digital model of the cultural relic to be denoised includes:

[0012] Collecting original data of the digital model of the cultural relic, and preprocessing the original data of the digital model of the cultural relic using a preset preprocessing method; wherein the preprocessing includes coordinate alignment, defect repair, and texture mapping;

[0013] The pre-processed original data of the digital model of cultural relics are subjected to a preset standardization process to obtain original feature data of the digital model of cultural relics; wherein the standardization process includes redundancy removal, pixel-level segmentation, feature extraction and data enhancement.

[0014] In one possible implementation, the preset noise reduction model includes an encoder, a geometric transformation layer, and a decoder connected in sequence, wherein the encoder includes a plurality of coding layers connected in sequence;

[0015] One encoding layer is used to perform one iterative encoding;

[0016] The geometric transformation layer is used to perform the geometric transformation on the low-dimensional feature data to obtain the features to be reconstructed;

[0017] The encoder is used to perform the decoding of the feature to be reconstructed to obtain the reconstructed feature data of the digital model of the cultural relic.

[0018] In a possible implementation, performing a linear transformation on the input features of this iterative encoding using the preset denoising model to obtain a linear transformation result includes:

[0019] Performing a linear transformation on the input features by using the weights and biases of the target coding layer through the target coding layer to obtain the linear transformation result;

[0020] Among them, the target coding layer is the coding layer used to perform this iterative coding. If the target coding layer is the first coding layer, the input feature of the target coding layer is the original feature data. If the target coding layer is not the first coding layer, the input feature of the target coding layer is the output feature of the previous coding layer of the target coding layer.

[0021] In a possible implementation, performing a projection transformation on the linear transformation result by using the preset denoising model and a corresponding linear weight matrix and a nonlinear weight matrix to obtain a projection transformation result includes:

[0022] The linear transformation result is operated by the target coding layer using the hyperbolic tangent function to obtain a hyperbolic tangent result; the linear weight matrix of the target coding layer is multiplied by the linear transformation result to obtain a linear operation result, the nonlinear weight matrix of the target coding layer is multiplied by the hyperbolic tangent result to obtain a nonlinear operation result, and the linear operation result, the nonlinear operation result and the projection bias of the target coding layer are added to obtain the projection transformation result.

[0023] In a possible implementation, normalizing the projection transformation result using the preset denoising model to obtain an activation output for this iterative encoding includes:

[0024] Performing standard normalization on the projection transformation result through the target coding layer to obtain a standard normalization result; scaling and translating the standard normalization result to obtain the activation output;

[0025] If the target coding layer is not the last coding layer, the activation output is output as the output feature to the next coding layer of the target coding layer through the target coding layer; if the target coding layer is the last coding layer, the activation output is output as the low-dimensional feature data to the geometric transformation layer through the target coding layer.

[0026] In a possible implementation, performing a geometric transformation on the low-dimensional feature data using the preset denoising model to obtain features to be reconstructed includes:

[0027] The geometric transformation layer uses a skew-symmetric transformation function and a scale transformation function to construct a geometric transformation matrix of the low-dimensional feature data; after performing a matrix exponential operation on the geometric transformation matrix, it is multiplied with the low-dimensional feature data to obtain the feature to be reconstructed.

[0028] In one possible implementation, the decoder includes a plurality of decoding layers connected in sequence, and decoding the to-be-reconstructed features through the preset denoising model includes:

[0029] The decoding layer of each layer is sequentially performing a first nonlinear projection on the input feature based on a first nonlinear mapping function to obtain a first mapping feature; the first mapping feature is added to the bias of the decoding layer, and then multiplied by the transpose of the weight of the decoding layer to obtain a second mapping feature; the second mapping feature is nonlinearly projected based on the second nonlinear mapping function to obtain a nonlinear projection result of the decoding layer; wherein, if the decoding layer is the first decoding layer, the input feature of the decoding layer is the feature to be reconstructed; if the decoding layer is not the first decoding layer, the input feature of the decoding layer is the nonlinear projection result of the previous decoding layer;

[0030] The decoder adds the nonlinear projection results of all the decoding layers to obtain the reconstructed feature data.

[0031] In a possible implementation, adding the nonlinear projection results of all the decoding layers by the decoder to obtain the reconstructed feature data includes:

[0032] Sequentially adding the nonlinear projection result of the decoding layer to the feature accumulation result output by the previous decoding layer through each decoding layer to obtain the feature accumulation result of the decoding layer;

[0033] If the decoding layer is not the last decoding layer, outputting the feature accumulation result of the decoding layer to the next decoding layer through the decoding layer;

[0034] If the decoding layer is the last decoding layer, the feature accumulation result of the decoding layer is output as the reconstructed feature data through the decoding layer.

[0035] In a possible implementation, the denoising method for the digital model of cultural relics further includes:

[0036] Acquiring training data, wherein the training data includes original feature data of a plurality of sample cultural relic digital models;

[0037] Iterations are performed until the preset training completion condition is reached. Any iteration is a target iteration, which includes:

[0038] Inputting the original feature data of each of the sample cultural relic digital models into the improved autoencoder to be trained, and obtaining the reconstructed feature data of each of the sample cultural relic digital models output by the improved autoencoder;

[0039] Based on the reconstructed feature data and original feature data of all sample cultural relic digital models of the target iteration, a preset enhanced loss function is used to calculate the loss value of the target iteration; the enhanced loss function is constructed based on an error function and a regularization function that introduces periodic penalties;

[0040] Based on the loss value of the target iteration, the model parameters in the model parameter set are updated using the gradient descent method, wherein the model parameters include the linear weight matrix and the nonlinear weight matrix corresponding to each iteration encoding;

[0041] Determine whether the preset training completion conditions are met;

[0042] If it is reached, the iteration is stopped, and the improved autoencoder is configured with each model parameter in the updated model parameter set to obtain the preset noise reduction model;

[0043] If not reached, execute the next iteration.

[0044] A second aspect of the present application provides a noise reduction device for a digital model of cultural relics, comprising:

[0045] The original data acquisition unit is used to obtain the original feature data of the digital model of the cultural relic to be denoised;

[0046] A model denoising unit is used to input the original feature data into a preset denoising model, and perform multiple iterative encodings through the preset denoising model to obtain low-dimensional feature data, wherein each iterative encoding includes: performing linear transformation on the input feature of this iterative encoding to obtain a linear transformation result; performing projection transformation on the linear transformation result using the corresponding linear weight matrix and nonlinear weight matrix to obtain a projection transformation result; normalizing the projection transformation result to obtain the activation output of this iterative encoding, and using the activation output of this iterative encoding as the input feature of the next iterative encoding; wherein the input feature of the first iterative encoding is the original feature data, and the activation output of the last iterative encoding is the low-dimensional feature data; performing geometric transformation on the low-dimensional feature data through the preset denoising model to obtain the feature to be reconstructed; decoding the feature to be reconstructed through the preset denoising model to obtain the reconstructed feature data of the digital model of the cultural relic.

[0047] A third aspect of the present application provides an electronic device, comprising at least one processor and a memory connected to the processor, wherein:

[0048] The memory is used to store computer programs;

[0049] The processor is used to execute the computer program so that the electronic device can implement the noise reduction method of the digital model of cultural relics according to the first aspect or any implementation method of the first aspect.

[0050] By means of the above technical solution, the present application provides a method and related device for denoising a digital model of cultural relics, which obtains the original feature data of the digital model of cultural relics to be denoised; inputs the original feature data into a preset denoising model; performs multiple iterative encodings through the preset denoising model to obtain low-dimensional feature data, performs geometric transformation on the low-dimensional feature data through the preset denoising model to obtain features to be reconstructed; decodes the features to be reconstructed through the preset denoising model to obtain reconstructed feature data of the digital model of cultural relics. Wherein, each iterative encoding includes: performing a linear transformation on the input features of this iterative encoding to obtain a linear transformation result; performing a projection transformation on the linear transformation result using the corresponding linear weight matrix and nonlinear weight matrix to obtain a projection transformation result; normalizing the projection transformation result to obtain the activation output of this iterative encoding, and using the activation output of this iterative encoding as the input feature of the next iterative encoding; wherein, the input feature of the first iterative encoding is the original feature data, and the activation output of the last iterative encoding is the low-dimensional feature data; it can be seen that this solution improves the accuracy and fidelity of the digital model of cultural relics by combining linear projection transformation and nonlinear projection transformation to project the features during the encoding stage. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The above and other features, advantages, and aspects of the various embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that the originals and elements are not necessarily drawn to scale.

[0052] Figure 1 A schematic flow chart of a method for digital modeling of cultural relics provided in this application;

[0053] Figure 2 A schematic diagram of the architecture of a noise reduction system for a digital model of cultural relics provided in this application;

[0054] Figure 3 A schematic diagram showing raw data of a digital model of cultural relics;

[0055] Figure 4 A flowchart of a specific method for constructing a preset noise reduction model provided in an embodiment of the present application;

[0056] Figure 5 A schematic diagram of the structure of an improved autoencoder provided in an embodiment of the present application;

[0057] Figure 6 A schematic flow chart of a method for denoising a digital model of cultural relics provided in an embodiment of the present application;

[0058] Figure 7 A schematic diagram comparing noise suppression rates provided in the embodiments of the present application;

[0059] Figure 8 A schematic diagram showing a comparison of geometric consistency indicators provided in an embodiment of the present application;

[0060] Figure 9 A schematic diagram showing a comparison of topological fidelity provided in an embodiment of the present application;

[0061] Figure 10 A schematic diagram comparing the processing time efficiency provided in the embodiments of the present application;

[0062] Figure 11 A comparative schematic diagram of the digital models of cultural relics provided in the embodiments of this application;

[0063] Figure 12 A schematic structural diagram of a noise reduction device for a digital model of cultural relics provided in an embodiment of the present application;

[0064] Figure 13 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0065] The following describes the embodiments of the present application in conjunction with the accompanying drawings. The terms used in the implementation methods of the present application are only used to explain the specific embodiments of the present application and are not intended to limit the present application.

