An intelligent recognition system for 3D-like artifacts

Abstract

The present invention discloses an intelligent recognition system for 3D-like artifacts, which relates to the field of image recognition technology. The present invention obtains the shape and texture information of cultural relics through 3D scanning, and performs local and global feature extraction. In local feature extraction, for the recognition of core words and patterns, through feature capture of the CNN-RNN structure and local recognition of the UNet model, the core words and patterns are accurately recognized. Subsequently, through natural language processing technology, the recognition results are deeply understood and integrated with the historical and cultural background of the cultural relics to provide support for the comprehensive understanding of the cultural relics. The meanings of the core words and patterns are integrated with the internal structure of the cultural relics. Through the fusion of feature vectors and the analysis of the decision tree model, combined with historical documents, an evaluation model of the weathering degree is established to provide a basis for the classification and weathering degree judgment of cultural relics, realizing the overall understanding and comprehensive recognition of cultural relics.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, and particularly to an intelligent recognition system for 3D objects. Background Art

[0002] The identification of glass cultural relics is an important task involved by archaeologists when studying and analyzing archaeological relics. It requires a systematic observation of the appearance of glass products, including aspects such as shape, color, transparency, ornamentation, and degree of damage. These appearance features can provide clues about the possible age, manufacturing process, use, etc. of the glass products. Secondly, chemical analysis is one of the key steps in identifying glass cultural relics. Through technical means such as X-ray fluorescence analysis (XRF) or laser-induced breakdown spectroscopy (LIBS), archaeologists can determine the composition and its proportion of the glass, which helps to judge the manufacturing process and possible raw material sources of the glass products.

[0003] For example, the patent with publication number CN115238081A discloses a method, system, and readable storage medium for intelligent identification of cultural relics. The method includes obtaining cultural relic record data for reflecting the historical accumulation and modern practice scenarios of cultural relics, and generating a cultural relic knowledge graph based on the cultural relic record data; performing feature extraction on the collected cultural relic acquisition images containing the main object of the cultural relic to obtain feature data representing the features of the main object of the cultural relic; associating and matching the feature data with the classification data recorded in the preset image database and adapted to the main features of the cultural relic to obtain cultural relic classification information for cultural relic identification; based on the obtained cultural relic identification result, querying and feedbacking the cultural relic association information from the cultural relic knowledge graph. The implementation of this method can improve the efficiency and accuracy of cultural relic identification.

[0004] Although the above solution has the above advantages, traditional glass cultural relic identification usually based on SIFT and SURF feature extraction algorithms. Such extraction methods cannot fully capture the complex features of cultural relics, especially the extraction of detailed features such as core words and ornamentation, lacking adaptability to complex ornamentation and different styles of words, resulting in low recognition accuracy. At the same time, the recognition result cannot be effectively associated with the background information of the cultural relic, limiting the overall understanding ability of the cultural relic. Therefore, there is an urgent need for an intelligent recognition system for 3D objects to solve such problems. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art, the present invention provides an intelligent recognition system for 3D objects, which solves the problems in the prior art that there is a lack of adaptability to complex ornamentation and different styles of words, resulting in low recognition accuracy, and the recognition result cannot be effectively associated with the background information of the cultural relic, limiting the overall understanding ability of the cultural relic.

[0006] To achieve the above object, the present invention is implemented through the following technical solutions:

[0007] The present invention provides an intelligent recognition system for 3D-like artifacts, including:

[0008] A data acquisition module, which is responsible for obtaining 3D data from ancient glass cultural relics and preprocessing the collected 3D data, including denoising, filtering, and alignment; extracting and identifying classification features from the preprocessed 3D data, including surface texture and shape features; extracting local features and global features in the data acquisition module, where the local features extract the feature information of the local area of the cultural relic, and the global features extract the global feature information of the overall shape and size of the cultural relic;

