A multimodal object similarity comparison method, system, device, and medium

Abstract

The application discloses a kind of multimodal object similarity comparison method, system, equipment and medium.Method includes: the reference object and the comparison object of the same modality are vectorized to obtain reference vector group and comparison vector group;Determine the similarity between corresponding vectors in two vector groups after row number alignment, record the reference vector serial number of which similarity is lower than first threshold;If the first ratio of serial number quantity and effective reference vector quantity reaches second threshold, it is determined that it meets consistency requirement;Otherwise, effective comparison vector is sequentially rearranged, similarity calculation and serial number recording steps are repeated, determine the second ratio based on the reference vector serial number after deduplication, if the second ratio reaches second threshold, it is determined that it meets consistency requirement, if the second ratio still does not reach second threshold after all effective comparison vectors complete a round of sequential rearrangement, it is determined that it does not meet consistency requirement.The scheme can improve the accuracy and reliability of similarity comparison of two objects under the same modality.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, system, device, and medium for comparing the similarity of multimodal objects. Background Technology

[0002] Existing multimodal object similarity comparison methods typically perform only a single vector similarity calculation after vectorization when comparing multimodal objects. They do not consider issues such as inconsistent row numbers in the vector groups of two objects or possible vector order shifts, which leads to some effective features not being matched or being misjudged as inconsistent, thus affecting the accuracy and reliability of the final similarity determination. Summary of the Invention

[0003] This invention provides a multimodal object similarity comparison method, system, device, and medium to solve the problem in the prior art that the final judgment result is inaccurate due to the failure to consider the inconsistency of the number of rows in the vector group and the vector order offset.

[0004] According to one aspect of the present invention, a multimodal object similarity comparison method is provided, the method comprising: Vectorize the reference object and the comparison object of the same mode to obtain the reference vector group and the comparison vector group; Align the reference vector group and the comparison vector group by row number; Determine the similarity between corresponding vectors in the row-aligned reference vector group and comparison vector group, and record the index of the reference vector with a similarity lower than the first threshold; If the first ratio of the number of serial numbers to the number of valid reference vectors in the reference vector group reaches the second threshold, then the two objects are determined to meet the consistency requirement. Otherwise, the valid comparison vectors in the comparison vector group are rearranged in order, and the above similarity calculation and sequence number recording steps are repeated based on the rearranged comparison vector group. The second ratio is determined based on the sequence number of the deduplicated reference vector. If the second ratio reaches the second threshold, the two objects are determined to meet the consistency requirement. If the second ratio still does not reach the second threshold after all valid comparison vectors have been rearranged in order, the two objects are determined not to meet the consistency requirement. Among them, the effective reference vector refers to the original reference vector in the reference vector group before row alignment, and the effective comparison vector refers to the original comparison vector in the comparison vector group before row alignment.

[0005] According to another aspect of the present invention, a multimodal object similarity comparison system is provided, the system comprising: The vectorization processing module is used to vectorize the reference object and the comparison object of the same mode to obtain the reference vector group and the comparison vector group; The row alignment module is used to align the row numbers of the reference vector group and the comparison vector group. The similarity calculation and recording module is used to determine the similarity between corresponding vectors in the reference vector group and the comparison vector group after row alignment, and to record the index of the reference vector with a similarity lower than the first threshold. The initial determination module is used to determine that two objects meet the consistency requirement if the first ratio of the number of serial numbers to the number of valid reference vectors in the reference vector group reaches the second threshold. The cyclic comparison module is used to rearrange the order of the valid comparison vectors in the comparison vector group when the first ratio does not reach the second threshold, and repeat the above similarity calculation and sequence number recording steps based on the rearranged comparison vector group. The second ratio is determined based on the sequence number of the deduplicated reference vector. If the second ratio reaches the second threshold, it is determined that the two objects meet the consistency requirement. If the second ratio still does not reach the second threshold after all valid comparison vectors have completed one round of sequential rearrangement, it is determined that the two objects do not meet the consistency requirement. Among them, the effective reference vector refers to the original reference vector in the reference vector group before row alignment, and the effective comparison vector refers to the original comparison vector in the comparison vector group before row alignment.

[0006] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the multimodal object similarity comparison method according to any embodiment of the present invention.

[0007] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the multimodal object similarity comparison method according to any embodiment of the present invention.