[0066] The embodiments of the present application are described below in conjunction with the accompanying drawings. Those skilled in the art will appreciate that, with the development of technology and the emergence of new scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0067] The terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the terms used in this way can be interchangeable under appropriate circumstances, and this is merely a way of distinguishing the objects of the same attributes when describing them in the embodiments of the present application. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, so that the process, method, system, product or equipment comprising a series of units need not be limited to those units, but may include other units that are not clearly listed or inherent to these processes, methods, products or equipment.

[0068] After research, it was found that digital models of cultural relics usually have complex geometric shapes and delicate texture features, such as carved patterns on bronzes, cracks on pottery, and surface textures on stone tools. These features are easily over-smoothed or destroyed during noise processing, resulting in the loss of detailed information of the original cultural relics. In this regard, an embodiment of the present application provides a denoising method for digital models of cultural relics, the purpose of which is to improve the accuracy and fidelity of digital models of cultural relics. The present application can be applied to the field of artificial intelligence technology, and specifically can be applied to denoising the original digital model of cultural relics using a preset denoising model to obtain a high-precision and high-fidelity digital model of cultural relics, so as to facilitate the research, restoration, display and archiving of cultural relics.

[0069] Figure 1 This is a flowchart of a method for reducing noise in a digital model of cultural relics provided in this application, such as Figure 1 As shown, the method includes:

[0070] S101: Obtain original feature data of a digital model of a cultural relic to be denoised.

[0071] In this embodiment, the original feature data of the digital model of the cultural relic to be denoised is obtained by preprocessing and standardizing the digital model of the cultural relic.

[0072] In an optional embodiment, raw data of a digital model of a cultural relic is collected and preprocessed using a predetermined preprocessing method, wherein the preprocessing includes coordinate alignment, defect repair, and texture mapping. The preprocessed raw data of the digital model of the cultural relic is then subjected to a predetermined standardization process to obtain raw feature data of the digital model of the cultural relic, wherein the standardization process includes redundancy removal, pixel-level segmentation, feature extraction, and data enhancement.

[0073] S102: Inputting the original feature data of the digital model of the cultural relic into a preset noise reduction model.

[0074] In this embodiment, the preset denoising model is used to encode, geometrically transform, and decode the original feature data of the digital model of cultural relics to achieve denoising of the digital model of cultural relics. The preset denoising model is constructed and trained based on an autoencoder.

[0075] In an optional embodiment, a preset denoising model is obtained by training a pre-constructed improved autoencoder. The structure of the preset denoising model includes an encoder, a geometric transformation layer, and a decoder. The encoder includes a plurality of encoding layers connected in sequence, and the decoder includes a plurality of decoding layers connected in sequence, wherein the encoding layers and the decoding layers are symmetrically distributed. The model parameters of the preset denoising model include a linear weight matrix and a nonlinear weight matrix. During the training process, the model parameters are iteratively updated, and the improved autoencoder is configured with the model parameters after the training completion conditions are met, thereby obtaining the preset denoising model for denoising the digital model of cultural relics.

[0076] In an optional embodiment, the loss function pre-configured for training the preset noise reduction model is an enhanced loss function, and the specific training method includes:

[0077] The training data is obtained. The training data includes original feature data of a plurality of sample cultural relic digital models. The method for obtaining the original feature data of the sample cultural relic digital models is described in S101.

[0078] Iterations are performed until the preset training completion conditions are met. Any iteration is the target iteration, which includes:

[0079] The original feature data of each sample cultural relic digital model is input into the improved autoencoder to be trained, and the reconstructed feature data of each sample cultural relic digital model output by the improved autoencoder is obtained.

[0080] The iterative loss value is calculated based on the reconstructed and original feature data of all sample digital models of cultural relics in the target iteration using a preset enhanced loss function. The enhanced loss function is constructed based on an error function that introduces a periodic penalty and a regularization function.

[0081] Based on the loss value of the target iteration, the gradient descent method is used to update the model parameters in the model parameter set.

[0082] Determine whether the preset training completion conditions are met.

[0083] If it is reached, the iteration is stopped, and the improved autoencoder is configured with each model parameter in the updated model parameter set to obtain the preset denoising model.

[0084] If not reached, execute the next iteration.

[0085] It should be noted that the loss function is not limited to the enhanced loss function, and the model parameters updated through iteration include the linear weight matrix and the nonlinear weight matrix of each coding layer.

[0086] S103: Perform multiple iterations of encoding using a preset noise reduction model to obtain low-dimensional feature data.

[0087] In this embodiment, each iterative encoding process includes: performing a linear transformation on the input features of the current iterative encoding process to obtain a linear transformation result. Using the corresponding linear weight matrix and nonlinear weight matrix, the linear transformation result is projected to obtain a projection transformation result. The projection transformation result is normalized to obtain the activation output of the current iterative encoding process, and the activation output of the current iterative encoding process is used as the input feature of the next iterative encoding process. The input features of the first iterative encoding process are the original feature data, and the activation output of the last iterative encoding process is the low-dimensional feature data.

[0088] In an optional embodiment, the preset denoising model includes a sequentially connected encoder, a geometric transformation layer, and a decoder. The encoder includes a plurality of sequentially connected encoding layers, and the decoder includes a plurality of sequentially connected decoding layers. A decoding layer is used to perform a single iterative encoding, and the geometric transformation layer is used to perform a geometric transformation on low-dimensional feature data to obtain features to be reconstructed. The encoder is used to decode the features to be reconstructed to obtain reconstructed feature data of the digital model of the cultural relic.

[0089] Based on the structure of the above-mentioned preset denoising model, the specific method for obtaining low-dimensional feature data by performing multiple iterative encodings through the preset denoising model is as follows: linearly transform the input features through each coding layer in sequence to obtain a linear transformation result. Project the linear transformation result using the corresponding linear weight matrix and nonlinear weight matrix to obtain a projection transformation result. Normalize the projection transformation result to obtain an activation output, and output the activation output. Specifically, the original feature data is input to the first coding layer in the encoder, and the encoding operation is performed in sequence starting from the first coding layer. The input features of the next coding layer are the output features of the previous coding layer, that is, the activation output, until the last coding layer outputs low-dimensional feature data. The low-dimensional feature data is the activation output obtained after the last coding layer performs iterative encoding.

[0090] In an optional embodiment, performing a linear transformation on the input features through the target coding layer to obtain the linear transformation result is specifically implemented as follows: performing a linear transformation on the input features through the target coding layer using the weights and biases of the target coding layer to obtain the linear transformation result. Wherein, if the target coding layer is the first coding layer, the input features of the target coding layer are the original feature data of the digital model of the cultural relic; if the target coding layer is not the first coding layer, the input features of the target coding layer are the output features of the coding layer immediately preceding the target coding layer.

[0091] In an optional embodiment, the linear transformation result is projected by the target coding layer using the corresponding linear weight matrix and the nonlinear weight matrix to obtain the projection transformation result. Specifically, the projected transformation result is achieved as follows:

[0092] The linear transformation result is calculated using the hyperbolic tangent function to obtain the hyperbolic tangent result. The linear transformation result is multiplied by the linear weight matrix of the target coding layer and the linear transformation result to obtain the linear operation result. The nonlinear operation result is multiplied by the nonlinear weight matrix of the target coding layer and the hyperbolic tangent result to obtain the nonlinear operation result. The linear operation result, the nonlinear operation result, and the projection bias of the target coding layer are added together to obtain the projection transformation result. The weights, biases, and projection biases of the target coding layer are all model parameters obtained through iterative training updates.

[0093] In an optional embodiment, the specific implementation of normalizing the projection transformation result to obtain the activation output is:

[0094] The projection transformation result is normalized by the target coding layer to obtain a normalized result. The normalized result is scaled and translated to obtain an activation output. If the target coding layer is not the last coding layer, the activation output is output as an output feature to the next coding layer after the target coding layer. If the target coding layer is the last coding layer, the activation output is output as low-dimensional feature data to the geometric transformation layer through the target coding layer.

[0095] S104: Performing geometric transformation on the low-dimensional feature data using a preset denoising model to obtain features to be reconstructed.

[0096] In this embodiment, there are multiple specific methods for geometric transformation, for example, performing one or multiple geometric transformations on the low-dimensional feature data, wherein the multiple geometric transformations include skew symmetry transformation and scale transformation.

[0097] In an optional embodiment, a geometric transformation layer constructs a geometric transformation matrix for the low-dimensional feature data using a skew-symmetric transformation function and a scale transformation function. After performing a matrix exponential operation on the geometric transformation matrix, it is multiplied by the low-dimensional feature data to obtain the features to be reconstructed, which are then output to the decoder.

[0098] S105 , decoding the features to be reconstructed by using a preset noise reduction model to obtain reconstructed feature data of the digital model of the cultural relic.

[0099] In an optional embodiment, a decoder decodes the features to be reconstructed, obtaining reconstructed feature data of the digital model of the cultural relic output by the decoder. Specifically, the input features are first decoded and reconstructed sequentially through each decoding layer in the decoder, obtaining reconstructed feature data of the digital model of the cultural relic output by the last decoding layer in the decoder. Furthermore, the decoder sums the nonlinear projection results of all decoding layers to obtain the reconstructed feature data.