[0009] A character and pattern recognition module, which uses deep learning and image processing technologies to recognize the core characters and patterns of ancient glass cultural relics and uses them as key features for subsequent processing; divides the cultural relic image into different regions, and then separately recognizes the core characters and pattern patterns in the cultural relic;

[0010] A historical and cultural background definition module, which, based on natural language processing technology, performs semantic understanding on the meanings of the recognized core characters and patterns, including the historical and cultural background of the cultural relic;

[0011] A cultural relic structure kernel design module, which integrates the meanings of the core characters and patterns output by the historical and cultural background definition module with the internal structure of the cultural relic; the internal structure of the cultural relic refers to the local features and global features output by the data acquisition module;

[0012] A classification and recognition module, which uses the extracted features and information, and adopts machine learning and deep learning algorithms to classify and recognize the cultural relics, uses the extracted features to identify the cultural relics and judge the degree of weathering, and outputs the corresponding recognition results.

[0013] The present invention is further set as follows: The local feature extraction step includes:

[0014] Performing feature extraction on each local area P i where i = 1, 2,..., N, and N is the number of local areas;

[0015] Performing local normal estimation, using the least squares method to fit the surface of each local area, and calculating the surface normal vector where, is the position vector of the j-th point in the local area, and k is the number of points in the local area;

[0016] Using the normal vector to calculate the curvature tensor K i , and then obtaining the principal curvatures k max and kmin , wherein, represents the curvature in the x - direction at (x, y, z), represents the change in curvature in the x - direction in the local region P i , represents the unit vector along the x - direction, represents the change in curvature simultaneously in the x - and y - directions in the local region P i , represents the product of the unit vectors along the x - direction and the y - direction, represents the curvature in the y - direction, that is, the change in curvature in the y - direction in the local region P i , represents the unit vector along the y - direction, H i represents the mean curvature, G i is the Gaussian curvature;

[0017] Then, using the color information and texture information of the local region, the gray - level co - occurrence matrix (GLCM) is adopted to extract the color information and texture information features;

[0018] The present invention is further set as: the global feature extraction step includes:

[0019] Adopting spherical harmonic transform to represent the three - dimensional shape of the cultural relic as a linear combination of spherical harmonic functions:

[0020] Performing dimensional measurement of the global features, calculating the dimensions of the bounding box and enclosing box of the cultural relic to describe the overall size and shape of the cultural relic;

[0021] The present invention is further set as: the recognition method of the core words and decorative patterns is:

[0022] Performing image pre - processing, using the DnCNN model to denoise the cultural relic image, and adopting the CLAHE algorithm to enhance the local contrast of the image;

[0023] Based on the pre - processed image data, using the DeepLab model to perform semantic segmentation on the cultural relic image;

[0024] Using a deep - learning model with a CNN - RNN structure that combines a convolutional neural network and a recurrent neural network to capture the core word and decorative features;

[0025] Using the UNet model to perform local segmentation on the cultural relic image, then inputting the segmented local region into the CNN - RNN model for recognition, and then selecting local feature descriptors SIFT and SURF to extract key features;

[0026] The present invention is further configured as follows: The core character and ornamentation feature capture step includes:

[0027] First, use a pre-trained CNN model as a feature extractor, input the cultural relic image into the CNN model, and obtain the feature representation of the image at a certain layer in the model through forward propagation;

[0028] Let the input cultural relic image be I. After passing through the CNN model, the feature map at the l-th layer is F (l) , F (l) = CNN(I);

[0029] Adopt a support vector machine SVM classifier. The loss function is to minimize the classification error and maximize the classification margin. Given the feature X and the label Y, the optimization objective of the SVM is expressed as: Among them, i1 is the retrieval of training samples, n1 is the number of samples, w is the weight vector, b is the bias term, C is the regularization parameter, (x i1 , y i1 ) is the training sample;

[0030] The training process minimizes the loss function through the gradient descent method and updates the model parameters w and b;

[0031] The present invention is further configured as follows: The way to perform semantic understanding on the core characters and ornamentations is:

