A bronze inscription recognition method and system

Abstract

The application discloses a bronze inscription recognition method and system, and relates to the field of inscription recognition, and comprises: a field acquisition module, which is used for scanning and acquiring the global intrinsic physical field of the inscription casting area of the bronze ware, and acquiring the multi-dimensional field source original data corresponding to the area; a field decoupling module, which is used for decoupling and separating the field dimension of the multi-dimensional field source original data, and eliminating the field signal data irrelevant to the inscription casting behavior; the application realizes scanning and acquisition of the global intrinsic physical field of the inscription casting area of the bronze ware, decoupling and separating the field dimension, accurately eliminating the field signal data irrelevant to the inscription casting, acquiring the effective field signal corresponding to the inscription casting, improving the recognition stability and accuracy of the damaged and blurred inscription, and finally outputting the standardized and structured inscription data, thereby providing support for the archaeological research of the bronze inscription.

Description

Technical Field

[0001] This invention relates to the field of inscription recognition technology, specifically to a method and system for recognizing inscriptions on bronze artifacts. Background Technology

[0002] Bronze inscriptions are China's earliest systematic historical archives, carrying important information on ancient politics, economy, and military affairs. They are the core basis for archaeological research and textual research on the evolution of writing, and their accurate identification has irreplaceable value for verifying historical records, supplementing history, and inheriting ancient culture.

[0003] The invention patent application with application number 202211007559.8 discloses a method, device, computer equipment and storage medium for identifying bronze inscriptions. The application aims to solve the problem that "in order to observe or restore the inscription information, it is often necessary to physically or chemically remove the rust on it, but this may damage the bronze. The existing technology is difficult to accurately identify the inscriptions on bronze without damaging the bronze."

[0004] However, existing technologies still cannot effectively solve the problem of incompleteness and blurring of bronze inscriptions caused by erosion over time, making it difficult to accurately identify such damaged inscriptions. This fails to meet the actual needs of archaeological work for efficient and accurate identification of inscriptions. Furthermore, in such identification scenarios, traditional optical identification relies on surface vision, which cannot identify rusted / damaged inscriptions, and cleaning can easily damage cultural relics. CT scans only obtain morphological information and do not have physical field characteristics, making it impossible to distinguish between casting and rust.

[0005] Therefore, we propose a method and system for identifying inscriptions on bronze artifacts. Summary of the Invention

[0006] In view of the above-mentioned shortcomings of the existing technology, the present invention provides a method and system for identifying inscriptions on bronzes, which can effectively solve the problems of the existing technology.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions; This invention discloses a system for identifying inscriptions on bronze artifacts, comprising: The system comprises the following modules: a field acquisition module for scanning and acquiring the intrinsic physical field of the entire bronze inscription casting area, obtaining the original multi-dimensional field source data corresponding to that area; a field decoupling module for decoupling and separating the original multi-dimensional field source data in the field dimension, removing field signal data unrelated to the inscription casting behavior; an extraction module for locating and extracting the stroke topology nodes formed by the inscription casting from the decoupled field signal data, establishing an independent node feature set; a reconstruction module for connecting the stroke topology nodes in the node feature set according to the inherent arrangement logic of the inscription casting, generating a complete inscription stroke topology link; an identification module for comparing the complete inscription stroke topology link with a preset inscription topology feature library link by link, identifying the corresponding character attribution of the inscription; and an output module for integrating the character attribution identification results, corresponding topology link information, and spatial location information, outputting standardized data and completing structured storage. The field acquisition module is interactively connected to the field decoupling module via a wireless network signal. The field decoupling module is interactively connected to the extraction module via a wireless network signal. The extraction module is interactively connected to the reconstruction module via a wireless network signal. The reconstruction module is interactively connected to the recognition module via a wireless network signal. The recognition module is interactively connected to the output module via a wireless network signal.

[0008] Furthermore, the field acquisition module synchronously scans and acquires the intrinsic physical field of the entire area of ​​the bronze inscription casting region through a three-axis linkage scanning path with the same spatial origin. The intrinsic physical field of the entire area includes the residual magnetic field, stress field and micro-topography elevation field of the casting region. The field acquisition module uses the spatial coordinates of the scanning sampling points as a unified index to stitch together the multi-physical field sampling values ​​under the same spatial coordinates to generate multi-dimensional field source raw data. At the same time, it dynamically adapts the density of the scanning sampling points, increasing the sampling point density in the edge area of ​​the casting marks and decreasing the sampling point density in the flat area without casting marks.

[0009] Furthermore, the dynamic adaptation of the scan sampling point density follows the following: ; In the formula: The density of scan sampling points at planar coordinates (x, y) within the inscription casting area; The preset base sampling point density; The gradient magnitude of the pre-scan field signal at the plane coordinates (x, y); is the preset baseline gradient threshold; k is the preset density adaptation coefficient; The preset maximum sampling point density; This is the preset minimum sampling point density.

[0010] Furthermore, the field decoupling module first aligns the original multi-dimensional field source data according to spatial coordinates to construct a field dimension feature matrix. It then performs singular value decomposition on the feature matrix to obtain the eigenvectors and singular values ​​corresponding to each field dimension. Finally, based on the field mutation characteristics corresponding to the inscription casting behavior, it decouples and separates the field signals of each dimension, eliminating field signal data unrelated to the inscription casting behavior. ; In the formula: The unit eigenvector corresponding to the j-th field dimension Standard casting feature unit vector corresponding to pre-defined inscription casting behavior Cosine similarity; express and The angle between the vectors; Let j be the singular values ​​corresponding to the field dimensions; It is the sum of the singular values ​​corresponding to all field dimensions; This is a preset threshold for the proportion of singular values.