[0008] According to another aspect of the present invention, a computer program product is provided, the computer program product comprising a computer program that, when executed by a processor, implements the multimodal object similarity comparison method according to any embodiment of the present invention.

[0009] The technical solution of this invention involves vectorizing reference objects and comparison objects of the same modality to obtain a reference vector group and a comparison vector group; aligning the rows of the reference vector group and the comparison vector group; determining the similarity between corresponding vectors in the aligned reference vector group and the comparison vector group, and recording the index of the reference vector with a similarity lower than a first threshold; if the ratio of the number of indexes to the number of valid reference vectors in the reference vector group reaches a second threshold, then the two objects are determined to meet the consistency requirement; otherwise, the valid comparison vectors in the comparison vector group are rearranged in order, and the above similarity calculation and index recording steps are repeated based on the rearranged comparison vector group. A second ratio is determined based on the deduplicated reference vector indexes. If the second ratio reaches the second threshold, then the two objects are determined to meet the consistency requirement; if the second ratio still does not reach the second threshold after all valid comparison vectors have undergone one round of rearrangement, then the two objects are determined not to meet the consistency requirement. This scheme eliminates calculation errors or feature omissions caused by inconsistent vector numbers by aligning the row numbers of the reference vector group and the comparison vector group; and effectively overcomes the misjudgment problem caused by vector order offset through multiple rounds of rearrangement comparison, thereby effectively improving the accuracy and reliability of similarity comparison between two objects in the same modality.

[0010] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart of a multimodal object similarity comparison method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a multimodal object similarity comparison system provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of an electronic device that implements the multimodal object similarity comparison method of the present invention. Detailed Implementation

[0013] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0014] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0015] Figure 1 This is a flowchart illustrating a multimodal object similarity comparison method provided in an embodiment of the present invention. This embodiment is applicable to situations where a similarity comparison is performed between a reference object and a comparison object of the same modality to determine their degree of consistency. This method can be executed by a multimodal object similarity comparison system, which can be implemented in hardware and / or software and can be configured in an electronic device. Figure 1 As shown in the figure, the multimodal object similarity comparison method provided in this embodiment specifically includes the following steps: S110. Vectorize the reference object and the comparison object of the same mode to obtain the reference vector group and the comparison vector group.

[0016] In this context, the reference object can refer to the modal data used as a benchmark for comparison, while the comparison object refers to another modal data that needs to be compared with the reference object in terms of similarity. Both belong to the same modal type, such as both being text data or audio data. Vectorization processing refers to the process of converting non-numerical modal data into numerical vectors using a specific algorithm. Each vector represents a local or global feature in the original data, and each component of the vector represents the value of a certain dimension.

[0017] A reference vector set can be a collection of all reference vectors obtained after vectorizing a reference object, arranged in order. This set can be viewed as a matrix, where each row is a reference vector, and the number of rows equals the number of reference vectors. A comparison vector set can be a collection of all comparison vectors obtained after vectorizing a comparison object, arranged in order. Its structure is similar to that of the reference vector set, and the dimension of each comparison vector is exactly the same as that of the reference vector.

[0018] In this embodiment of the invention, a reference object and a comparison object to be processed can be obtained, and it can be verified whether their modal types are completely consistent. If the modal types are inconsistent, the comparison process is terminated or a modal mismatch message is output. If they are consistent, the corresponding vectorization algorithm is selected according to the modal type, and the reference object and the comparison object are vectorized respectively to obtain the corresponding reference vector group and comparison vector group. The reference vector in the reference vector group and the comparison vector in the comparison vector group have the same dimension.

[0019] S120. Align the rows of the reference vector group and the comparison vector group.

[0020] In this embodiment of the invention, to address the issue of inconsistent row counts (vector counts) in the vector groups obtained after vectorization of the reference and comparison objects, and to ensure that the two vector groups can be sequentially matched for subsequent similarity calculations, row count alignment processing can be performed on the aforementioned generated reference and comparison vector groups. Specifically, the vector group with the smaller row count can be expanded to equal the row count of the vector group with the larger row count; the specific expansion methods may include, but are not limited to, zero-vector padding, repeated vector padding, etc., and this embodiment does not impose any restrictions on these methods. It should be understood that the padded vectors are invalid vectors and do not participate in the subsequent valid vector statistics.

[0021] S130. Determine the similarity between corresponding vectors in the row-aligned reference vector group and comparison vector group, and record the index of the reference vector with a similarity lower than the first threshold.

