CN-121999479-A - Ceramic cultural relic full-dimension identification system and method based on multi-mode visual large model and microscopic feature fusion

CN121999479ACN 121999479 ACN121999479 ACN 121999479ACN-121999479-A

Abstract

The invention relates to the technical field of cultural relic identification and vision large model fusion, in particular to a ceramic cultural relic full-dimensional identification system and method based on multi-mode vision large model and microscopic feature fusion, wherein the system and method quantizes data value through full-dimensional adaptation indexes of fusion data quality, feature integrity and mode cooperativity, and after screening high-quality applicable data, the rest data are calibrated by taking an optimal mode as a reference, so that cross-mode deviation is eliminated, and a data foundation is built; the method comprises the steps of converting core features into topological nodes, calculating association strength, establishing feature association edges with guide, generating traceable and interpretable identification guide links, fitting ceramic identification strong logic requirements, screening optimal links, combining a pre-training model, relying on a confidence layering judgment mechanism, balancing identification precision and cautiousness, effectively solving black box pain points of a traditional model, and improving reliability of identification results.

Inventors

SHI JUNCHAO
DING KE
AN NA
MI LIN
Li Xiaozhun
GUO TONGTONG

Assignees

北京潘家园数智科技文化有限公司
京古云(北京)信息科技有限公司

Dates

Publication Date: 20260508
Application Date: 20260122

Claims (10)

1. The ceramic cultural relic full-dimensional identification system based on the fusion of the multi-mode visual large model and the microscopic features is characterized by comprising an identification applicable data selection unit, an identification guide link generation unit and an optimal identification link execution unit; The identification applicable data selecting unit is used for acquiring all mode visual data of the ceramic cultural relics, acquiring full-dimensional adaptation indexes of all mode visual data, marking corresponding identification applicable data based on comparison results of the full-dimensional adaptation indexes and full-dimensional adaptation thresholds, marking the identification applicable data with the maximum full-dimensional adaptation index value as reference mode data, and calibrating the rest identification applicable data based on the reference mode data; The identification guide link generation unit takes core features in all identification applicable data as nodes of a topological model, calculates association strength of any two nodes, establishes feature association edges between the two nodes when the association strength is higher than an association strength threshold value, establishes guidance of the feature association edges, and generates a plurality of identification guide links according to the feature association edges and guidance thereof; the optimal identification link execution unit is used for acquiring the identification link comprehensive index of each identification guide link, marking the identification link comprehensive index with the largest numerical value as an optimal link, combining a pre-trained identification model based on the node characteristics of the optimal link, outputting an identification result and confidence level, and judging the result.
2. The ceramic cultural relic full-dimension identification system based on multi-modal visual large model and microscopic feature fusion as claimed in claim 1, wherein the full-dimension adaptation index of modal visual data is obtained by obtaining the data quality coefficient of one modal visual data Coefficient of feature integrity Modal co-factor By the formula Calculating to obtain the full-dimensional adaptation index of the modal visual data ; 、、 Are weight coefficients.
3. The ceramic relic full-dimension identification system based on multi-modal visual large model and microscopic feature fusion as recited in claim 2, wherein the data quality coefficient of modal visual data 。
4. The ceramic cultural relic full-dimensional identification system based on multi-modal visual large model and microscopic feature fusion as defined in claim 2, wherein the feature integrity coefficient Selecting one mode visual data, defining all core features contained in the mode visual data, obtaining feature importance weights of all core features, obtaining feature coverage of all core features, carrying out weighted summation calculation on the feature importance weights and the feature coverage of all core features, and calculating to obtain feature integrity coefficients 。
5. The ceramic cultural relic full-dimension identification system based on multi-modal visual large model and microscopic feature fusion as defined in claim 2, wherein the modal coordination coefficient The acquisition method comprises the steps of acquiring a modal feature vector of each modal visual data, selecting one modal visual data, acquiring feature similarity of the selected one modal visual data and each other modal visual data, and calculating by adopting cosine similarity to obtain a plurality of images For all of Summing, dividing by (total number of modal visual data n-1), and averaging to obtain modal synergistic coefficient 。
6. The ceramic cultural relic full-dimension identification system based on multi-modal visual large model and microscopic feature fusion as set forth in claim 1, wherein the identification link comprehensive index of the identification guide link is obtained by selecting an identification guide link and obtaining the correlation strength cumulative value of the identification guide link Mean value of importance Modal visual coverage By the formula Calculating to obtain the comprehensive index of the identification link of the identification guide link ; 、、 Are weight coefficients.
7. The ceramic cultural relic full-dimensional identification system based on multi-modal visual large model and microscopic feature fusion according to claim 6, wherein the correlation strength integrated value of the identification guide link The acquisition mode comprises the steps of acquiring the association strength of each characteristic association edge in the identification guide link, performing product calculation on the association strength of all characteristic association edges, and obtaining an association strength accumulated value by calculation 。
8. The ceramic cultural relic full-dimension identification system based on multi-modal visual large model and microscopic feature fusion as defined in claim 6, wherein the importance mean of the identification guide link The acquisition mode comprises the steps of acquiring the characteristic importance weight of each node in the identification guide link, carrying out summation mean value calculation on the characteristic importance weight of each node, and obtaining an importance mean value through calculation 。
9. The ceramic relic full-dimension identification system based on multi-modal visual large model and microscopic feature fusion as recited in claim 6, wherein the modal visual coverage of the guide link is identified The acquisition mode of (1) acquiring the total number of the mode visual data covered in the coverage of the authentication guide link By the formula The calculation result shows that the method comprises the steps of, Is the total number of modalities of the modal visual data.
10. The ceramic cultural relic full-dimension identification method based on the multi-mode visual large model and the microscopic feature fusion is applied to the ceramic cultural relic full-dimension identification system based on the multi-mode visual large model and the microscopic feature fusion, and is characterized by comprising the following steps: Selecting identification applicable data of ceramic cultural relics; step two, taking the core features in the authentication application data as nodes of a topology model, establishing feature association edges and guiding, and generating a plurality of authentication guiding links according to the feature association edges and guiding thereof; Step three, acquiring an identification link comprehensive index of each identification guide link, marking the identification link comprehensive index with the largest numerical value as an optimal link, combining a pre-trained identification model based on node characteristics of the optimal link, outputting an identification result and confidence level, and judging the result.