[0011] Furthermore, the extraction module first performs spatial gradient calculation on the decoupled field signal data to obtain the field gradient magnitude and gradient direction of each sampling point. The sampling points whose field gradient magnitude exceeds the preset gradient threshold and whose gradient direction conforms to the casting stroke direction constraint are marked as stroke edge sampling points. Then, the topological skeleton is extracted from all stroke edge sampling points to obtain the single-pixel width skeleton line of the inscription stroke. The endpoints, intersections, and inflection points of the skeleton line are uniformly marked as stroke topological nodes. The extraction module extracts the spatial coordinates, field gradient features, and neighborhood field distribution features corresponding to each stroke topology node, and establishes an independent feature vector for each topology node. The feature vectors of all topology nodes together constitute the node feature set.

[0012] Furthermore, the reconstruction module performs link connection on the stroke topology nodes in the node feature set based on the consistency of stroke direction and spatial distance constraints, to generate a complete inscription stroke topology link. The necessary and sufficient condition for two stroke-based topology nodes to be linked is set as follows: ; In the formula: for , Euclidean distance in a spatial coordinate system; , For any two stroke topology nodes to be connected in the node feature set; The preset maximum connection distance threshold; for , The angle between the gradient directions of the corresponding skeleton lines; The preset maximum directional angle threshold; After two nodes that meet the necessary and sufficient conditions are connected, they form a continuous stroke topology link in the order of connection. Finally, all connected stroke topology links together constitute a complete inscription stroke topology link.

[0013] Furthermore, during the operation phase of the recognition module, the scale and rotation of the generated complete inscription stroke topology links are first normalized to eliminate feature deviations caused by spatial scale and deflection angle, so as to obtain the standard topology link features to be recognized. Then, the standard topology link features are compared with the standard character topology features in the preset inscription topology feature library on a link-by-link basis to calculate the isomorphism matching degree between the features to be recognized and the standard features in the library. When the isomorphic matching degree exceeds the preset matching threshold, the character information of the corresponding standard character in the library is marked as the character belonging to the stroke topology link of the inscription; if the isomorphic matching degree does not exceed the preset matching threshold, the link is marked as a character to be verified, and its link features are extracted and stored in the feature set to be verified.

[0014] Furthermore, the output module uses the spatial coordinate system of the inscription casting area as a reference to bind the identified inscription character attribution results, corresponding stroke topology link information, and character spatial position information one by one, generating a standardized data group with the character as the smallest unit. The standardized data set includes text information of characters, topology link vector data, spatial bounding box coordinates, and recognition matching degree information; The output module sorts and encodes all standardized data groups according to the inherent arrangement order of bronze inscriptions, generates a structured dataset, and finally stores the structured dataset into a preset inscription database. At the same time, it generates a standardized export file that is compatible with general archaeological information systems.

[0015] On the other hand, a method for identifying inscriptions on bronze artifacts includes: A three-axis coordinate system is established with the geometric center of the inscription casting area as the origin. Multi-dimensional intrinsic physical field data are collected synchronously, and spatially aligned and stitched to generate multi-dimensional field source raw data, dynamically adapting the sampling density. A feature matrix is ​​constructed from the multi-dimensional field source data and singular value decomposition is performed. Field signals are decoupled according to casting behavior characteristics, irrelevant interference data is removed, and effective field signals are retained. Gradient calculation is performed on the effective field signals, stroke edges are marked, and single-pixel skeleton lines are extracted. Stroke topology nodes are calibrated, and node feature sets are constructed. Based on stroke direction and spatial distance constraints, node connection conditions are verified, and node link connections are completed according to the inherent logic of casting, generating complete inscription topology links. The topology links are scaled and rotated for normalization, and isomorphic comparison is performed with the standard inscription topology feature library. The character assignment is determined according to the matching degree, and characters that do not meet the standard are marked as characters to be verified. The recognition results, topology links, and spatial information are bound to generate standardized data, and sorting and encoding are completed according to the inscription arrangement, realizing structured storage and standardized file export.

[0016] Compared with the known prior art, the technical solution provided by this invention has the following beneficial effects: This invention achieves spatial alignment of multiple field source data by scanning and acquiring the intrinsic physical field of the entire area of ​​bronze inscription casting and decoupling the field dimension. It accurately eliminates field signal data unrelated to the inscription casting, obtains the effective field signal corresponding to the inscription casting, and balances the acquisition accuracy and data processing efficiency through unified spatial coordinates. It avoids the interference of bronze surface corrosion and wear on inscription recognition, locates the topological nodes of inscription strokes and restores the complete topological links of inscription strokes. Based on standardized topological feature comparison, it completes the identification of inscription element attribution, improves the stability and accuracy of the identification of damaged and blurred inscriptions, and finally outputs standardized and structured inscription data, providing further technical support for the archaeological research of bronze inscriptions. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0018] Figure 1 A schematic diagram of the structure of an inscription identification system on bronze artifacts; Figure 2 This is a flowchart illustrating a method for identifying inscriptions on bronze artifacts. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0020] The present invention will be further described below with reference to embodiments.