[0100] In an optional embodiment, the decoding and reconstruction operation includes one or more nonlinear projections and accumulation of the nonlinear projection results. For example, each decoding layer in the decoder sequentially performs nonlinear projection and feature accumulation on the input features based on corresponding decoding parameters to obtain reconstructed feature data output by the decoder. Specifically, each decoding layer sequentially performs a first nonlinear projection on the input features based on a first nonlinear mapping function to obtain a first mapping feature. The first mapping feature is added to the bias of the decoding layer and then multiplied by the transpose of the weight of the decoding layer to obtain a second mapping feature. The second mapping feature is then nonlinearly projected based on the second nonlinear mapping function to obtain a nonlinear projection result of the decoding layer. If the decoding layer is the first decoding layer, the input features of the decoding layer are the features to be reconstructed. If the decoding layer is not the first decoding layer, the input features of the decoding layer are the nonlinear projection results of the previous decoding layer. The weights and biases of the decoding layer are model parameters obtained through iterative training updates. The nonlinear projection results of all decoding layers are summed up by the decoder to obtain the reconstructed feature data.

[0101] In an optional embodiment, the specific method of adding the nonlinear projection results of all decoding layers through the decoder to obtain the reconstructed feature data is as follows:

[0102] The nonlinear projection result of the decoding layer is sequentially added to the feature accumulation result output by the previous decoding layer through each decoding layer to obtain the feature accumulation result of the decoding layer; if the decoding layer is not the last decoding layer, the feature accumulation result of the decoding layer is output to the next decoding layer through the decoding layer; if the decoding layer is the last decoding layer, the feature accumulation result of the decoding layer is output as the reconstructed feature data through the decoding layer.

[0103] In another optional embodiment, the decoder also includes an accumulation module, and the specific method of adding the nonlinear projection results of all decoding layers through the decoder to obtain the reconstructed feature data is: obtaining the nonlinear projection results output by each decoding layer through the accumulation module, adding the nonlinear projection results output by all decoding layers, and using the added result as the reconstructed feature data.

[0104] It can be seen from the above technical solution that the embodiment of the present application provides a method for denoising a digital model of cultural relics, which obtains the original feature data of the digital model of cultural relics to be denoised; inputs the original feature data into a preset denoising model; performs multiple iterative encodings through the preset denoising model to obtain low-dimensional feature data, performs geometric transformation on the low-dimensional feature data through the preset denoising model to obtain features to be reconstructed; decodes the features to be reconstructed through the preset denoising model to obtain reconstructed feature data of the digital model of cultural relics. Wherein, each iterative encoding includes: performing a linear transformation on the input features of this iterative encoding to obtain a linear transformation result; performing a projection transformation on the linear transformation result using the corresponding linear weight matrix and nonlinear weight matrix to obtain a projection transformation result; normalizing the projection transformation result to obtain the activation output of this iterative encoding, and using the activation output of this iterative encoding as the input feature of the next iterative encoding; wherein, the input feature of the first iterative encoding is the original feature data, and the activation output of the last iterative encoding is the low-dimensional feature data; it can be seen that this solution improves the accuracy and fidelity of the digital model of cultural relics by projecting the features in combination with linear projection transformation and nonlinear projection transformation in the encoding stage.

[0105] Furthermore, the specific implementation methods of the method for reducing the noise of the digital model of cultural relics provided in the embodiment of the present application include multiple methods. Optionally, the present application can be applied to Figure 2 The denoising system 200 for the digital model of artifacts is shown, Figure 2 As shown, the noise reduction system 200 includes a data preprocessing module 201, a data standardization module 202, a data storage module 203, a machine learning modeling module 204, a preset noise reduction model 205 and a model application module 206.

[0106] The data preprocessing module 201 is used to collect the original data of the digital model of the cultural relics and preprocess the original data using a preset preprocessing method.

[0107] In this embodiment, the data source of the original data of the digital model of cultural relics is mainly based on the three-dimensional scanning model provided by museums and cultural relics protection institutions. The original data collection method includes structured light scanning and laser scanning, and the data storage format of the original data adopts the standard OBJ and STL file formats. Figure 3 A schematic diagram illustrating the raw data of a digital model of cultural relics.

[0108] In this embodiment, the preprocessing method includes but is not limited to operations such as coordinate alignment, defect repair, and texture mapping, which are used to ensure the consistency and integrity of the data.

[0109] In summary, the data acquisition module collects and preprocesses the raw data of the digital model of cultural relics, providing high-quality data for subsequent model training and application.

[0110] In this embodiment, the data standardization module 202 is used to perform a preset standardization process on the preprocessed raw data of the digital cultural relic model to obtain the raw feature data of the digital cultural relic model, and store the feature data of the digital cultural relic model in the data storage module 203. The raw feature data of the digital cultural relic model includes a multidimensional feature tensor of the digital cultural relic model. Optionally, the standardization process includes: 1. performing dimensionality reduction and de-redundancy on the raw data to reduce redundant information in the raw data; 2. performing pixel-level segmentation and feature extraction on the raw data to generate a multidimensional feature tensor; and 3. performing data enhancement on the multidimensional feature tensor, where data enhancement includes rotation, scaling, and adding simulated noise.

[0111] In this embodiment, the data storage module 203 is used to store data from various stages of the noise reduction process for the digital cultural relic model through a multi-layered data storage architecture, including raw data, preprocessed raw data, and standardized raw data. Furthermore, the data storage module implements rapid data retrieval, supports data screening based on criteria such as cultural relic type, time, and acquisition source, and provides backup and encryption mechanisms to ensure data security and sustainability.

[0112] In this embodiment, the machine learning modeling module 204 is used to construct a preset denoising model 205, which is used to: construct the basic architecture of the improved autoencoder, train and verify the autoencoder based on the original feature data of the digital model of the cultural relics, and optimize the model parameters of the autoencoder to obtain the preset denoising model.

[0113] In this embodiment, the model application module 206 is used to use the trained preset denoising model to perform denoising on the digital model of cultural relics to be denoised, and obtain the reconstructed feature data output by the preset denoising model as the denoising result of the digital model of cultural relics.

[0114] Furthermore, the embodiment of the present application provides a specific method for constructing a preset noise reduction model applied to a machine learning modeling module, such as Figure 4 As shown, this method specifically includes:

[0115] S401. Construct a basic architecture of the improved autoencoder and initialize the model parameter set of the autoencoder.

[0116] In this embodiment, the preset noise reduction model is constructed based on an improved autoencoder. Figure 5 A schematic diagram of the structure of an improved autoencoder provided in an embodiment of the present application is shown in FIG. Figure 5As shown, the improved autoencoder is an autoencoder based on geometric transformation and topological constraints. The improved autoencoder includes an encoder, a geometric transformation layer, and a decoder. The encoder includes a plurality of encoding layers connected in sequence, and the decoder includes a plurality of decoding layers connected in sequence. The encoding layer and the decoding layer are symmetrically distributed. The number of encoding layers and the number of decoding layers are both , the codec layer includes the encoding layer and the decoding layer, and the number of codec layers is E.

[0117] In this embodiment, the model parameter set includes encoding parameters of each encoding layer and decoding parameters of each decoding layer. The encoding parameters include weights, biases, linear weight matrices, nonlinear weight matrices and projection biases, and the decoding parameters include weights and biases.

[0118] In this embodiment, n represents the number of iterations. After the nth iteration, the model parameter set of the improved autoencoder is A set of model parameters including the encoder and the decoder model parameter set When n=0, the initialized model parameter set is The set of model parameters of the initialized encoder included in And the model parameter set of the initialized decoder , respectively expressed as:

[0119] ,in, Indicates the The weights of the encoding layer, Indicates the The bias of the coding layer, Indicates the The linear weight matrix of the encoding layer, Indicates the The nonlinear weight matrix of the coding layer, Indicates the Projection bias of the coding layer;

[0120] , Indicates the The weights of the decoding layer, Indicates the Bias for the decoding layer.

[0121] S402: Obtain training data.

[0122] In this embodiment, the training data includes original feature data of a plurality of sample cultural relic digital models.

[0123] It should be noted that based on the need for noise reduction, the original feature data of multiple types of digital models of cultural relics are extracted from the data storage module, and the original feature data of the digital models of cultural relics are balanced so that the number of original feature data of each type of digital models of cultural relics is roughly the same, avoiding the long-tail effect in the training model.

[0124] In this embodiment, according to the pre-set training configuration, the training data is divided into batches to obtain multiple sample sets, and S403 is iterated based on the sample sets batch by batch until a preset training completion condition is met.

[0125] S403: Input the original feature data of the sample cultural relic digital model into the encoder, sequentially encode the input features based on corresponding encoding parameters through each encoding layer, and output low-dimensional feature data through the last encoding layer.

[0126] In this embodiment, when , No. The input features of the encoding layer are the original feature data. , No. The input feature of the encoding layer is The output features of the coding layer. When n=1, the coding parameters of the coding layer are the initialized coding parameters, that is, , when n>1, the coding parameters of the coding layer are the coding parameters updated after the n-1th iteration, that is, .

[0127] In this embodiment, based on forward propagation, the original feature data of the sample cultural relic digital model is input into the first coding layer, and the input features are encoded in sequence through each coding layer based on the corresponding coding parameters, and the low-dimensional feature data is output through the last coding layer (the Lth coding layer). Taking the encoding layer as an example, each encoding layer encodes the input features based on the corresponding encoding parameters in the following ways:

[0128] A1, No. The coding layer uses The encoding layer weights and biases perform a linear transformation on the input features to obtain a linear transformation result.

[0129] In this embodiment, The input features of the encoding layer are the original feature data. The input feature of the encoding layer is -1 The output feature of the encoding layer, that is, The activation output of the encoding layer is denoted as , will Weights of the encoding layer Multiply Add the Bias of the coding layer , to achieve The linear transformation of , refer to formula (1) for details, as follows:

[0130] (1).

[0131] A2, The coding layer uses the linear weight matrix and the nonlinear weight matrix to perform projection transformation on the linear transformation result to obtain a projection transformation result.