[0022] In this embodiment of the invention, after determining that the reference vector group and the comparison vector group have completed row alignment, the similarity between the corresponding vectors can be calculated according to the vector index in both groups. For example, the similarity between reference vector 1 and comparison vector 1 can be calculated sequentially, the similarity between reference vector 2 and comparison vector 2 can be calculated sequentially, and so on. Then, the calculated similarity is compared with a pre-configured first threshold TH1. If the similarity is greater than TH1, it means that the corresponding vector pair is highly similar and belongs to invalid or non-critical information. If the similarity is less than or equal to TH1, it means that the corresponding vector pair has differences or is significantly different and belongs to valid or critical information.

[0023] For vector pairs with a similarity ≤ TH1, the index of the reference vector can be recorded, which is the original position index of the reference vector in the reference vector group. Furthermore, the similarity score and the current round number can also be recorded for subsequent deduplication statistics.

[0024] S140. If the first ratio of the number of serial numbers to the number of valid reference vectors in the reference vector group reaches the second threshold, then the two objects are determined to meet the consistency requirement.

[0025] The consistency requirement can refer to the similarity between the reference object and the comparison object in terms of content, features, or semantics reaching a preset standard. A valid reference vector can refer to the original reference vectors in the reference vector group before row alignment.

[0026] In this embodiment of the invention, the number N of reference vector indices that meet the similarity ≤ TH1 recorded above, and the number M of valid reference vectors in the reference vector group, that is, the original number of reference vectors before row alignment, can be counted respectively. Then, the ratio between the two numbers is calculated and denoted as the first ratio P = N / M. If the first ratio P reaches the pre-configured second threshold TH2, it is determined that the reference object and the comparison object meet the preset consistency requirements.

[0027] S150. Otherwise, rearrange the order of the valid comparison vectors in the comparison vector group, and repeat the above similarity calculation and sequence number recording steps based on the rearranged comparison vector group. Determine the second ratio based on the sequence number of the deduplicated reference vector. If the second ratio reaches the second threshold, it is determined that the two objects meet the consistency requirement. If the second ratio still does not reach the second threshold after all valid comparison vectors have completed one round of sequential rearrangement, it is determined that the two objects do not meet the consistency requirement.

[0028] Here, a valid comparison vector can refer to the original comparison vectors in the comparison vector group before row alignment. Order reordering refers to the operation of changing the order of valid comparison vectors in the comparison vector group. For example, a cyclic shift can be used to move the first few vectors to the end to generate a new vector order combination. A round of order reordering refers to the process of sequentially performing cyclic shifts on all valid comparison vectors in the comparison vector group, ensuring that each valid comparison vector is moved to the beginning (or end) of the vector group at least once. That is, it involves traversing all possible cyclic shift orders until the comparison vector group is restored to its original order.

[0029] In this embodiment of the invention, if the first ratio P does not reach the second threshold TH2, the following order rearrangement and multi-round comparison process is executed: ① The order of the valid comparison vectors in the comparison vector group is changed according to a preset rule to obtain a new comparison vector group after rearrangement. For example, a cyclic shift method can be used to move the first n valid comparison vectors to the end of the vector group, and the remaining vectors are moved forward n positions in sequence.

[0030] ② Using the rearranged new comparison vector group, perform the similarity calculation and filtering process in step S130 again with the original reference vector group, and record the reference vector number that satisfies the similarity ≤ TH1 in this round.

[0031] ③ Combine the reference vector sequence number obtained in this round with the sequence numbers obtained in all previous rounds (including the initial round and all previous rearrangement rounds), remove duplicate sequence numbers, and obtain the total number of unique sequence numbers after deduplication, N_total. Calculate the ratio between this total number N_total and the total number of valid reference vectors M, and denot it as the second ratio P_total=N_total / M.

[0032] ④ If the second ratio P_total reaches the preset second threshold TH2, it is determined that the reference object and the comparison object meet the consistency requirements, and the comparison process is terminated.

[0033] ⑤ If the second ratio P_total does not reach the preset second threshold TH2, check whether all valid comparison vectors have completed one round of sequential rearrangement (i.e., each possible cyclic displacement order has been covered). If one round has not been completed, return to step ① to continue the next sequential rearrangement of the comparison vector group. If one round has been completed and the new second ratio still does not reach the second threshold, it is determined that the reference object and the comparison object do not meet the consistency requirements, and the comparison process is terminated.