[0021] Example 1: This embodiment provides a system for identifying inscriptions on bronze artifacts, such as... Figure 1 As shown, it includes: The field acquisition module is used to scan and acquire the intrinsic physical field of the entire area of ​​bronze inscription casting area, and obtain the original data of multi-dimensional field source corresponding to the area. The field acquisition module synchronously scans and acquires the intrinsic physical field of the entire area of ​​the bronze inscription casting area through a three-axis linkage scanning path with the same spatial origin. The intrinsic physical field of the entire area includes the residual magnetic field, stress field and micro-topography elevation field of the casting area. The field acquisition module uses the spatial coordinates of the scanning sampling points as a unified index to stitch together the multi-physical field sampling values ​​under the same spatial coordinates to generate multi-dimensional field source raw data. At the same time, it dynamically adapts the density of the scanning sampling points, increasing the sampling point density in the edge area of ​​the casting marks and decreasing the sampling point density in the flat area without casting marks. The three-axis linkage scanning path uses the geometric center point of the bronze inscription casting area as the only fixed spatial origin, constructing an orthogonal X-axis, Y-axis, and Z-axis three-axis scanning coordinate system. The X-axis and Y-axis form a scanning plane parallel to the inscription casting reference plane, and the Z-axis is the normal axis perpendicular to the reference plane. The scanning execution end is equipped with a multi-physics field sensing unit, and the three-axis drive mechanism corresponds to the lateral translation of the X-axis, the longitudinal translation of the Y-axis, and the normal follow-up of the Z-axis, respectively. The motion control signals of the three axes share the same origin clock reference and spatial coordinate encoding system, realizing the synchronous linkage of the three-axis motion. During the scanning process, the X-axis and Y-axis drive the sensing unit to complete the planar full-area traversal scanning of the casting area, and the Z-axis performs normal position tracking compensation in real time according to the surface morphology of the casting area to ensure that the normal distance between the sensing unit and the surface of the casting area is always kept within the preset sampling distance range. The spatial coordinates of all scanning sampling points are uniformly calibrated with this fixed origin as the reference, so as to achieve spatial alignment and dimensional matching of multi-physics sampling data under the same spatial coordinates. The dynamic adaptation of the scan sampling point density follows the following rules: ; In the formula: The density of scan sampling points at planar coordinates (x, y) within the inscription casting area; The preset base sampling point density; The gradient magnitude of the pre-scan field signal at the plane coordinates (x,y) represents the degree of abrupt change in the field signal at that location; is the preset baseline gradient threshold, corresponding to the upper limit of the field signal gradient in the flat area without casting marks; k is the preset density adaptation coefficient; The preset maximum sampling point density; The preset minimum sampling point density; The above formula uses a preset basic sampling density as a benchmark, and performs an exponential operation to adjust the density based on the degree of abrupt change in the casting marks reflected by the field signal gradient magnitude. Then, it uses the maximum and minimum sampling density thresholds to limit the boundaries, which can automatically increase the sampling density at the edge of the casting marks and decrease the sampling density in flat areas, thus balancing recognition accuracy and scanning efficiency. Where k is a constant greater than 0, which is proportional to the sampling signal-to-noise ratio of the multi-physics sensing unit and the target accuracy requirement for inscription recognition; The field decoupling module is used to decouple and separate the original data of multi-dimensional field sources in the field dimension, and remove field signal data that are unrelated to the inscription casting behavior; Among them, the casting process will cause abrupt changes in the field signal (remanence enhancement, stress concentration, elevation change). After decoupling, only such abrupt change signal is retained, which can completely eliminate the interference of corrosion and substrate noise on inscription recognition. The field decoupling module first aligns the original multi-dimensional field source data according to spatial coordinates to construct a field dimension feature matrix. It then performs singular value decomposition on the feature matrix to obtain the eigenvectors and singular values ​​corresponding to each field dimension. Next, based on the field mutation characteristics corresponding to the inscription casting behavior, it decouples and separates the field signals of each dimension, eliminating field signal data unrelated to the inscription casting behavior. ; In the formula: The unit eigenvector corresponding to the j-th field dimension Standard casting feature unit vector corresponding to pre-defined inscription casting behavior The cosine similarity represents the correlation between the j-th field dimension and the casting behavior; express and The angle between the vectors; Let j be the singular values ​​corresponding to the field dimensions, representing the proportion of field signal energy in that field dimension; It is the sum of the singular values ​​corresponding to all field dimensions; This is a preset threshold for the proportion of singular values; The above formula uses cosine similarity to measure the correlation between field dimension and casting behavior. By combining the singular value energy ratio and exponential operation to set the elimination threshold, it can accurately separate casting-related field signals and interference data, so as to eliminate the dimensional differences and baseline shifts of different fields and make field signal processing purer and more accurate. When performing singular value decomposition on the feature matrix, the original multidimensional field source data aligned according to spatial coordinates is first constructed into an M×N dimensional field dimension feature matrix F, where M is the total number of scanning sampling points, N is the total number of field dimensions of the acquired intrinsic physical fields, and the element F(i,j) in the i-th row and j-th column of matrix F is the field signal sampling value of the i-th spatial sampling point in the j-th field dimension. Next, the feature matrix F is preprocessed by centering. This involves subtracting the arithmetic mean of all elements in each column from the elements in that column, resulting in the centered feature matrix. Eliminate dimensional differences and baseline shifts in different field dimensions; Finally, the centered M×N feature matrix Perform singular value decomposition, the decomposition formula is: Where U is an M×N dimensional left singular matrix, corresponding to the eigenvectors of the spatial distribution of sampling points; Σ is an N×N dimensional diagonal matrix, with the elements on the diagonal arranged in descending order of value, representing the singular values ​​corresponding to each field dimension. V is an N×N dimensional right singular matrix, whose column vectors are the unit eigenvectors corresponding to each field dimension, and the j-th column vector... The unit eigenvector corresponding to the j-th field dimension, and the singular values One-to-one correspondence; The extraction module is used to locate and extract the stroke topology nodes formed by the inscription casting in the decoupled field signal data, and establish an independent node feature set. The extraction module first performs spatial gradient calculation on the decoupled field signal data to obtain the field gradient magnitude and gradient direction of each sampling point. The sampling points whose field gradient magnitude exceeds the preset gradient threshold and whose gradient direction conforms to the casting stroke direction constraint are marked as stroke edge sampling points. Then, the topological skeleton is extracted from all stroke edge sampling