[0132] In this embodiment, a hyperbolic tangent function is used to operate on the linear transformation result, thereby introducing nonlinear characteristics and obtaining a hyperbolic tangent result. The linear transformation result is multiplied by a linear weight matrix to obtain a linear operation result, and the nonlinear weight matrix is multiplied by the hyperbolic tangent result to obtain a nonlinear operation result. The linear operation result, the nonlinear operation result, and the projection bias are added together to obtain a projection transformation result.

[0133] Specifically, the method for the coding layer to perform projection transformation on the linear transformation result using the linear weight matrix and the nonlinear weight matrix is shown in formula (2):

[0134] = (2);

[0135] In formula (2), Indicates the The projection transformation result of the coding layer, Indicates the Projection transformation function of the coding layer. For the The linear weight matrix of the encoding layer, For the The nonlinear weight matrix of the coding layer, is the hyperbolic tangent function; For the The bias term of the projection transformation function of the coding layer, also known as the projection bias.

[0136] It should be noted that this step realizes the refinement of the projection transformation function based on the requirement of modeling the nonlinear characteristics of high-frequency noise in the characteristic data of the digital model of the sample cultural relics, thereby enhancing the modeling ability of high-frequency noise. There are often high-frequency details in the characteristic data of the digital model of the sample cultural relics, such as texture or engraving lines. These details are easily ignored in the noise reduction process. The encoder operates on the linear transformation result through the hyperbolic tangent function, thereby introducing nonlinear features, and through the adjustment of the linear weight matrix and the nonlinear weight matrix, it can better retain these high-frequency details while reducing noise.

[0137] A3, The encoding layer uses a nonlinear activation function to The projection transformation result of the coding layer is normalized to obtain the first The activation output of the encoding layer is output as the output feature.

[0138] In this embodiment, The calculation formula of the activation output of the encoding layer is shown in formula (3):

[0139] (3);

[0140] In formula (3), For the The activation output of the encoding layer, represents a nonlinear activation function. Optionally, the nonlinear activation function adopts a Sigmoid activation function or a normalized activation function. The normalized activation function is used to perform standard normalization on the projection transformation result before scaling and translating it. For the specific normalized activation function, see formula (4):

[0141] (4);

[0142] In formula (4), Represents the normalized result of the projection transformation result, is the scaling factor used to scale the normalized result of the projection transformation. is the translation factor, which is used to translate the normalized result of the projection transformation. For the The projection transformation result of the coding layer is preferably: Set to 0.3, Set to 2.

[0143] In formula (4), For the The mean of the projection transformation results of the coding layer. The mean calculation method is shown in formula (5):

[0144] (5);

[0145] In formula (5), for The dimension of for The index of the dimension, that is, It is composed of the eigenvalue of the first dimension to the eigenvalue of the Mth dimension. for No. eigenvalues.

[0146] In formula (4), For the The standard deviation of the projection transformation result of the coding layer. The standard deviation calculation method is shown in formula (6):

[0147] (6);

[0148] In this embodiment, if , that is, The coding layer is not the last coding layer. The encoding layer activates the features Output to Coding layer, if , that is, The coding layer is the last coding layer. The encoding layer activates the features Output to the geometric transformation layer.

[0149] It should be noted that this step uses a normalized activation function to perform nonlinear activation on the projection transformation results during the forward propagation process, achieving normalization and thus enabling adaptive matching of different feature distributions. The morphology and distribution of the sample artifact digital models vary greatly, such as the significant difference between the complex patterns on the surface of bronze artifacts and the smooth texture of stone artifacts. Through the nonlinear activation function, the scaling and translation factors are used to make targeted adjustments to the features, ensuring that the denoising and reconstruction processes are adapted to the diverse feature distributions, thereby improving the autoencoder's generalization ability for various types of sample artifact digital models.

[0150] S404 , performing multiple geometric transformations on the output features of the encoder through the geometric transformation layer in the autoencoder to obtain features to be reconstructed, and outputting the features to be reconstructed to the decoder.

[0151] In this embodiment, the encoder outputs the activation feature through the Lth encoding layer ,The output features of the encoder are low-dimensional feature data, and the multiple geometric transformations include skew symmetric transformation and scale transformation.

[0152] In this embodiment, a geometric transformation layer is used to construct a geometric transformation matrix of the low-dimensional feature data output by the encoder using a skew-symmetric transformation function and a scale transformation function. After performing a matrix exponential operation on the geometric transformation matrix, it is multiplied with the low-dimensional feature data to obtain the features to be reconstructed. Specifically, the method of multiple geometric transformations is shown in formula (7):

[0153] (7);

[0154] In formula (7), represents the geometric transformation function, is the matrix exponential operation, where Represents the construction function of the geometric transformation matrix, which is used to Perform skew symmetry transformation and scale transformation. Specifically, the construction function of the geometric transformation matrix is shown in formula (8):

[0155] (8);

[0156] In formula (8), represents the skew-symmetric transformation function, represents the scaling function, is the first layer of the last coding layer (the Lth coding layer). eigenvalues, is the total dimension of the low-dimensional feature data, is the dimension index of the low-dimensional feature data, is the skew-symmetric transformation weight of the eigenvalue, is the scale transformation weight of the eigenvalue, optionally, and Can be pre-configured to 0.3 and 0.7 respectively.

[0157] In summary, the construction of the skew-symmetric transformation matrix and the scale transformation matrix of the low-dimensional feature data is realized through formula (8), and the multiple geometric transformations of the skew-symmetric transformation and scale transformation are realized through formula (7).

[0158] It should be noted that this step uses multiple geometric transformations to perform spatial geometric adjustments on the low-dimensional feature data encoded by the encoder, thereby realizing in-depth utilization of the three-dimensional structural information of the cultural relics, avoiding the problem that it is difficult to capture global shape features at the pixel level alone, and compressing complex sample cultural relic digital models into potential features. The geometric transformation layer can significantly enhance the autoencoder's ability to capture the three-dimensional structure of cultural relics. Sample cultural relic digital models usually have complex geometric details and noise characteristics. It is difficult to fully capture the overall shape by relying solely on traditional pixel-level processing. The geometric transformation layer ensures that the global shape and key features of the cultural relics are retained during the noise reduction process by adjusting the input features at the spatial level. For example, for cultural relics with asymmetric carvings, the geometric transformation layer can effectively maintain their unique geometric form without being disturbed by random noise.

[0159] S405 , sequentially performing nonlinear projection and feature accumulation on the input features based on corresponding decoding parameters through each decoding layer in the decoder to obtain reconstructed feature data output by the decoder.

[0160] In this embodiment, when , No. The input features of the decoding layer are the features to be reconstructed , that is ,when , No. The input feature of the decoding layer is Nonlinear projection results of the decoding layer And the feature accumulation result ,when , No. The output feature of the decoding layer is Nonlinear projection results of the decoding layer And the feature accumulation result ,when , No. The output feature of the decoding layer is Feature accumulation results of the decoding layer , that is, the decoder passes the The decoding layer outputs the reconstructed features.

[0161] When n=1, the decoding parameters of the decoding layer are the initialized decoding parameters, that is, , when n>1, the decoding parameters of the decoding layer are the decoding parameters updated after the n-1th iteration, that is, .

[0162] First Taking the decoding layer as an example, each decoding layer performs nonlinear projection and feature accumulation on the input features based on the corresponding decoding parameters, including:

[0163] B1, The decoding layer is based on the first nonlinear mapping function Perform the first nonlinear projection on the input features to obtain the first mapping features.

[0164] In this embodiment, The first mapping feature of the decoding layer The calculation method of is shown in formula (9):

[0165] (9);

[0166] In formula (9), is the Sigmoid function, is the Hadamard product operation.

[0167] B2, The decoding layer combines the first mapping feature with the After adding the bias of the decoding layer, multiply it by the transposed The weights of the decoding layer are used to obtain the second mapping features.

[0168] B3. Based on The second nonlinear mapping function of the decoding layer Perform nonlinear projection on the second mapping feature to obtain the first The nonlinear projection result of the decoding layer.

[0169] In this embodiment, Nonlinear projection results of the decoding layer The calculation method of is shown in formula (10):

[0170] (10);

[0171] In formula (10), Indicates the The second nonlinear mapping function of the decoding layer, preferably, the second nonlinear mapping function adopts the ReLU nonlinear function, For the the transpose of the decoder layer's weights, For the The bias of the decoding layer, Represents the second mapping feature, that is, the second mapping feature is equal to the sum of the first mapping feature and the After the bias of the decoding layer is added, it is multiplied by the The transpose of the decoding layer's weights.

[0172] B4, the The nonlinear projection result of the decoding layer plus the The feature accumulation result of the decoding layer is obtained Feature accumulation result of the decoding layer.

[0173] In this embodiment, The calculation method of the feature accumulation result of the decoding layer is shown in formula (11):

[0174] (11).

[0175] B5. Output the output features.

[0176] In this embodiment, when , No. The output feature of the decoding layer is Nonlinear projection results of the decoding layer And the feature accumulation result , No. The decoding layer outputs the output features to the Decoding layer. , No. The output feature of the decoding layer is Feature accumulation results of the decoding layer , No. The decoding layer accumulates the features The reconstructed feature data is output to the loss module, where the loss module is configured in the machine learning modeling module.

[0177] In summary, S405 restores the digital model of the sample cultural relic layer by layer through the decoder of the autoencoder for the features after geometric transformation. Among them, the features to be reconstructed, which are output through multiple geometric transformations of the geometric transformation layer, retain the features of the feature topological relationship that is helpful for reconstruction. Furthermore, the features are projected nonlinearly multiple times through the decoding layer to achieve multiple nonlinear transformations of the features, thereby realizing topological constraints and ensuring that the spatial structure of the data at different levels remains consistent. It can be seen that this step uses topological constraints to make the output reconstructed feature data consistent with the original feature data in topological structure, thereby reducing reconstruction errors and retaining cultural relic details. Through topological constraints, the reconstruction model can avoid shape breaks or topological errors while restoring geometric details. For example, for broken pottery fragments, topological constraints can help the model more accurately splice the original complete shape during reconstruction.