[0034] The multimodal object similarity comparison method provided by this invention eliminates calculation errors or feature omissions caused by inconsistent vector numbers by aligning the rows of the reference vector group and the comparison vector group. It achieves precise focusing on key features with low similarity by recording the index of reference vectors with similarity below a first threshold and calculating a first ratio for initial judgment. When the initial judgment fails, it effectively overcomes the misjudgment problem caused by vector order shift by rearranging the order of valid comparison vectors in the comparison vector group and repeating the similarity calculation, and then determining a second ratio based on the deduplicated index for cumulative judgment. Simultaneously, it uses a preset second threshold as the quantification judgment standard and terminates early when the cumulative threshold is reached, and determines inconsistencies when the threshold is not reached after all rearrangements, ensuring the objectivity, efficiency, and finite-step convergence of the judgment process, significantly improving the accuracy and reliability of similarity comparison between two objects in the same modality.

[0035] Furthermore, based on the above embodiments of the invention, the reference object and the comparison object of the same modality are vectorized to obtain a reference vector group and a comparison vector group, including: Select the appropriate vectorization method according to the modality type; the modality type can be one of text, image, audio, or video. The reference object and the comparison object are vectorized separately according to the vectorization method to obtain the corresponding reference vector group and comparison vector group; wherein, each reference vector in the reference vector group and each comparison vector in the comparison vector group have the same dimension.

[0036] In this embodiment of the invention, the process of vectorizing the reference object and the comparison object specifically includes: S1. Determine the modality type of the reference object and the comparison object. The modality type can be one of text, image, audio, or video. Based on the determined modality type, select the vectorization method corresponding to that modality. For example, text modality can use Term Frequency-Inverse Document Frequency (TF-IDF) or Bidirectional Encoder Representations from Transformers (BERT) sentence vectorization algorithm; image modality can use Convolutional Neural Network (CNN) feature extraction algorithm; audio modality can use Mel-frequency cepstral coefficients (MFCC) feature extraction algorithm; and video modality can extract features frame by frame and combine them into a vector sequence, etc.

[0037] S2. Segment the reference object, and then vectorize each segment according to the vectorization method selected in the previous step to obtain a set of intermediate vectors corresponding to each segment; then, aggregate the intermediate vectors to obtain the reference vectors of the corresponding segments, and finally form a reference vector group from the reference vectors corresponding to each segment.

[0038] Similarly, the comparison objects are also vectorized as described above to obtain a comparison vector group consisting of multiple comparison vectors. During vectorization, uniform feature extraction parameters (such as fixed window size and feature dimension) or subsequent dimension mapping or alignment operations can be used to ensure that each reference vector in the reference vector group has the exact same dimension as each comparison vector in the comparison vector group, meaning that all vectors have the same length (number of components). This guarantees the feasibility of subsequent vector-wise similarity calculations.

[0039] Furthermore, the reference vector group can be statically stored so that it can be directly read and used when performing similarity comparisons with multiple different comparison objects in the future, avoiding repeated vectorization processing on the same reference object, thereby improving comparison efficiency and saving computing resources.

[0040] Furthermore, based on the above embodiments of the invention, row alignment is performed on the reference vector group and the comparison vector group, including: Count the number of rows in the first row of the reference vector group and the number of rows in the second row of the comparison vector group respectively; If the number of rows in the first row is not equal to the number of rows in the second row, then for the vector group with the smaller number of rows, fill the end with zero vectors with the same dimension as the original vectors, until the number of rows in the two vector groups is equal.

[0041] In this embodiment of the invention, the row number alignment process may include: S1. Count the number of rows in the first row of the reference vector group (R1) and the number of rows in the second row of the comparison vector group (R2). The row count can be performed using conventional methods for counting rows in vector groups in linear algebra, such as directly counting the total number of vectors in the vector group.

[0042] S2. Compare R1 and R2. If R1 equals R2, no padding is needed, meaning the two vector groups are already aligned. If R1 does not equal R2, determine the vector group with the smaller number of rows.

[0043] S3. For vector groups with fewer rows, one or more zero vectors can be added to the end until the number of rows in the vector group equals the number of rows in the original vector group with more rows. After filling, the two vector groups will have the same number of rows, i.e., R = max(R1, R2). The dimension of each zero vector added is exactly the same as the dimension of the original vectors in the vector group.