[0032] Collect historical and cultural background text data related to cultural relics; use the Skip-gram model to map the words in the text data to a high-dimensional vector space;

[0033] The loss function of the Skip-gram model is: Among them, w t is the context vocabulary, w t+j is the target vocabulary in the context, T is the total number of contexts, p(w t+j |w t ) represents the conditional probability of the target vocabulary given the context, t represents the current word in the training corpus, j represents the relative position of the word in the context window, and m represents the size of the context window;

[0034] Adopt a pre-trained language model to perform semantic understanding and text generation on the text data; use the text representation generated by the model to infer semantic information;

[0035] Integrate the meanings of the core characters and ornamentations obtained from semantic understanding with historical and cultural background knowledge, and combine the regional, age, and usage information of the cultural relics;

[0036] Output the historical and cultural background definitions obtained after semantic understanding of the meanings of the core characters and ornamentations as one of the outputs of the cultural relic intelligent recognition system;

[0037] The present invention is further configured such that the manner of integrating the internal structure of the cultural relic with the meanings of the core words and patterns is as follows:

[0038] Perform element-by-element superposition on the feature vectors, and connect the feature vectors of the cultural relic with the semantic vectors of the cultural relic;

[0039] Then map the fused feature vectors to a high-dimensional representation space through a non-linear mapping function;

[0040] Adopt a decision tree model to analyze the importance of each dimension in the fused feature vectors, and determine the features that are most representative and discriminatory for the identification and classification of cultural relics;

[0041] Use the fused feature vectors as the comprehensive representation of the cultural relic;

[0042] The present invention is further configured such that the manner of integrating the internal structure of the cultural relic with the meanings of the core words and patterns is as follows:

[0043] Perform element-by-element superposition on the feature vectors, and connect the feature vectors of the cultural relic with the semantic vectors of the cultural relic;

[0044] Then map the fused feature vectors to a high-dimensional representation space through a non-linear mapping function;

[0045] Adopt a decision tree model to analyze the importance of each dimension in the fused feature vectors, and determine the features that are most representative and discriminatory for the identification and classification of cultural relics;

[0046] Use the fused feature vectors as the comprehensive representation of the cultural relic;

[0047] The present invention is further configured such that in the classification and recognition module, perform feature selection and preprocessing on the extracted features and information; use the extracted features and information to train with a recurrent neural network;

[0048] While classifying the cultural relics, use the extracted features to judge the weathering degree of the cultural relics;

[0049] The judgment of the weathering degree is evaluated by establishing an evaluation model of the weathering degree in combination with historical documents;

[0050] Then output the results of classification and judgment of the weathering degree;

[0051] The present invention is further configured such that the construction method of the evaluation model of the weathering degree is as follows:

[0052] Based on the indicators related to the weathering degree of ancient glass cultural relics, including surface color change, pattern clarity, and surface flatness, define and quantify each indicator;

[0053] Calculate the corresponding feature descriptors based on the data of surface texture and shape feature recognition collected by the data acquisition module;

[0054] Select the support vector machine and random forest method to establish a weathering degree evaluation model based on the extracted feature data;

[0055] Construct a corresponding weathering degree model for each index and evaluate the weathering degree with new cultural relic data.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] In the present invention, through deep learning and natural language processing technologies, comprehensive recognition and in-depth understanding of cultural relics are realized, including the recognition of core words and patterns and the integration of historical and cultural backgrounds, improving the comprehensiveness of recognition and in-depth understanding;

[0058] In the present invention, by combining multi-curvature calculation, texture feature extraction, and spherical harmonic transform for multi-angle feature extraction, the accuracy and robustness of feature extraction are improved, enabling the system to more comprehensively describe the features of cultural relics;

[0059] In the present invention, by combining various cultural relic features and historical literature information, an evaluation model of weathering degree is established to realize comprehensive consideration and accurate evaluation of the weathering degree of cultural relics. Brief Description of the Drawings