points to obtain the single-pixel width skeleton line of the inscription stroke. The endpoints, intersections, and inflection points of the skeleton line are uniformly marked as stroke topological nodes. The extraction module extracts the spatial coordinates, field gradient features, and neighborhood field distribution features of each stroke topology node, and establishes an independent feature vector for each topology node. The feature vectors of all topology nodes together constitute the node feature set. The process of extracting the topological skeleton from all stroke edge sampling points includes the following steps: Using all the marked stroke edge sampling points as boundary references, a closed connected region of the cast stroke is constructed based on the spatial coordinates of the sampling points. For each sampling point within the closed connected region, the Euclidean distance from the sampling point to the nearest stroke edge sampling point is calculated by combining its field gradient magnitude and spatial location information, and denoted as the field distance feature value of the sampling point. For all sampling points within a closed connected region, prioritize them according to the field distance feature value from high to low. Sampling points are removed sequentially according to the sorting result. After each removal, the topological connectivity of the closed connected region is checked. If the topological connectivity of the closed connected region does not change after removing the sampling point, the removal of the sampling point is completed. If the topological connectivity changes after removal, the sampling point is retained. The removal and verification operations are repeated until the lines formed by the remaining sampling points within the closed connected region retain only a single pixel width of connected structure. The obtained single-pixel width connected structure is subjected to topological smoothing and denoising processing to remove isolated short line segments with a length less than a preset length threshold. At the same time, the jagged edges of the lines are corrected by segmented fitting based on the stroke direction constraint, and finally the single-pixel width skeleton line of the inscription stroke is obtained. The reconstruction module is used to connect the stroke topology nodes in the node feature set according to the inherent arrangement logic of the inscription casting, and generate a complete inscription stroke topology link. The reconstruction module connects the stroke topology nodes in the node feature set based on the consistency of stroke direction and spatial distance constraints to generate a complete inscription stroke topology link. The necessary and sufficient condition for two stroke-based topology nodes to be linked is set as follows: ; In the formula: for , Euclidean distance in a spatial coordinate system; , For any two stroke topology nodes to be connected in the node feature set; The preset maximum connection distance threshold; for , The angle between the gradient directions of the corresponding skeleton lines; The preset maximum directional angle threshold; The above formula takes the Euclidean distance between nodes and the angle between stroke directions as core parameters. Through exponential operation, the angle constraint is incorporated into the distance threshold, clarifying the necessary and sufficient conditions for node connection. This not only conforms to the spatial distribution law of inscription strokes but also follows the direction logic of cast strokes, making the topology link reconstruction more in line with the real form of inscriptions and improving the integrity of stroke restoration. After two nodes that meet the necessary and sufficient conditions are connected, they form a continuous stroke topology link in the order of connection. Finally, all connected stroke topology links together constitute a complete inscription stroke topology link. The identification module is used to compare the complete stroke topology link of the inscription with the preset inscription topology feature library link by link to identify the character attribution of the corresponding inscription. During the recognition module's operation phase, the scale and rotation of the generated complete inscription stroke topology links are first normalized to eliminate feature deviations caused by spatial scale and deflection angle, thus obtaining the standard topology link features to be recognized. Then, the standard topology link features are compared with the standard character topology features in the preset inscription topology feature library on a link-by-link basis to calculate the isomorphic matching degree between the features to be recognized and the standard features in the library. When the isomorphic matching degree exceeds the preset matching threshold, the character information of the corresponding standard character in the library is marked as the character belonging to the stroke topology link of the inscription; if the isomorphic matching degree does not exceed the preset matching threshold, the link is marked as a character to be verified, and its link features are extracted and stored in the feature set to be verified. When calculating the isomorphic matching degree between the feature to be identified and the standard features in the library, the standard topological link feature to be identified and the standard character topological feature to be compared in the library are first decomposed into basic topological units containing topological nodes, node connection relationships, stroke direction, stroke intersection and closed structure according to the unified decomposition rules. Then, it is checked whether the number of topological nodes of the two types of features is within the preset error tolerance range. If it exceeds the range, the matching degree is 0. If it meets the range, a one-to-one mapping relationship is established between the topological nodes of the two types of features. If the full node mapping cannot be completed, the matching degree is 0. After the mapping is completed, the matching scores of the four dimensions of node connection relationship, stroke direction, intersection and closed structure, and relative spatial position of nodes are calculated respectively. The scores of each dimension are fused according to the preset weight to obtain the isomorphic matching degree with a value range of 0 to 1. The closer the value is to 1, the higher the topological isomorphism between the two. The preset inscription topology feature library is a pre-built standardized feature library. Each standard character in the library corresponds to a unique character text attribution information, as well as standard topology link features generated using the same normalization and topology decomposition rules as the features to be identified. The output module is used to integrate the identification results of the inscription characters, the corresponding topological link information and spatial location information, output standardized data and complete structured storage. The output module uses the spatial coordinate system of the inscription casting area as a reference to bind the identified inscription character attribution results, corresponding stroke topology link information, and character spatial position information one by one, generating a standardized data group with the character as the smallest unit. The standardized data set includes textual information of characters, topological link vector data, spatial bounding box coordinates, and recognition matching information; The output module sorts and encodes all standardized data groups according to the inherent arrangement order of bronze inscriptions, generates a structured dataset, and finally stores the structured dataset into a preset inscription database, while generating a standardized export file that is compatible with general archaeological information systems. The field acquisition module is interconnected with the field decoupling module via a wireless network signal. The field decoupling module is interconnected with the extraction module via a wireless network signal. The extraction module is interconnected with the reconstruction module via a wireless network signal. The reconstruction module is interconnected with the recognition module via a wireless network signal. The recognition module is interconnected with the output module via a wireless network signal.