Description

Ceramic cultural relic full-dimension identification system and method based on multi-mode visual large model and microscopic feature fusion Technical Field The invention relates to the technical field of cultural relic identification and visual large model fusion, in particular to a ceramic cultural relic full-dimension identification system and method based on multi-mode visual large model and microscopic feature fusion. Background In the field of ceramic cultural relics identification, application of multi-mode visual data has become an important development direction, but the prior art has a remarkable short plate in a data processing link. The traditional method is often used for directly collecting and using macroscopic, microscopic, spectral and other multi-mode data, lacks a comprehensive evaluation mechanism for data adaptation value, and cannot effectively screen low-quality, feature missing or poor-cooperativity data, so that invalid information interferes with identification results. Meanwhile, the natural deviation exists in the cross-modal data due to the differences of the acquisition equipment and the characteristic scale, the existing calibration mode is mostly limited to simple numerical normalization, and the physical attribute of the unbound ceramic cultural relics is precisely aligned with the process logic, so that effective coordination of different modal characteristics is difficult to form, and the improvement of the identification precision is restricted. The current ceramic cultural relic identification technology has obvious defects in aspects of feature fusion and logic modeling. The prior proposal often carries out simple splicing or isolated use on the core characteristics of each mode, fails to deeply dig the internal association among the characteristics, and does not consider the causal relationship and the identification priority of the ceramic process. This lack of structured modeling results in fuzzy feature-related logic, and the authentication process is difficult to trace back and interpret. The core requirement of cultural relic authentication on strong logic support is not met, so that the authentication conclusion lacks enough persuasion, and the authentication challenge brought by the increasing refinement of the imitation technology cannot be effectively met. The traditional authentication model and the result judging mechanism have limitations, and the authentication precision and the cautiousness are difficult to balance. Some techniques rely on a single model to process complex features, either because of the strong interpretability of the model but insufficient capture of high-dimensional features, or because of the lack of logical support of results caused by the use of black box deep learning models. Meanwhile, most of the existing result judgment is single conclusion output, a layering mechanism based on confidence coefficient is lacking, the reliability degree of the result cannot be clearly distinguished, supplementary verification guidance is not provided for the suspicious situation, misjudgment risks are easy to occur, and high requirements of the cultural relic identification field on the rigor and the safety are difficult to adapt. Disclosure of Invention Aiming at the defects existing in the prior art, the invention aims to provide a ceramic cultural relic full-dimensional identification system and method based on multi-mode visual large model and microscopic feature fusion. In order to achieve the above purpose, the present invention provides the following technical solutions: The ceramic cultural relic full-dimensional identification system based on the fusion of the multi-mode visual large model and the microscopic features comprises an identification applicable data selection unit, an identification guide link generation unit and an optimal identification link execution unit; The identification applicable data selecting unit is used for acquiring all mode visual data of the ceramic cultural relics, acquiring full-dimensional adaptation indexes of all mode visual data, marking corresponding identification applicable data based on comparison results of the full-dimensional adaptation indexes and full-dimensional adaptation thresholds, marking the identification applicable data with the maximum full-dimensional adaptation index value as reference mode data, and calibrating the rest identification applicable data based on the reference mode data; The identification guide link generation unit takes core features in all identification applicable data as nodes of a topological model, calculates association strength of any two nodes, establishes feature association edges between the two nodes when the association strength is higher than an association strength threshold value, establishes guidance of the feature association edges, and generates a plurality of identification guide links according to the feature association edges and guidance thereof; th