[0060] Figure 1 It is a framework diagram of the intelligent recognition system for 3D cultural relics of the present invention. Detailed Embodiments

[0061] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0062] It should be noted that the terms "first", "second", etc. in the description, claims and above-mentioned drawings of the present invention are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present invention described here can be implemented in an order other than those illustrated or described here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0063] The present invention will be further described in detail below with reference to the accompanying drawings:

[0064] Embodiment

[0065] Please refer to Figure 1 , the present invention provides an intelligent recognition system for 3D-like artifacts, including:

[0066] A data acquisition module, which is responsible for obtaining 3D data from ancient glass cultural relics and preprocessing the collected 3D data, including denoising, filtering, and alignment; extracting and identifying classification features from the preprocessed 3D data, including surface texture and shape features; local features and global features are extracted in the data acquisition module, the local features extract the feature information of the local area of the cultural relic, and the global features extract the global feature information of the overall shape and size of the cultural relic;

[0067] The local feature extraction steps include:

[0068] Feature extraction is performed on each local area P i where i = 1, 2,..., N, and N is the number of local areas;

[0069] Local normal estimation is performed, the surface of each local area is fitted using the least squares method, and the surface normal vector is calculated where, is the position vector of the j-th point in the local area, and k is the number of points in the local area;

[0070] Using the normal vector calculate the curvature tensor K i , and then obtain the principal curvatures k max and k min , where, represents the curvature in the x direction at (x, y, z), Indicates in the local area P i the curvature change in the x - direction, represents the unit vector along the x - direction, Indicates in the local area P i the curvature change simultaneously in the x - and y - directions, represents the product of the unit vectors along the x - direction and the y - direction, represents the curvature along the y - direction, that is, in the local area P i the curvature change in the y - direction, represents the unit vector along the y - direction, H i represents the mean curvature, G i is the Gaussian curvature;

[0071] Then, using the color information and texture information of the local area, the gray - level co - occurrence matrix GLCM is adopted to extract the color information and texture information features;

[0072] The global feature extraction steps include:

[0073] Using spherical harmonic transform to represent the three - dimensional shape of the cultural relic as a linear combination of spherical harmonic functions:

[0074] Conducting the dimension measurement of the global features, calculating the dimensions of the bounding box and enclosing box of the cultural relic to describe the overall size and shape of the cultural relic;

[0075] The core words and patterns recognition module, using deep learning and image - processing techniques, recognizes the core words and patterns of ancient glass cultural relics and uses them as key features for subsequent processing; the cultural relic image is segmented into different regions, and then the core words and pattern patterns in the cultural relic are recognized respectively;

[0076] The recognition methods for the core words and pattern patterns are:

[0077] Conducting image pre - processing, using the DnCNN model to denoise the cultural relic image and adopting the CLAHE algorithm to enhance the local contrast of the image;

[0078] Based on the pre - processed image data, using the DeepLab model for semantic segmentation of the cultural relic image;

[0079] Using a deep - learning model with a CNN - RNN structure that combines a convolutional neural network and a recurrent neural network to capture the core word and pattern features;

[0080] Using the UNet model to perform local segmentation on the cultural relic image, then inputting the segmented local regions into the CNN - RNN model for recognition, and then selecting local feature descriptors SIFT and SURF to extract key features;

[0081] The steps for capturing the core words and ornamentation features include:

[0082] First, use a pre-trained CNN model as a feature extractor. Input the cultural relic image into the CNN model, and through forward propagation, obtain the feature representation of the image at a certain layer in the model;

[0083] Let the input cultural relic image be I. After passing through the CNN model, the feature map at the l-th layer is F (l) , F (l) = CNN(I);

[0084] Adopt a support vector machine SVM classifier. The loss function is to minimize the classification error and maximize the classification margin. Given the feature X and the label Y, the optimization objective of SVM is expressed as: Among them, i1 is the retrieval of training samples, n1 is the number of samples, w is the weight vector, b is the bias term, C is the regularization parameter, (x i1 , y i1 ) is the training sample;