[0178] In this step, based on the need for higher-precision reconstruction of complex cultural relic data, the decoder function is refined and calculated. The cultural relic data is restored layer by layer through multi-layer nonlinear mapping. At the same time, multiple geometric transformations and topological constraints are combined to significantly reduce the reconstruction error. For example, when reconstructing a bronze artifact with a severely weathered surface, this process can accurately restore its internal details without affecting the overall shape due to the noise of the weathering layer. Based on B1~B4, at the nth iteration, the decoder function is expressed as formula (12):

[0179] (12);

[0180] Among them, if n=1, Represents the decoder model parameter set after initialization. If n>1, Represents the model parameters of the decoder updated after the n-1th iteration, that is, the decoding parameters. Unlike the traditional method of one-time encoding and decoding, this step proposes a "layer-by-layer restoration" scheme, that is, during the decoding process, each decoding layer performs multiple nonlinear mappings on the features, so that each decoding layer can correct the error in a targeted manner, gradually approaching the details of the original cultural relic data, thereby improving the overall restoration quality.

[0181] S406 , calculating the loss value between the reconstructed feature data and the original feature data based on a preset enhancement loss function through a loss module.

[0182] In this embodiment, the enhanced loss function is constructed based on the error function and the regularization function that introduces periodic penalties.

[0183] In this embodiment, the error function with periodic penalty is used to calculate the loss error of each sample cultural relic digital model. For example, at the nth iteration, the sample set includes the original feature data of N sample cultural relic digital models, which is recorded as ~ After obtaining the reconstructed feature data of the sample cultural relic digital model through S403~406, the loss error of each sample cultural relic digital model is calculated based on the error function with periodic penalty. The reconstructed feature data of the i-th (1≤i≤N) sample cultural relic digital model Mi is taken as And the original feature data As an example, the error function calculation based on the introduction of periodic penalty and The loss error The method is shown in formula (13):

[0184] (13);

[0185] In formula (12), To introduce the error function of periodic penalty, is the adjustment coefficient, is the cosine function, is a constant pi, preferably, Set to 0.5.

[0186] In this embodiment, the regularization function is used to calculate the regularization term of the nth iteration The method is shown in formula (14):

[0187] (14);

[0188] In formula (13), represents the regularization function, It represents the set of model parameters after the n-1th iteration update, that is, the set of model parameters currently to be updated. is the regularization coefficient of the autoencoder, is the L2 norm, preferably, Set to 0.3. Represents the weight of the e-th codec layer, which includes the encoding layer and the decoding layer. Indicates the total number of encoding and decoding layers, that is, ,in, ~ express ~ , ~ express ~ .

[0189] In this embodiment, the loss value of the nth iteration is calculated based on the enhanced loss function See formula (15):

[0190] (15);

[0191] Where, is the enhanced loss function of the autoencoder, is the number of samples input to the autoencoder, that is, the number of original feature data in the sample set of the nth iteration.

[0192] In summary, this step calculates the loss value through the loss module, optimizes model performance by introducing a periodic penalty, and improves the adaptability of the autoencoder in suppressing noise and preserving geometric details through regularization. Specifically, the noise in the feature data of the sample digital model of the cultural relic is often unevenly distributed. The enhanced loss function can adapt to this noise distribution while emphasizing the precise matching of the reconstruction result with the original data. For example, for a damaged piece of porcelain, this loss function can guide the model to effectively ignore the noise in the missing area while accurately reconstructing the well-preserved part.

[0193] Furthermore, the enhanced loss function constructed based on the error function and regularization function with the introduction of periodic penalty is summarized as formula (16):

[0194] (16);

[0195] It should be noted that for any sample cultural relic digital model Mi, is the input to the encoder , Output of the decoder .

[0196] S407: Use the gradient descent method to update each model parameter in the model parameter set of the autoencoder.

[0197] In this embodiment, the gradient descent method is used to update each model parameter in the model parameter set of the autoencoder. Taking weights as an example, the specific method of weight updating includes:

[0198] C1. Calculate the gradient of the enhanced loss function with respect to the weights , the calculation method is shown in formula (17):

[0199] (17);

[0200] In formula (17), is the symbol of partial derivative.

[0201] C2, based on the gradient and learning rate, updates the weights of each encoding and decoding layer in the autoencoder. The update method is shown in formula (18):

[0202] (18);

[0203] In formula (18), is the parameter update operation, represents the weight of the e-th codec layer after the n-th iteration update, Represents the weight of the e-th codec layer after the n-1th iteration update. is the learning rate of the autoencoder. Preferably, Set to 0.01.

[0204] S408, determine whether the preset training completion condition is reached, if so, stop the iteration, and use the model parameter set of the autoencoder Configure the model parameters of the autoencoder to obtain the preset denoising model. If not, return to S403 to perform the n+1th iteration.

[0205] In this embodiment, the training completion condition includes the number of iterations reaching a preset iteration threshold or the loss value being less than a preset loss value threshold. For example, if the preset maximum number of iterations is 1000 and the preset loss value threshold is 0.001, if n=1000 or , the training completion condition is met.

[0206] It should be noted that when the training completion condition is not met, steps S403 to S408 are repeated until the training completion condition is met, which means that the model training is completed.

[0207] As can be seen from the above technical solutions, the method for constructing a preset denoising model provided in the embodiment of the present application, by training an improved autoencoder, obtains a preset denoising model, which overcomes the technical problem of poor pixel-level denoising effect of digital models of cultural relics, mainly reflected in the following aspects:

[0208] 1. Traditional autoencoders have a weak ability to capture the geometric details and global morphology of digital models of cultural relics, making it difficult to fully preserve the key features and detailed information of cultural relics. This solution uses an improved autoencoder and a combination of linear and nonlinear weight matrices in the encoding phase to achieve precise modeling of high-frequency noise, ensuring that the geometric details and key morphology of the model are preserved during the denoising process. This achieves high-precision denoising and detail preservation for the complex geometric structures of digital cultural relic models. When processing data on cultural relics with carvings and textures, it can effectively remove noise while preserving key details, improving the restoration of the geometric morphology of digital cultural relic models by the preset denoising model.

[0209] 2. The topological structure of three-dimensional cultural relic data (such as surface connectivity and hole shape) is crucial for authenticating cultural relics. However, in traditional denoising and reconstruction methods, traditional autoencoders lack effective constraints on the consistency of the cultural relic's topological structure during the denoising process, which can lead to shape discontinuities or topological errors in the reconstructed digital model. This solution uses an improved autoencoder to apply topological constraints during the decoder phase to ensure that the reconstructed image maintains topological consistency with the original model, thereby reducing reconstruction errors and preserving cultural relic details. This achieves topological consistency during the denoising process, ensuring that the reconstructed digital model remains consistent with the original in terms of connectivity, closure, and number of holes, effectively reducing shape discontinuities and topological errors.

[0210] 3. Faced with diverse digital models of cultural relics, traditional autoencoders lack the ability to dynamically adapt to differences in feature distribution, resulting in uneven noise reduction effects on different types of digital models of cultural relics. This solution uses an improved autoencoder. In the encoder encoding stage, based on the adaptive feature normalization method, scaling factors and translation factors are introduced in the forward propagation process to normalize the features, thereby achieving adaptive matching of the feature distribution of different cultural relic features. This improves the adaptability of the autoencoder to diverse digital models of cultural relics, and demonstrates stable noise reduction performance in the noise reduction tasks of different types of digital models of cultural relics, such as bronze, pottery, and stone tools, ensuring the generalization of the model under different data distributions.

[0211] 4. Since the noise distribution in digital models of cultural relics is often uneven, traditional autoencoders fail to achieve effective noise reduction in high-frequency noise areas. The fidelity and accuracy of the reconstructed feature data are difficult to meet the needs of cultural relic protection and display. This scheme introduces a periodic penalty mechanism into the loss function to adapt to the uneven noise distribution while ensuring that the reconstruction results accurately match the original data. This significantly reduces the reconstruction error of the digital model of cultural relics after denoising, retains more local details, and improves the overall performance of the preset denoising model.

[0212] It should be noted that the machine learning modeling module is also used to tune the preset noise reduction model, including adjusting hyperparameters such as the number of network layers, number of neurons, activation function and learning rate to improve the noise reduction effect. For specific tuning timing and methods, please refer to the existing technology.

[0213] Furthermore, the embodiment of the present application provides a specific implementation method of a denoising method for a digital model of cultural relics applied to a model application module. Figure 6 This is a specific implementation flow chart of a method for reducing noise of a digital model of cultural relics provided in an embodiment of the present application, such as Figure 6 As shown, this method specifically includes:

[0214] S601: Obtain original feature data of the digital model of the cultural relic to be denoised.

[0215] In this embodiment, the method for obtaining the original feature data of the digital model of the cultural relics includes: collecting the original data of the digital model of the cultural relics, and preprocessing the original data through a preset preprocessing method, and performing a preset standardization process on the preprocessed original data of the digital model of the cultural relics to obtain the original feature data of the digital model of the cultural relics. The specific method of obtaining the original feature data can be found in the above embodiment.

[0216] S602: Inputting the original feature data of the digital model of the cultural relic into a preset noise reduction model.

[0217] In this embodiment, the preset noise reduction model is constructed based on an improved autoencoder, which includes an encoder, a geometric transformation layer, and a decoder connected in sequence. The improved autoencoder is an autoencoder based on geometric transformation and topological constraints. The improved autoencoder includes an encoder, a geometric transformation layer, and a decoder. The encoder includes a plurality of encoding layers connected in sequence, and the decoder includes a plurality of decoding layers connected in sequence. The encoding layers and the decoding layers are symmetrically distributed. The number of encoding layers and the number of decoding layers are both , No. The coding layer is Layer coding layer, The decoding layer is Layer decoding layer.