[0022] In this embodiment, the field acquisition module scans and acquires the intrinsic physical field of the entire area of ​​the bronze inscription casting region, obtaining the original multi-dimensional field source data corresponding to the region. The field decoupling module simultaneously decouples and separates the original multi-dimensional field source data in the field dimension, removing field signal data that is irrelevant to the inscription casting behavior. The extraction module then extracts the stroke topology nodes formed by the inscription casting from the decoupled field signal data, establishing an independent node feature set. The reconstruction module further connects the stroke topology nodes in the node feature set according to the inherent arrangement logic of the inscription casting, generating a complete inscription stroke topology link. The recognition module then compares the complete inscription stroke topology link with the preset inscription topology feature library link by link to identify the character attribution of the corresponding inscription. Finally, the output module integrates the character attribution recognition results, corresponding topology link information, and spatial location information, outputting standardized data and completing structured storage.

[0023] In the above embodiments, the system can acquire multi-dimensional data of the inscription casting area through multi-physics field scanning in the actual work of bronze archaeological research and digital archiving of cultural relics. It can effectively deal with the identification problems caused by surface corrosion and damage of cultural relics, balance the acquisition efficiency and detail accuracy with a dynamically adapted sampling method, complete character recognition by restoring the topological link of inscription strokes and standardizing comparison, and finally output structured data compatible with existing archaeological systems, reducing the workload of manual identification and subjective errors.

[0024] It should be noted that: The residual magnetic field was acquired using a fluxgate magnetometer with a range of ±100μT and a resolution of 1nT. The stress field was acquired using a micro-strain sensor with a range of ±5000με and a resolution of 0.1με. The micro-topography elevation field was acquired using a laser displacement sensor with a range of ±5mm and a resolution of 0.1μm. The three types of sensors were integrated into a coaxial non-contact probe at the triaxial scanning execution end, which was fixed by a flexible bracket and maintained at a constant normal distance of 1-3mm from the surface of the bronze artifact. The entire process was conducted without physical contact to avoid damage to the artifact.

[0025] The three-axis mechanism is driven by a high-precision linear motor. The maximum scanning stroke of the X and Y axes covers 200mm×200mm with a positioning accuracy of ±0.01mm. The Z axis adopts closed-loop servo follow-up control. It relies on the laser displacement sensor to provide real-time feedback of surface elevation data and dynamically adjusts the probe height to maintain the sampling interval. The pre-scan uses low-density rapid traversal to obtain the field signal gradient distribution, providing input basis for dynamic sampling density. In addition, the three-axis mechanism supports replaceable stroke modules: 50mm×50mm stroke for small devices and segmented splicing scanning for large devices; the Z-axis adopts laser non-contact follow-up with a minimum spacing of 1mm and no physical contact throughout the entire process; Before data collection, the electromagnetic shielding cover (shielding effectiveness ≥60dB@10kHz-1GHz) is turned on to eliminate environmental magnetic field interference; a standard bronze test block without casting or corrosion is used for zero-point calibration, which is automatically calibrated every 10 minutes to correct sensor drift error.

[0026] A right-handed Cartesian coordinate system was established with the geometric center of the inscription casting area as the origin. The coordinates of the sampling points were encoded in a three-dimensional floating-point (X,Y,Z) format with precision retained to four decimal places. The remanence, stress, and elevation data under the same coordinate were concatenated in a binary stream format of "coordinate + field 1 value + field 2 value + field 3 value", temporarily stored in a high-speed cache, and then written to an SSD solid-state drive.

[0027] An M×3 feature matrix is ​​constructed based on the total number of sampling points M and the field dimension N=3. The matrix elements are the standardized field signal values ​​of the corresponding coordinates. The mean of the field signal is calculated column by column for central preprocessing, and each element is subtracted from the mean to eliminate the dimension and baseline offset. The singular value decomposition adopts the double diagonalized SVD algorithm, which can be accelerated by Python numpy library or FPGA hardware to quickly output the eigenvectors and singular values.

[0028] The standard casting feature vector vc is obtained by extracting and calibrating the average feature vector after multi-field acquisition and decoupling of known clear Western Zhou bronze inscription samples. The singular value proportion threshold λth is set to 0.05. The cosine similarity elimination threshold is calibrated by casting field signal correlation experiment. Only field data strongly correlated with casting behavior are retained, and interference such as corrosion and substrate noise are accurately eliminated.

[0029] The gradient threshold was determined by statistically analyzing the gradient magnitude of clear inscription sample fields and taking the 95th percentile. The direction of the cast strokes was constrained to four main directions of bronze inscriptions: horizontal, vertical, and 45° / 135°. The topological skeleton extraction adopted an improved Zhang-Suen thinning algorithm. Gaussian filtering was used for smoothing and denoising single-pixel skeleton lines. The length threshold for isolated short line segments was set to 0.2mm to remove noisy and small line segments.

[0030] The endpoints, intersections, and inflection points of the skeleton line are determined by the connectivity of the pixel neighborhood. The endpoint connectivity is 1, the intersection connectivity is ≥3, and the inflection point is a pixel with a change in direction angle ≥30°. The topological node feature vector contains 8-dimensional quantized features, including spatial coordinates, field gradient magnitude, gradient direction, and the mean of the field distribution at 8 points in the neighborhood, thus completing the standardized construction of the node feature set.

[0031] The maximum connection distance threshold dth is set to 1.5mm, and the maximum direction angle threshold αth is set to 30°. The parameters are experimentally calibrated based on the conventional size and direction rules of the bronze inscription strokes. The node links are connected in a depth-first traversal order. When multiple nodes are connected in conflict, the node with the closer distance and smaller direction angle is selected first to complete the link construction, which fits the actual stroke shape of the inscription.