[0085] In the training process, minimize the loss function through the gradient descent method and update the model parameters w and b;

[0086] The historical and cultural background definition module, based on natural language processing technology, conducts semantic understanding on the meanings of the identified core words and ornamentation, including the historical and cultural background of the cultural relic;

[0087] The way to conduct semantic understanding on the core words and ornamentation is as follows:

[0088] Collect historical and cultural background text data related to the cultural relic; use the Skip-gram model to map the words in the text data to a high-dimensional vector space;

[0089] The loss function of the Skip-gram model is: Among them, w t is the context vocabulary, w t+j is the target vocabulary in the context, T is the total number of contexts, p(w t+j |w t ) represents the conditional probability of the target vocabulary given the context, t represents the current word in the training corpus, j represents the relative position of the word in the context window, and m represents the size of the context window;

[0090] Adopt a pre-trained language model to conduct semantic understanding and text generation on the text data; use the text representation generated by the model to infer semantic information;

[0091] Integrate the meanings of the core words and patterns obtained from semantic understanding with historical and cultural background knowledge, and combine the geographical location, age, and usage information of the cultural relics to deeply understand the cultural connotations represented by the core words and patterns;

[0092] Output the historical and cultural background definitions obtained after semantic understanding of the meanings of the core words and patterns, as one of the outputs of the cultural relic intelligent recognition system;

[0093] Cultural relic structure core design module, which integrates the meanings of the core words and patterns output by the historical and cultural background definition module with the internal structure of the cultural relics; the internal structure of the cultural relics refers to the local features and global features output by the data acquisition module;

[0094] The way to integrate the internal structure of the cultural relics with the meanings of the core words and patterns is as follows:

[0095] Perform element-by-element addition of the feature vectors, and connect the feature vector of the cultural relic with the semantic vector of the cultural relic;

[0096] Then map the fused feature vector to a high-dimensional representation space through a non-linear mapping function;

[0097] Adopt a decision tree model to analyze the importance of each dimension in the fused feature vector, and determine the features that are most representative and discriminative for the identification and classification of cultural relics;

[0098] Take the fused feature vector as the comprehensive representation of the cultural relic;

[0099] The way to analyze the importance of each dimension in the fused feature vector is as follows:

[0100] Take the fused feature vector as the input data, and at the same time prepare the corresponding cultural relic labels and categories as the output;

[0101] Use the CART decision tree algorithm to construct a decision tree model according to each dimension in the feature vector and the corresponding cultural relic labels;

[0102] After training, analyze the importance of each dimension in the fused feature vector through the feature information provided by the decision tree model;

[0103] Based on the feature importance scores of each node, perform the final cultural relic identification and classification features;

[0104] Classification and recognition module, which uses machine learning and deep learning algorithms to classify and recognize cultural relics by using the extracted features and information, uses the extracted features to identify cultural relics and judge the degree of weathering, and outputs the corresponding recognition results;

[0105] In the classification and recognition module, feature selection and preprocessing are performed on the extracted features and information; the extracted features and information are used to train a recurrent neural network;

[0106] While classifying cultural relics, the extracted features are used to judge the weathering degree of cultural relics;

[0107] The judgment of the weathering degree is evaluated by establishing an evaluation model of the weathering degree in combination with historical documents;

[0108] Then the results of classification and weathering degree judgment are output;

[0109] The construction method of the evaluation model of the weathering degree is as follows:

[0110] Based on the indicators related to the weathering degree of ancient glass cultural relics, including surface color change, ornament clarity, and surface flatness, each indicator is defined and quantified;

[0111] Based on the surface texture and shape feature recognition data obtained by the data acquisition module, the corresponding feature descriptors are calculated;

[0112] Support vector machines and random forest methods are selected to establish an evaluation model of the weathering degree based on the extracted feature data;

[0113] An corresponding weathering degree model is constructed for each indicator, and the weathering degree of new cultural relic data is evaluated.