[0218] In this embodiment, the model parameter set includes encoding parameters for each encoding layer and decoding parameters for each decoding layer. The encoding parameters include weights, biases, linear weight matrices, nonlinear weight matrices, and projection biases, and the decoding parameters include weights and biases. The model parameter set of the preset denoising model is the model parameter set updated after the nth iteration, where the nth iteration is the last iteration.

[0219] Set the encoding parameters of the preset denoising model Expressed as:

[0220] ;

[0221] in, Indicates the The weights of the encoding layer, Indicates the The bias of the coding layer, Indicates the The linear weight matrix of the encoding layer, Indicates the The nonlinear weight matrix of the coding layer, Indicates the Projection bias for the coding layer.

[0222] ;

[0223] in, Indicates the The weights of the decoding layer, Indicates the Bias for the decoding layer.

[0224] It should be noted that the structure and construction method of the preset noise reduction model can refer to the above embodiment.

[0225] S603. Perform linear transformation on the input features using the corresponding weights and biases through each coding layer in sequence to obtain a linear transformation result, perform projection transformation on the linear transformation result using a linear weight matrix and a nonlinear weight matrix to obtain a projection transformation result, normalize the projection transformation result using a nonlinear activation function to obtain an activation output, and output the activation output as the output feature, and output the low-dimensional feature data to the geometric transformation layer through the last coding layer.

[0226] In this embodiment, the input features are encoded in sequence through each coding layer based on the corresponding coding parameters, and low-dimensional feature data is output through the last coding layer. For a specific encoding method, see S403 above.

[0227] S604 , performing multiple geometric transformations on the output features of the encoder through the geometric transformation layer in the autoencoder to obtain features to be reconstructed, and outputting the features to be reconstructed to the decoder.

[0228] In this embodiment, the multiple geometric transformations include skew symmetry transformation and scale transformation, and details can be found in the above S404.

[0229] S605. Perform a first nonlinear projection on the input feature based on the first nonlinear mapping function through the decoding layers in the decoder in sequence to obtain a first mapping feature, add the first mapping feature and the bias, and multiply it by the transposed weight to obtain a second mapping feature, perform a nonlinear projection on the second mapping feature based on the second nonlinear mapping function to obtain a nonlinear projection result, add the nonlinear projection result to the feature accumulation result of the previous decoding layer to obtain the feature accumulation result of the decoding layer, and output the feature accumulation result of the last decoding layer as the reconstructed feature data.

[0230] In this embodiment, nonlinear projection and feature accumulation are performed on the input features based on corresponding decoding parameters in each decoding layer in the decoder in sequence to obtain reconstructed feature data output by the decoder. For details, please refer to the above S405.

[0231] It should be noted that when each step in S603 to 605 corresponds to the above-mentioned S403 to 405, the model parameters to be updated in the training phase in S403 to 405 are replaced with the model parameters of the preset denoising model in S602.

[0232] As can be seen from the above technical solution, the embodiment of the present application provides a method for denoising a digital model of cultural relics. The method denoises the digital model of cultural relics based on a preset denoising model and outputs reconstructed feature data. Since the preset denoising model is constructed based on an improved autoencoder, the improved autoencoder, based on the traditional autoencoder, adopts a combination of linear weight matrices and nonlinear weight matrices in the feature encoding stage of the autoencoder to achieve fine modeling of high-frequency noise, ensuring that the geometric details and key forms of the model are retained during the denoising process. In addition, an adaptive feature normalization method is used to enable the features to dynamically adapt to the feature distribution of different digital models of cultural relics. Under the premise of achieving denoising of the digital model of cultural relics, a higher-precision reconstruction effect can also be achieved.

[0233] Furthermore, a geometric transformation layer is used after the encoder to perform skew-symmetry and scale transformations on the features output by the encoder. Topological constraints are then applied to the decoder to reduce noise while maintaining the details and topological structure of the artifact. This improves noise suppression and geometric consistency, more accurately preserves the shape and details of the three-dimensional artifact, and enhances the overall reconstruction quality and visual consistency. Therefore, the model application module uses the reconstructed feature data output by the preset denoising model as the denoising result of the digital artifact model, achieving high-precision denoising of the three-dimensional artifact digital model while maintaining the consistency of geometric details and topological structure, thereby improving the reconstruction quality and adaptability of the model.

[0234] In summary, this scheme combines geometric transformation, topological constraints and adaptive feature normalization methods to achieve high-precision denoising of three-dimensional cultural relic digital models based on an improved autoencoder algorithm, while maintaining the consistency of geometric details and topological structure, thereby improving the reconstruction quality and adaptability of the model.

[0235] Furthermore, the original feature data of the sample digital models of cultural relics in the test data was input into a trained traditional autoencoder and an improved autoencoder (i.e., a pre-set denoising model). The denoising results were obtained by using the traditional autoencoder to denoise the test data using the traditional algorithm, and by using the pre-set denoising model to denoise the test data using the improved algorithm. This was done to test the performance of the pre-set denoising model. Specifically, the improved autoencoder was based on geometric transformations and topological constraints, and the test data included sample digital models of cultural relics, i.e., samples 1 to 10.

[0236] Figure 7A noise suppression rate comparison diagram provided in an embodiment of the present application, wherein the horizontal axis identifies each sample cultural relic digital model with sample numbers 1 to 10, and the vertical axis identifies the noise suppression rate of the traditional autoencoder and the preset noise reduction model. Figure 7 The example shows the comparison of the noise suppression rate between the traditional autoencoder and the preset denoising model. The noise suppression rate is a performance indicator to measure the effectiveness of the denoising algorithm in the noise removal task. It is often used in the field of image denoising and is calculated by comparing the signal-to-noise ratio of the signal before and after processing. For example, Figure 7 As shown, the improved autoencoder has significant improvements in all samples.

[0237] Figure 8 A schematic diagram comparing geometric consistency indicators provided in an embodiment of the present application, wherein the horizontal axis identifies each sample digital model of cultural relics with sample numbers 1 to 10, and the vertical axis identifies the geometric consistency indicators of the traditional autoencoder and the preset denoising model. Figure 8 This paper demonstrates the comparison between the traditional autoencoder and the preset denoising model in terms of geometric consistency index. The geometric consistency index is used to evaluate the ability of a 3D model to retain the original geometric form during the denoising or reconstruction process. By comparing the differences between the geometric forms (such as point clouds or meshes) of the preset denoising model and the traditional autoencoder, the effect of the model in retaining shape and details is reflected. Figure 8 It can be seen that the improved autoencoder has better performance in preserving the geometric shape of cultural relics.

[0238] Figure 9 A schematic diagram comparing topological fidelity provided in an embodiment of the present application, wherein the horizontal axis identifies each sample digital model of cultural relics with sample numbers 1 to 10, and the vertical axis identifies the topological fidelity of the traditional autoencoder and the preset denoising model. Figure 9 The example compares the topological fidelity of traditional autoencoders and pre-set denoising models. Topological fidelity measures whether the integrity of the original topological structure of a 3D model is maintained during denoising, reconstruction, or other processing. It mainly evaluates whether the topological properties of the model have changed, such as connectivity, number of holes, closure, and shape consistency. Figure 9 As shown in the figure, the improved autoencoder significantly outperforms the traditional algorithm in terms of topological structure consistency in preserving artifact details.

[0239] Figure 10 A schematic diagram comparing the processing time efficiency provided in an embodiment of the present application, wherein the horizontal axis identifies each sample digital model of cultural relics with sample numbers 1 to 10, and the vertical axis identifies the processing time efficiency of the traditional autoencoder and the preset denoising model. Figure 10 This example shows the comparison of the processing time efficiency between the traditional autoencoder and the preset denoising model. Figure 10 As shown in Figure 3, the improved autoencoder has obvious advantages in processing time efficiency.

[0240] Furthermore, we compared the reconstruction error metrics of the improved autoencoder with those of other autoencoders (such as traditional autoencoders, variational autoencoders, and generative adversarial network models) on different test datasets to analyze the reconstruction performance of each autoencoder. The reconstruction error metrics include:

[0241] Mean square error: measures the mean of the squared difference between the original data and the reconstructed data. The smaller the value, the better the reconstruction effect.

[0242] Mean absolute error: measures the mean absolute difference between the original data and the reconstructed data. The smaller the value, the better the reconstruction effect.

[0243] Table 1 illustrates the comparison of the reconstruction error indicators of various autoencoders, as follows:

[0244] Table 1 Comparison of reconstruction errors

[0245]

[0246] According to Table 1, the mean square error of the improved autoencoder on all data sets is lower than that of other algorithms, indicating that it has a better noise suppression effect during the reconstruction process and can more accurately restore the original data. Compared with the traditional autoencoder, the mean square error of the improved autoencoder is reduced by about 55% on average. Compared with the generative adversarial network method, the mean square error of the improved autoencoder is reduced by about 30% on average.

[0247] According to Table 1, the mean absolute error of the improved autoencoder is the lowest on all datasets, indicating that it retains local features and details more accurately; compared with the traditional autoencoder, the mean absolute error of the improved autoencoder is reduced by about 35% on average.

[0248] To further demonstrate the technical effect of a noise reduction method provided in an embodiment of the present application, Figure 11 The following is a comparison diagram of the digital model of cultural relics before and after noise reduction. Figure 11 As shown, (a) is the digital model of the cultural relic before noise reduction, and (b) is the digital model of the cultural relic after noise reduction by the preset noise reduction model provided in the embodiment of the present application. Figure 11It can be seen that the original digital model of the cultural relic has strong noise, and there are problems such as surface irregularities, blurred details, and unclear edges. The denoising method for the three-dimensional cultural relic model provided by the embodiment of the present application effectively denoises the digital model of the cultural relic. Specifically, through the combination of geometric transformation layers and topological constraints, the cultural relic model can better retain its shape and details during denoising. Adaptive feature normalization improves its adaptability to different types of cultural relics. The enhanced loss function and optimized training strategy ensure higher reconstruction accuracy. Based on this, the noise of the digital model of the cultural relic is removed or suppressed, and the details, texture, shape, etc. of the model surface are more clearly presented. Moreover, after denoising, the digital model of the cultural relic maintains the accuracy of its original shape and structure.