[0032] The inscription topological feature library contains more than 2,000 commonly used characters in bronze inscriptions from the Western Zhou to the Spring and Autumn periods. Each character is collected in multiple fields and topologically reconstructed to generate standard topological features. Scale normalization scales the minimum bounding box to 32×32 pixels. Rotation normalization corrects the stroke direction through principal component analysis. Topological isomorphism comparison is weighted by the number of nodes, connection relationship, stroke direction, and intersection structure, with weights of 0.2, 0.3, 0.3, and 0.2 respectively. The matching threshold is set to 0.85.

[0033] Characters that do not meet the matching threshold are automatically stored in a temporary feature set. After manual verification by archaeologists, the correct character features are incorporated into the topological feature library for iterative updates, improving the system's ability to recognize and adapt to rare and damaged characters.

[0034] The standardized data set is stored in XML format and includes character text, topological vectors, bounding box coordinates, and matching degree core fields. It exports XML / CSV dual-format files to adapt to general archaeological information systems. The inscription database uses a MySQL relational database and is stored in separate tables according to bronze artifact number, inscription location, character information, and topological data, supporting fast retrieval and export.

[0035] The dynamic sampling density base value is set at 50 points / mm², the density adaptation coefficient k is set at 2.0, the baseline gradient threshold G0 is set at 0.1V / mm, and the maximum / minimum sampling density ρmax = 200 points / mm² and ρmin = 10 points / mm². All parameters were calibrated through multiple archaeological sample experiments.

[0036] See the system example in the above embodiments for an application example of the system: An archaeological institute was conducting an inscription decipherment study on a bronze ding vessel from the mid-Western Zhou Dynasty in its collection. Due to long-term underground burial, the inscription area on the inner wall of the ding exhibited layered corrosion and some strokes were worn and broken. Traditional rubbing and optical identification methods could not fully extract the inscription information, and manual identification revealed several unidentifiable damaged characters. Therefore, this system was adopted for the identification work. Before implementation, the system completed the calibration of basic parameters, preset core parameters such as sampling density, sampling distance range, gradient threshold, and matching threshold, and imported the pre-built Western Zhou bronze inscription topological feature library. Each standard character in the library corresponds to a unique text attribution information and standardized topological link features.

[0037] During implementation, the system first constructs an orthogonal three-axis scanning coordinate system with the geometric center point of the inscription casting area as the unique fixed spatial origin. The X and Y axes form a scanning plane parallel to the inscription casting reference plane, and the Z axis is the normal axis perpendicular to the reference plane. The scanning execution end is equipped with three types of sensing units: remanence, stress, and micro-topography elevation. The scanning is completed through a three-axis synchronous linkage mechanism. The X and Y axes drive the sensing units to complete the full planar traversal of the inscription area, and the Z axis performs real-time directional position compensation according to the surface topography of the vessel, ensuring that the normal distance between the sensing unit and the surface of the vessel remains stable within the preset sampling range. During the scanning process, the system automatically adjusts the sampling point density according to the degree of abrupt change in the field signal of the pre-scan, increasing the sampling density in the edge area of ​​the casting marks and decreasing the sampling density in the flat area without casting marks. Finally, using the spatial coordinates of the sampling points as a unified index, the multi-physics field sampling values ​​under the same coordinate are dimensionally stitched together to obtain the original multi-dimensional field source data of the inscription area.

[0038] Subsequently, the system aligns the original data of the multidimensional field source according to spatial coordinates and constructs a field dimension feature matrix. First, the matrix is ​​preprocessed by centering to eliminate the dimensional differences and baseline offsets of different field dimensions. Then, the preprocessed matrix is ​​decomposed by singular value to obtain the eigenvectors and singular values ​​corresponding to each field dimension. Combined with the pre-calibrated standard features of casting behavior, the correlation between each field dimension and casting behavior is calculated. Finally, interference field signals unrelated to inscription casting are removed, and the strongly correlated effective field signal data are retained.

[0039] Next, the system performs spatial gradient calculation on the decoupled effective field signal to obtain the field gradient magnitude and gradient direction of each sampling point. Sampling points whose gradient magnitude exceeds a preset threshold and whose gradient direction conforms to the casting stroke direction constraint are marked as stroke edge sampling points. A closed connected region of the casting stroke is constructed with the edge sampling points as the boundary. The Euclidean distance from each sampling point in the region to the nearest edge sampling point is calculated as the field distance feature value. Redundant sampling points are eliminated in order of feature value from high to low. After each elimination, the topological connectivity of the region is checked. If the connectivity remains unchanged, the elimination is completed. If it changes, the sampling point is retained. After the loop operation, a connected structure with a single pixel width is obtained. The structure is smoothed and denoised to eliminate isolated short line segments that are too short and to correct the jagged edges of the lines. Finally, a single pixel width skeleton line of the inscription stroke is obtained. The endpoints, intersections, and inflection points of the skeleton line are uniformly marked as stroke topological nodes. At the same time, the spatial coordinates, field gradient features, and neighborhood field distribution features of each node are extracted to establish an independent feature vector for each node. The feature vectors of all nodes together constitute the node feature set.

[0040] Subsequently, the system connects the stroke topology nodes in the node feature set based on the consistency of stroke direction and spatial distance constraints. The connection is completed when the spatial Euclidean distance between two nodes to be connected and the angle between the gradient directions of the corresponding skeleton lines meet the preset constraints. Continuous stroke topology links are formed in the connection order, and finally all connected links together constitute a complete inscription stroke topology link.