[0114] In the present invention, the shape and texture information of cultural relics are obtained through 3D scanning, and local and global feature extraction is carried out. In local feature extraction, curvature calculation and texture feature extraction are used to describe the curvature and texture features of local regions, and the feature information of local regions of cultural relics is mined; in the global feature extraction stage, the spherical harmonic transform is used to represent the three-dimensional shape of cultural relics as a linear combination of spherical harmonic functions, describing the overall shape and size features of cultural relics; for the recognition of core words and ornaments, through the feature capture of the CNN-RNN structure and the local recognition of the UNet model, the core words and ornaments are accurately recognized; subsequently, through natural language processing technology, the recognition results are deeply understood and integrated with the historical and cultural background of cultural relics to provide support for the comprehensive understanding of cultural relics, and the meanings of core words and ornaments are integrated with the internal structure of cultural relics. Through the fusion of feature vectors and the analysis of decision tree models, the most representative and discriminatory features for the recognition and classification of cultural relics are extracted. In the classification and recognition module, a recurrent neural network is used for training, and an evaluation model of the weathering degree is established in combination with historical documents, providing a basis for the classification and weathering degree judgment of cultural relics, and realizing the overall understanding and comprehensive recognition of cultural relics.

[0115] The above content is only to illustrate the technical idea of the present invention and shall not be used to limit the protection scope of the present invention. Any modification made on the basis of the technical solution according to the technical idea proposed by the present invention shall fall within the protection scope of the claims of the present invention.