[0249] The above describes a method for reducing the noise of a digital model of cultural relics provided by an embodiment of the present application. The following describes an apparatus for executing the above method for reducing the noise of a digital model of cultural relics.

[0250] See also Figure 12 , Figure 12 This is a schematic diagram of the structure of a noise reduction device for a digital model of cultural relics provided in an embodiment of the present application. Figure 12 As shown, the noise reduction device 1200 for the digital model of cultural relics includes:

[0251] The original data acquisition unit 1201 is used to acquire the original feature data of the digital model of the cultural relic to be denoised;

[0252] The model denoising unit 1202 is used for the model denoising unit, and is used to input the original feature data into a preset denoising model, and perform multiple iterative encodings through the preset denoising model to obtain low-dimensional feature data, each iterative encoding includes: performing linear transformation on the input feature of this iterative encoding to obtain a linear transformation result; performing projection transformation on the linear transformation result using the corresponding linear weight matrix and nonlinear weight matrix to obtain a projection transformation result; normalizing the projection transformation result to obtain the activation output of this iterative encoding, and using the activation output of this iterative encoding as the input feature of the next iterative encoding; wherein the input feature of the first iterative encoding is the original feature data, and the activation output of the last iterative encoding is the low-dimensional feature data; performing geometric transformation on the low-dimensional feature data through the preset denoising model to obtain the feature to be reconstructed; decoding the feature to be reconstructed through the preset denoising model to obtain the reconstructed feature data of the digital model of the cultural relic.

[0253] In a possible implementation, when the original data acquisition unit is used to acquire the original feature data of the digital model of the cultural relic to be denoised, it is specifically used to:

[0254] Collecting original data of the digital model of the cultural relic, and preprocessing the original data of the digital model of the cultural relic using a preset preprocessing method; wherein the preprocessing includes coordinate alignment, defect repair, and texture mapping;

[0255] The pre-processed original data of the digital model of cultural relics are subjected to a preset standardization process to obtain original feature data of the digital model of cultural relics; wherein the standardization process includes redundancy removal, pixel-level segmentation, feature extraction and data enhancement.

[0256] In one possible implementation, the preset noise reduction model includes an encoder, a geometric transformation layer, and a decoder connected in sequence, wherein the encoder includes a plurality of coding layers connected in sequence;

[0257] One encoding layer is used to perform one iterative encoding;

[0258] The geometric transformation layer is used to perform the geometric transformation on the low-dimensional feature data to obtain the features to be reconstructed;

[0259] The encoder is used to perform the decoding of the feature to be reconstructed to obtain the reconstructed feature data of the digital model of the cultural relic.

[0260] In one possible implementation, the model denoising unit is configured to perform a linear transformation on the input features of the current iterative encoding using the preset denoising model, and to obtain a linear transformation result, specifically for:

[0261] Performing a linear transformation on the input features by using the weights and biases of the target coding layer through the target coding layer to obtain the linear transformation result;

[0262] Among them, the target coding layer is the coding layer used to perform this iterative coding. If the target coding layer is the first coding layer, the input feature of the target coding layer is the original feature data. If the target coding layer is not the first coding layer, the input feature of the target coding layer is the output feature of the previous coding layer of the target coding layer.

[0263] In one possible implementation, the model denoising unit is configured to perform a projection transformation on the linear transformation result using the preset denoising model and the corresponding linear weight matrix and the nonlinear weight matrix, and to obtain the projection transformation result, specifically for:

[0264] The linear transformation result is operated by the target coding layer using the hyperbolic tangent function to obtain a hyperbolic tangent result; the linear weight matrix of the target coding layer is multiplied by the linear transformation result to obtain a linear operation result, the nonlinear weight matrix of the target coding layer is multiplied by the hyperbolic tangent result to obtain a nonlinear operation result, and the linear operation result, the nonlinear operation result and the projection bias of the target coding layer are added to obtain the projection transformation result.

[0265] In a possible implementation, when the model denoising unit is configured to perform normalization processing on the projection transformation result using the preset denoising model to obtain the activation output of this iterative encoding, it is specifically configured to:

[0266] The projection transformation result is standard normalized through the target coding layer to obtain a standard normalized result; the standard normalized result is scaled and translated to obtain the activation output; if the target coding layer is not the last coding layer, the activation output is output as an output feature to the next coding layer of the target coding layer through the target coding layer; if the target coding layer is the last coding layer, the activation output is output as the low-dimensional feature data to the geometric transformation layer through the target coding layer.

[0267] In a possible implementation, the model denoising unit is configured to perform a geometric transformation on the low-dimensional feature data using the preset denoising model to obtain the features to be reconstructed, specifically for:

[0268] The geometric transformation layer uses a skew-symmetric transformation function and a scale transformation function to construct a geometric transformation matrix of the low-dimensional feature data; after performing a matrix exponential operation on the geometric transformation matrix, it is multiplied with the low-dimensional feature data to obtain the feature to be reconstructed.

[0269] In one possible implementation, the decoder includes a plurality of decoding layers connected in sequence, and the model denoising unit is configured to decode the to-be-reconstructed feature using the preset denoising model, specifically for:

[0270] The decoding layer of each layer is sequentially passed through to perform a first nonlinear projection on the input feature based on a first nonlinear mapping function to obtain a first mapping feature; the first mapping feature is added to the bias of the decoding layer, and then multiplied by the transpose of the weight of the decoding layer to obtain a second mapping feature, and the second mapping feature is nonlinearly projected based on a second nonlinear mapping function to obtain a nonlinear projection result of the decoding layer; wherein, if the decoding layer is the first decoding layer, the input feature of the decoding layer is the feature to be reconstructed; if the decoding layer is not the first decoding layer, the input feature of the decoding layer is the nonlinear projection result of the previous decoding layer; the nonlinear projection results of all the decoding layers are added through the decoder to obtain the reconstructed feature data.

[0271] In one possible implementation, the model denoising unit is configured to add the nonlinear projection results of all the decoding layers through the decoder to obtain the reconstructed feature data, specifically to:

[0272] The nonlinear projection result of the decoding layer is sequentially added to the feature accumulation result output by the previous decoding layer through each decoding layer to obtain the feature accumulation result of the decoding layer; if the decoding layer is not the last decoding layer, the feature accumulation result of the decoding layer is output to the next decoding layer through the decoding layer; if the decoding layer is the last decoding layer, the feature accumulation result of the decoding layer is output as the reconstructed feature data through the decoding layer.

[0273] In one possible implementation, the device for reducing the noise of the digital model of cultural relics further includes a model training unit; the model training unit is used to:

[0274] Acquire training data, the training data including original feature data of a plurality of sample cultural relic digital models; perform iterations until a preset training completion condition is reached, with any iteration being a target iteration, the target iteration including:

[0275] Inputting the original feature data of each sample cultural relic digital model into an improved autoencoder to be trained, obtaining reconstructed feature data of each sample cultural relic digital model output by the improved autoencoder; calculating the loss value of the target iteration using a preset enhanced loss function based on the reconstructed feature data and the original feature data of all sample cultural relic digital models of the target iteration; the enhanced loss function is constructed based on an error function and a regularization function that introduces a periodic penalty;

[0276] Based on the loss value of the target iteration, the model parameters in the model parameter set are updated using the gradient descent method, wherein the model parameters include the linear weight matrix and the nonlinear weight matrix corresponding to each iteration encoding;

[0277] Determine whether the preset training completion condition is met; if so, stop the iteration, configure the improved autoencoder with the model parameters in the updated model parameter set, and obtain the preset denoising model; if not, execute the next iteration.

[0278] An electronic device is also provided in an embodiment of the present application. Figure 13 As shown, Figure 13 It shows a schematic diagram of the structure of an electronic device suitable for implementing the embodiments of the present application. The electronic device in the embodiments of the present application may include but is not limited to fixed terminals such as mobile phones, laptops, PDAs (personal digital assistants), PADs (tablet computers), desktop computers, etc. Figure 13 The electronic device shown is merely an example and should not limit the functions and scope of use of the embodiments of the present application.

[0279] like Figure 13 As shown, the electronic device may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 1301, which can perform various appropriate actions and processes based on programs stored in a read-only memory (ROM) 1302 or programs loaded from a storage device 1308 into a random access memory (RAM) 1303. When the electronic device is powered on, RAM 1303 also stores various programs and data required for the operation of the electronic device. Processing device 1301, ROM 1302, and RAM 1303 are interconnected via a bus 1304. An input / output (I / O) interface 1305 is also connected to bus 1304.

[0280] Typically, the following devices may be connected to the I / O interface 1305: an input device 1306 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 1307 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 1308 including, for example, a memory card, a hard disk, etc.; and a communication device 1309. The communication device 1309 may allow the electronic device to communicate with other devices wirelessly or by wire to exchange data. Figure 13 The electronic device is shown with various devices, but it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed instead.

[0281] An embodiment of the present application also provides a computer program product including computer-readable instructions. When the computer-readable instructions are executed on an electronic device, the electronic device implements any method for reducing the noise of a digital model of cultural relics provided in the embodiment of the present application.

[0282] A computer-readable storage medium is also provided in an embodiment of the present application. The storage medium carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement any noise reduction method for a digital model of cultural relics provided in an embodiment of the present application.

[0283] It should also be noted that the device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed across multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment. In addition, in the drawings of the device embodiments provided in this application, the connection relationship between the modules indicates that there is a communication connection between them, which can be specifically implemented as one or more communication buses or signal lines.