[0041] After reconstructing the link, the system first normalizes the scale and rotation of the complete inscription stroke topology link to eliminate feature deviations caused by spatial scale and deflection angle, obtaining the standard topology link features to be identified. Then, it decomposes the standard character topology features in the feature library into basic topology units containing topology nodes, node connection relationships, stroke direction, stroke intersection and closed structure according to unified rules. First, it checks whether the number of topology nodes of the two types of features is within the preset error tolerance range. If it exceeds the range, the matching degree is recorded as 0. If it meets the requirements, a one-to-one mapping relationship is established for the nodes. The matching degree of the node that cannot be fully mapped is recorded as 0. After the mapping is completed, the matching scores of the four core dimensions are calculated respectively. The isomorphic matching degree in the range of 0 to 1 is obtained by fusing according to the preset weight. If the matching degree exceeds the preset threshold, the corresponding standard character in the library is marked as the character belonging to the link. If it does not exceed the threshold, it is marked as the character to be verified. At the same time, its link features are extracted and stored in the feature set to be verified.

[0042] Finally, the system uses the spatial coordinate system of the inscription casting area as a reference to bind the character recognition results, corresponding stroke topology link information, and character spatial position information one by one, generating a standardized data group with the character as the smallest unit. The data group contains character text information, topology link vector data, spatial bounding box coordinates, and recognition matching degree information. Then, all data groups are sorted and encoded according to the inherent arrangement order of the inscription to generate a structured dataset, which is stored in the preset inscription database. At the same time, a standardized export file that is compatible with the archaeological information system of the institute is generated.

[0043] In this application, the system successfully identified 28 characters of inscription on the inner wall of the bronze ding. Among them, 6 damaged strokes that could not be identified by traditional methods were completely restored and the characters were identified. The identification results were verified by archaeologists, and the accuracy of character recognition met the standards of archaeological research. The output structured data can be directly connected to the institute's cultural relics information management system, providing support for the dating, inscription interpretation and digital archiving of the bronze ding.

[0044] Example 2: A method for identifying inscriptions on bronze artifacts includes: A three-axis coordinate system is established with the geometric center of the inscription casting area as the origin. Multi-dimensional intrinsic physical field data are collected synchronously, and spatial alignment and splicing are used to generate multi-dimensional field source raw data, dynamically adapting the sampling density. A feature matrix is ​​constructed from the multidimensional field source data and singular value decomposition is performed. The field signal is decoupled according to the casting behavior characteristics, irrelevant interference data is removed, and the effective field signal is retained. Gradient calculation is performed on the effective field signal, stroke edges are marked and single-pixel skeleton lines are extracted, stroke topology nodes are calibrated, and node feature sets are constructed. Based on the constraints of stroke direction and spatial distance, the node connection conditions are verified, the node link connection is completed according to the inherent logic of casting, and a complete inscription topology link is generated. The topology links are scaled and rotated for normalization, and are compared with the standard inscription topology feature library for isomorphism. The character attribution is determined according to the matching degree, and characters that do not meet the standard are marked as characters to be verified. The binding of identification results, topology links, and spatial information generates standardized data, which is sorted and encoded according to the inscription arrangement to achieve structured storage and standardized file export.

[0045] In summary, the system and method described above, through scanning and acquiring the intrinsic physical field of the entire bronze inscription casting area and decoupling the field dimension, accurately eliminate field signal data unrelated to the inscription casting, obtain the effective field signal corresponding to the inscription casting, achieve spatial alignment of multi-field source data with unified spatial coordinates, balance acquisition accuracy and data processing efficiency through dynamically adapted sampling point density, avoid interference from bronze surface corrosion and wear on inscription recognition, locate the topological nodes of inscription strokes and restore the complete topological link of inscription strokes, complete the identification of inscription element attribution based on standardized topological feature comparison, improve the stability and accuracy of identifying damaged and blurred inscriptions, and finally output standardized and structured inscription data, providing further technical support for the archaeological research of bronze inscriptions.

[0046] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A system for identifying inscriptions on bronze artifacts, characterized in that, include: The field acquisition module is used to scan and acquire the intrinsic physical field of the entire area of ​​bronze inscription casting area, and obtain the original data of multi-dimensional field source corresponding to the area. The field decoupling module is used to decouple and separate the original data of multi-dimensional field sources in the field dimension, and remove field signal data that are unrelated to the inscription casting behavior; The extraction module is used to locate and extract the stroke topology nodes formed by the inscription casting in the decoupled field signal data, and establish an independent node feature set. The reconstruction module is used to connect the stroke topology nodes in the node feature set according to the inherent arrangement logic of the inscription casting, and generate a complete inscription stroke topology link. The identification module is used to compare the complete stroke topology link of the inscription with the preset inscription topology feature library link by link to identify the character attribution of the corresponding inscription. The output module is used to integrate the identification results of the inscription characters, the corresponding topological link information and spatial location information, output standardized data and complete structured storage.

2. The inscription identification system on bronze artifacts according to claim 1, characterized in that, The field acquisition module synchronously scans and acquires the global intrinsic physical field of the bronze inscription casting area through a three-axis linkage scanning path with the same spatial origin. The global intrinsic physical field includes the residual magnetic field, stress field and micro-morphological elevation field of the casting area. The field acquisition module uses the spatial coordinates of the scanning sampling points as a unified index to stitch together the multi-physical field sampling values ​​under the same spatial coordinates to generate multi-dimensional field source raw data. At the same time, it dynamically adapts the density of the scanning sampling points, increasing the sampling point density in the edge area of ​​the casting marks and decreasing the sampling point density in the flat area without casting marks.

3. The inscription identification system on bronze artifacts according to claim 2, characterized in that, The dynamic adaptation of the scan sampling point density follows the following: ； In the formula: The density of scan sampling points at planar coordinates (x, y) within the inscription casting area; The preset base sampling point density; The gradient magnitude of the pre-scan field signal at the plane coordinates (x, y); is the preset baseline gradient threshold; k is the preset density adaptation coefficient; The preset maximum sampling point density; This is the preset minimum sampling point density.