Claims

1. An intelligent recognition system for 3D objects, characterized in that: include: The data acquisition module is responsible for acquiring 3D data from ancient glass artifacts and preprocessing the acquired 3D data, including denoising, filtering, and alignment; Extract and identify classification features from preprocessed 3D data, including surface texture and shape features; In the data acquisition module, local features and global features are extracted. Local features extract feature information of the local area of ​​the cultural relic, and global features extract global feature information of the shape and size of the cultural relic as a whole. The word pattern recognition module uses deep learning and image processing technology to identify the core words and patterns of ancient glass artifacts, and uses them as key features for subsequent processing; it divides the artifact image into different areas, and then identifies the core words and patterns in the artifacts respectively; The historical and cultural background definition module, based on natural language processing technology, provides semantic understanding of the meaning of the identified core characters and patterns, including the historical and cultural background of the cultural relics; The cultural relic structure core design module integrates the meaning of the core words and patterns output by the history and cultural background definition module with the internal structure of the cultural relic; the internal structure of the cultural relic refers to the local and global features output by the data acquisition module; The classification and recognition module uses the extracted features and information to classify and identify cultural relics using machine learning and deep learning algorithms, uses the extracted features to identify cultural relics and determine the degree of weathering, and outputs the corresponding recognition results; The local feature extraction steps include: For each local region P i Perform feature extraction, where i = 1, 2, ..., N, N is the number of local regions; Perform local normal estimation, use the least squares method to fit the surface of each local area, and calculate the surface normal vector in, is the position vector of the jth point in the local area, k is the number of points in the local area; Using the normal vector Calculate the curvature tensor K i , and then find the principal curvature k max and k min , in, represents the curvature in the x direction at (x,y,z), Indicates that in the local area P i In the x-direction, the curvature changes. represents the unit vector along the x direction, Indicates that in the local area P i In the case of a curve, the curvature changes in both the x and y directions. represents the product between unit vectors along the x-direction and the y-direction, represents the curvature along the y direction, that is, in the local area P i In the y direction, the curvature changes. represents the unit vector along the y direction, H i represents the mean curvature, G i is the Gaussian curvature; Then, the color information and texture information of the local area are used to extract the color information and texture information features using the gray level co-occurrence matrix GLCM; The global feature extraction steps include: The spherical harmonic transform is used to represent the three-dimensional shape of the artifact as a linear combination of spherical harmonic functions: Measure the size of global features and calculate the size of the bounding box and the bounding box of the cultural relics; The core characters and decorative patterns can be identified as follows: Perform image preprocessing, use the DnCNN model to denoise the cultural relics image, and use the CLAHE algorithm to enhance the local contrast of the image; Based on the preprocessed image data, the DeepLab model is used to perform semantic segmentation of cultural relic images; Use the CNN-RNN structure deep learning model that combines convolutional neural network and recurrent neural network to capture the core characters and decorative features; The UNet model is used to perform local segmentation on the cultural relic image, and then the segmented local area is input into the CNN-RNN model for recognition, and then the local feature descriptors SIFT and SURF are used to extract key features; The steps to capture core characters and decorative features include: First, use the pre-trained CNN model as a feature extractor, input the cultural relic image into the CNN model, and obtain the feature representation of the image at a certain layer in the model through forward propagation; Suppose the input cultural relic image is I. After passing through the CNN model, the feature map at the lth layer is F. (l) , F (l) =CNN(I); The support vector machine SVM classifier is used, and the loss function is to minimize the classification error and maximize the classification interval. Given the feature X and label Y, the optimization goal of SVM is expressed as: Among them, i1 is the retrieval of training samples, n1 is the number of samples, w is the weight vector, b is the bias term, C is the regularization parameter, (x i1 ,y i1 ) is a training sample; The training process minimizes the loss function through the gradient descent method and updates the model parameters w and b; The way to understand the core words and patterns semantically is: Collect historical and cultural background text data related to cultural relics; use the Skip-gram model to map words in the text data to a high-dimensional vector space; The loss function of the Skip-gram model is: Among them, w t is the context word, w t+j is the target word in the context, T is the total number of contexts, p(w t+j |w t ) represents the conditional probability of the target word given the context, t represents the current word in the training corpus, j represents the relative position of the word in the context window, and m represents the size of the context window; Use pre-trained language models to perform semantic understanding and text generation on text data; use the text representation generated by the model to infer semantic information; The meaning of the core characters and patterns obtained through semantic understanding is integrated with historical and cultural background knowledge, and combined with the region, age, and use information of the cultural relics; Output the historical and cultural background definitions obtained through semantic understanding of the meaning of the core characters and patterns as one of the outputs of the cultural relics intelligent identification system; The internal structure of the cultural relic is integrated with the meaning of the core words and patterns in the following ways: The feature vectors are superimposed element by element, and the feature vectors of the cultural relics are connected with the semantic vectors of the cultural relics; Then the fused feature vector is mapped to a high-dimensional representation space through a nonlinear mapping function; A decision tree model is used to analyze the importance of each dimension in the fused feature vector to determine the most representative and distinguishing features for the identification and classification of cultural relics; The fused feature vector is used as a comprehensive representation of the artifacts.

2. The intelligent recognition system for 3D objects according to claim 1, characterized in that: In the classification and recognition module, feature selection and preprocessing are performed on the extracted features and information; the extracted features and information are used to train the recurrent neural network; While classifying the cultural relics, the extracted features are used to determine the degree of weathering of the cultural relics; The weathering degree is determined by combining with historical documents and establishing a weathering degree assessment model for assessment; Then output the results of classification and weathering degree judgment.

3. The intelligent recognition system for 3D objects according to claim 2, characterized in that: The weathering degree assessment model is constructed as follows: Based on indicators related to the weathering degree of ancient glass artifacts, including surface color change, pattern clarity, and surface flatness, each indicator is defined and quantified; Based on the surface texture and shape feature recognition data obtained by the data acquisition module, the corresponding feature descriptors are calculated; Support vector machine and random forest method were used to establish a weathering degree assessment model based on the extracted feature data; A corresponding weathering degree model is constructed for each indicator, and the weathering degree is evaluated using new cultural relic data.

Images