[0284] Through the description of the above embodiments, those skilled in the art can clearly understand that the present application can be implemented by means of software plus necessary general hardware, and of course can also be implemented by special hardware including application-specific integrated circuits, special CPUs, special memories, special components, etc. In general, all functions performed by computer programs can be easily implemented with corresponding hardware, and the specific hardware structures used to implement the same function can also be diverse, such as analog circuits, digital circuits or special circuits, etc. However, for the present application, software program implementation is a better implementation method in most cases. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a readable storage medium, such as a computer's floppy disk, USB flash drive, mobile hard disk, ROM, RAM, magnetic disk or optical disk, etc., and includes a number of instructions to enable a computer device (which can be a personal computer, training equipment, or network equipment, etc.) to execute the methods described in each embodiment of the present application.

[0285] In the above embodiments, all or part of the embodiments may be implemented by software, hardware, firmware, or any combination thereof. When implemented by software, all or part of the embodiments may be implemented in the form of a computer program product.

[0286] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website, a computer, a training device or a data center by wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) mode to another website, computer, training device or data center. The computer-readable storage medium can be any available medium that a computer can store or a data storage device such as a training device, a data center that includes one or more available media integrations. The available medium can be a magnetic medium, (for example, a floppy disk, a hard disk, a tape), an optical medium (for example, a DVD), or a semiconductor medium (for example, a solid-state drive (SSD)).

Claims

1. A method for reducing noise of a digital model of cultural relics, characterized in that: include: Obtaining original feature data of the digital model of the cultural relic to be denoised; Inputting the original feature data into a preset noise reduction model; Low-dimensional feature data is obtained by performing multiple iterative encodings through the preset denoising model, and each iterative encoding includes: performing a linear transformation on the input features of this iterative encoding to obtain a linear transformation result; performing a projection transformation on the linear transformation result using a corresponding linear weight matrix and a nonlinear weight matrix to obtain a projection transformation result; normalizing the projection transformation result to obtain an activation output of this iterative encoding, and using the activation output of this iterative encoding as the input feature of the next iterative encoding; wherein the input feature of the first iterative encoding is the original feature data, and the activation output of the last iterative encoding is the low-dimensional feature data; Performing geometric transformation on the low-dimensional feature data using the preset denoising model to obtain features to be reconstructed; The features to be reconstructed are decoded by the preset noise reduction model to obtain reconstructed feature data of the digital model of the cultural relic.

2. The method for reducing noise of a digital model of cultural relics according to claim 1, characterized in that: The method of obtaining original feature data of the digital model of the cultural relic to be denoised includes: Collecting original data of the digital model of the cultural relic, and preprocessing the original data of the digital model of the cultural relic using a preset preprocessing method; wherein the preprocessing includes coordinate alignment, defect repair, and texture mapping; The pre-processed original data of the digital model of cultural relics are subjected to a preset standardization process to obtain original feature data of the digital model of cultural relics; wherein the standardization process includes redundancy removal, pixel-level segmentation, feature extraction and data enhancement.

3. The method for denoising a digital model of cultural relics according to claim 1, characterized in that: The preset noise reduction model includes an encoder, a geometric transformation layer and a decoder connected in sequence, wherein the encoder includes a plurality of coding layers connected in sequence; One coding layer is used to perform one iterative coding; The geometric transformation layer is used to perform the geometric transformation on the low-dimensional feature data to obtain the features to be reconstructed; The encoder is used to perform the decoding of the feature to be reconstructed to obtain reconstructed feature data of the digital model of the cultural relic.

4. The method for reducing noise of a digital model of cultural relics according to claim 3, characterized in that: The input features of this iterative encoding are linearly transformed using the preset denoising model, and the linear transformation results obtained include: Performing a linear transformation on the input features by using the weights and biases of the target coding layer through the target coding layer to obtain the linear transformation result; Among them, the target coding layer is the coding layer used to perform this iterative coding. If the target coding layer is the first coding layer, the input feature of the target coding layer is the original feature data. If the target coding layer is not the first coding layer, the input feature of the target coding layer is the output feature of the previous coding layer of the target coding layer.

5. The method for reducing noise of a digital model of cultural relics according to claim 4, characterized in that: Performing a projection transformation on the linear transformation result by using the preset denoising model and the corresponding linear weight matrix and the nonlinear weight matrix to obtain a projection transformation result, including: The linear transformation result is operated by the target coding layer using the hyperbolic tangent function to obtain a hyperbolic tangent result; the linear weight matrix of the target coding layer is multiplied by the linear transformation result to obtain a linear operation result, the nonlinear weight matrix of the target coding layer is multiplied by the hyperbolic tangent result to obtain a nonlinear operation result, and the linear operation result, the nonlinear operation result and the projection bias of the target coding layer are added to obtain the projection transformation result.

6. The method for reducing noise of a digital model of cultural relics according to claim 5, characterized in that: Normalizing the projection transformation result by the preset denoising model to obtain the activation output of this iterative encoding includes: Performing standard normalization on the projection transformation result through the target coding layer to obtain a standard normalization result; scaling and translating the standard normalization result to obtain the activation output; If the target coding layer is not the last coding layer, the activation output is output as the output feature to the next coding layer of the target coding layer through the target coding layer; if the target coding layer is the last coding layer, the activation output is output as the low-dimensional feature data to the geometric transformation layer through the target coding layer.

7. The method for reducing noise of a digital model of cultural relics according to claim 3, characterized in that: Performing geometric transformation on the low-dimensional feature data by using the preset denoising model to obtain features to be reconstructed includes: The geometric transformation layer uses a skew-symmetric transformation function and a scale transformation function to construct a geometric transformation matrix of the low-dimensional feature data; after performing a matrix exponential operation on the geometric transformation matrix, it is multiplied with the low-dimensional feature data to obtain the feature to be reconstructed.

8. The method for reducing noise of a digital model of cultural relics according to claim 3, characterized in that: The decoder includes a plurality of decoding layers connected in sequence, and decodes the features to be reconstructed through the preset noise reduction model, including: The decoding layer of each layer is sequentially performing a first nonlinear projection on the input feature based on a first nonlinear mapping function to obtain a first mapping feature; the first mapping feature is added to the bias of the decoding layer, and then multiplied by the transpose of the weight of the decoding layer to obtain a second mapping feature; the second mapping feature is nonlinearly projected based on the second nonlinear mapping function to obtain a nonlinear projection result of the decoding layer; wherein, if the decoding layer is the first decoding layer, the input feature of the decoding layer is the feature to be reconstructed; if the decoding layer is not the first decoding layer, the input feature of the decoding layer is the nonlinear projection result of the previous decoding layer; The decoder adds the nonlinear projection results of all the decoding layers to obtain the reconstructed feature data.

9. The method for reducing noise of a digital model of cultural relics according to claim 8, characterized in that: Adding the nonlinear projection results of all the decoding layers through the decoder to obtain the reconstructed feature data includes: Sequentially adding the nonlinear projection result of the decoding layer to the feature accumulation result output by the previous decoding layer through each decoding layer to obtain the feature accumulation result of the decoding layer; If the decoding layer is not the last decoding layer, outputting the feature accumulation result of the decoding layer to the next decoding layer through the decoding layer; If the decoding layer is the last decoding layer, the feature accumulation result of the decoding layer is output as the reconstructed feature data through the decoding layer.

10. The method for reducing noise of a digital model of cultural relics according to claim 1, characterized in that: The method for reducing noise of the digital model of cultural relics also includes: Acquiring training data, wherein the training data includes original feature data of a plurality of sample cultural relic digital models; Iterations are performed until the preset training completion condition is reached. Any iteration is a target iteration, which includes: Inputting the original feature data of each of the sample cultural relic digital models into the improved autoencoder to be trained, and obtaining the reconstructed feature data of each of the sample cultural relic digital models output by the improved autoencoder; Based on the reconstructed feature data and original feature data of all sample cultural relic digital models of the target iteration, a preset enhanced loss function is used to calculate the loss value of the target iteration; the enhanced loss function is constructed based on an error function and a regularization function that introduces periodic penalties; Based on the loss value of the target iteration, the model parameters in the model parameter set are updated using the gradient descent method, wherein the model parameters include the linear weight matrix and the nonlinear weight matrix corresponding to each iteration encoding; Determine whether the preset training completion conditions are met; If it is reached, the iteration is stopped, and the improved autoencoder is configured with each model parameter in the updated model parameter set to obtain the preset noise reduction model; If not reached, execute the next iteration.

11. A noise reduction device for a digital model of cultural relics, characterized in that: include: The original data acquisition unit is used to obtain the original feature data of the digital model of the cultural relic to be denoised; A model denoising unit is used to input the original feature data into a preset denoising model, and perform multiple iterative encodings through the preset denoising model to obtain low-dimensional feature data, wherein each iterative encoding includes: performing linear transformation on the input feature of this iterative encoding to obtain a linear transformation result; performing projection transformation on the linear transformation result using the corresponding linear weight matrix and nonlinear weight matrix to obtain a projection transformation result; normalizing the projection transformation result to obtain the activation output of this iterative encoding, and using the activation output of this iterative encoding as the input feature of the next iterative encoding; wherein the input feature of the first iterative encoding is the original feature data, and the activation output of the last iterative encoding is the low-dimensional feature data; performing geometric transformation on the low-dimensional feature data through the preset denoising model to obtain the feature to be reconstructed; decoding the feature to be reconstructed through the preset denoising model to obtain the reconstructed feature data of the digital model of the cultural relic.

12. An electronic device, characterized in that: comprising at least one processor and a memory connected to the processor, wherein: The memory is used to store computer programs; The processor is configured to execute the computer program so that the electronic device can implement the method for reducing noise of a digital model of cultural relics as claimed in any one of claims 1 to 10.

Images