4. The inscription identification system on bronze ware according to claim 1, characterized in that, The field decoupling module first aligns the original multi-dimensional field source data according to spatial coordinates to construct a field dimension feature matrix. It then performs singular value decomposition on the feature matrix to obtain the eigenvectors and singular values ​​corresponding to each field dimension. Finally, based on the field mutation characteristics corresponding to the inscription casting behavior, it decouples and separates the field signals of each dimension, eliminating field signal data unrelated to the inscription casting behavior. ； In the formula: The unit eigenvector corresponding to the j-th field dimension Standard casting feature unit vector corresponding to pre-calibrated inscription casting behavior Cosine similarity; express and The angle between the vectors; Let j be the singular values ​​corresponding to the j field dimensions; It is the sum of the singular values ​​corresponding to all field dimensions; This is a preset threshold for the proportion of singular values.

5. The inscription identification system on bronze artifacts according to claim 1, characterized in that, The extraction module first performs spatial gradient calculation on the decoupled field signal data to obtain the field gradient magnitude and gradient direction of each sampling point. The sampling points whose field gradient magnitude exceeds the preset gradient threshold and whose gradient direction conforms to the casting stroke direction constraint are marked as stroke edge sampling points. Then, the topological skeleton is extracted from all stroke edge sampling points to obtain the single-pixel width skeleton line of the inscription stroke. The endpoints, intersections, and inflection points of the skeleton line are uniformly marked as stroke topological nodes. The extraction module extracts the spatial coordinates, field gradient features, and neighborhood field distribution features corresponding to each stroke topology node, and establishes an independent feature vector for each topology node. The feature vectors of all topology nodes together constitute the node feature set.

6. The inscription identification system on bronze artifacts according to claim 1, characterized in that, The reconstruction module connects the stroke topology nodes in the node feature set based on the consistency of stroke direction and spatial distance constraints to generate a complete inscription stroke topology link. The necessary and sufficient condition for two stroke-based topology nodes to be linked is set as follows: ； In the formula: for , Euclidean distance in a spatial coordinate system; , For any two stroke topology nodes to be connected in the node feature set; The preset maximum connection distance threshold; for , The angle between the gradient directions of the corresponding skeleton lines; The preset maximum directional angle threshold; After two nodes that meet the necessary and sufficient conditions are connected, they form a continuous stroke topology link in the order of connection. Finally, all connected stroke topology links together constitute a complete inscription stroke topology link.

7. The inscription identification system on bronze artifacts according to claim 1, characterized in that, During the operation phase of the recognition module, the scale and rotation of the generated complete inscription stroke topology link are first normalized to eliminate feature deviations caused by spatial scale and deflection angle, so as to obtain the standard topology link feature to be recognized. Then, the standard topology link feature is compared with the standard character topology feature in the preset inscription topology feature library on a link-by-link basis to calculate the isomorphism matching degree between the feature to be recognized and the standard feature in the library. When the isomorphic matching degree exceeds the preset matching threshold, the character information of the corresponding standard character in the library will be marked as the character belonging to the stroke topology link of the inscription. If the isomorphic matching degree does not exceed the preset matching threshold, the link is marked as a character to be verified, and its link features are extracted and stored in the feature set to be verified.

8. A bronze inscription identification system according to claim 1, characterized in that, The output module uses the spatial coordinate system of the inscription casting area as a reference to bind the identified inscription character attribution results, corresponding stroke topology link information, and character spatial position information one by one, generating a standardized data group with the character as the smallest unit. The standardized data set includes text information of characters, topology link vector data, spatial bounding box coordinates, and recognition matching degree information; The output module sorts and encodes all standardized data groups according to the inherent arrangement order of bronze inscriptions, generates a structured dataset, and finally stores the structured dataset into a preset inscription database. At the same time, it generates a standardized export file that is compatible with general archaeological information systems.

9. A bronze inscription identification system according to claim 1, characterized in that, The field acquisition module is interactively connected to the field decoupling module via a wireless network signal. The field decoupling module is interactively connected to the extraction module via a wireless network signal. The extraction module is interactively connected to the reconstruction module via a wireless network signal. The reconstruction module is interactively connected to the recognition module via a wireless network signal. The recognition module is interactively connected to the output module via a wireless network signal.

10. A method for identifying inscriptions on bronze artifacts, wherein the method is an implementation method of the inscription identification system on bronze artifacts as described in any one of claims 1-9, characterized in that, include: A three-axis coordinate system is established with the geometric center of the inscription casting area as the origin. Multi-dimensional intrinsic physical field data are collected synchronously, and spatial alignment and splicing are used to generate multi-dimensional field source raw data, dynamically adapting the sampling density. A feature matrix is ​​constructed from the multidimensional field source data and singular value decomposition is performed. The field signal is decoupled according to the casting behavior characteristics, irrelevant interference data is removed, and the effective field signal is retained. Gradient calculation is performed on the effective field signal, stroke edges are marked and single-pixel skeleton lines are extracted, stroke topology nodes are calibrated, and node feature sets are constructed. Based on the constraints of stroke direction and spatial distance, the node connection conditions are verified, the node link connection is completed according to the inherent logic of casting, and a complete inscription topology link is generated. The topology links are scaled and rotated for normalization, and are compared with the standard inscription topology feature library for isomorphism. The character attribution is determined according to the matching degree, and characters that do not meet the standard are marked as characters to be verified. The binding of identification results, topology links, and spatial information generates standardized data, which is sorted and encoded according to the inscription arrangement to achieve structured storage and standardized file export.

Images