[0044] Furthermore, based on the above embodiments of the invention, the similarity between corresponding vectors in the row-aligned reference vector group and the comparison vector group is determined, and the index of the reference vector with a similarity lower than a first threshold is recorded, including: After aligning the rows, calculate the similarity between corresponding vectors in the reference vector group and the comparison vector group according to the vector index to obtain the similarity sequence. Each similarity in the similarity sequence is compared with the first threshold, and similarities greater than the first threshold are filtered out. Record the index of the reference vector corresponding to the retained similarity and the current round number.

[0045] In this embodiment of the invention, the process of similarity calculation and screening in the initial round specifically includes: S1. After row alignment, the i-th reference vector in the reference vector group and the i-th comparison vector in the comparison vector group can be extracted sequentially according to vector index i (i=1,2,…,R), and the similarity between them can be calculated. The similarity can be any measurable index of the closeness between vectors, such as cosine similarity or normalized Mahalanobis distance; this embodiment does not impose any restrictions. After calculating the similarity corresponding to all vector indices, a sequence containing R similarity values ​​can be obtained, i.e., the similarity sequence.

[0046] S2. For each similarity in the similarity sequence, it can be compared with a pre-configured first threshold TH1. If the similarity is greater than the first threshold TH1, it is considered that the similarity between the vector pairs corresponding to the similarity is high and the contribution to the determination of the difference between the objects is small. Therefore, it is filtered out and no information is recorded. If the similarity is less than or equal to the first threshold TH1, the similarity is retained as valid data for subsequent determination of consistency.

[0047] S3. For each retained similarity, record the index of its corresponding reference vector, as well as the round number of the current comparison round. For example, the initial round is round 1, and the round numbers after subsequent cyclic shifts are sequentially increased. This recorded data will serve as the basis for subsequent multi-round cumulative judgment and deduplication processing.

[0048] Furthermore, based on the above embodiments of the invention, if the first ratio of the number of serial numbers to the number of valid reference vectors in the reference vector group reaches a second threshold, then it is determined that the two objects meet the consistency requirement, including: Count the total number of recorded serial numbers; The total number of valid reference vectors in the reference vector group is counted. Determine the first ratio between the total number of sequence numbers and the total number of valid reference vectors; If the first ratio reaches the second threshold, then the reference object and the comparison object are determined to meet the consistency requirements.

[0049] In this embodiment of the invention, the consistency determination process for the initial round specifically includes: S1. Count the number N of reference vector indices that satisfy similarity ≤ TH1 recorded in this round, and the number M of valid reference vectors in the reference vector group.

[0050] S2. Calculate the first ratio P=N / M between the number N and the number M. This ratio reflects the proportion of reference vectors that differ significantly from the comparison vector in the current round to all valid reference vectors.

[0051] S3. Compare the calculated first ratio P with the preset second threshold TH2. If P≥TH2, it is determined that the reference object and the comparison object meet the consistency requirements, that is, the two have high consistency, and the comparison process is terminated. If P<TH2, it is determined that the consistency of the current round is insufficient, and the subsequent order rearrangement and multi-round comparison process needs to be entered.

[0052] Furthermore, based on the above embodiments of the invention, the effective comparison vectors in the comparison vector group are rearranged in order, and the above similarity calculation and sequence number recording steps are repeated based on the rearranged comparison vector group, including: If the first ratio does not reach the second threshold, then the valid comparison vectors are cyclically shifted according to the preset displacement step size, so as to move the valid comparison vectors that are currently at the front and whose number is equal to the preset displacement step size to the end of the comparison vector group. Based on the reference vector group and the rearranged comparison vector group, repeat the similarity calculation and sequence number recording steps to obtain the sequence number of the reference vector with a similarity lower than the first threshold in this round.

[0053] The preset displacement step size can refer to the number of valid comparison vectors that move from the front to the end of the comparison vector group in one cyclic displacement operation. It is represented by a positive integer n and satisfies 1≤n<the total number of valid comparison vectors.

[0054] In this embodiment of the invention, the process of rearranging the comparison vector order specifically includes: S1. If the first ratio P < the second threshold TH2, then according to the pre-configured preset displacement step size n, the n valid comparison vectors currently at the front of the comparison vector group are sequentially extracted and moved to the end of the entire comparison vector group. If the comparison vector group contains padding invalid zero vectors, the extracted n valid vectors are moved after all the padding zero vectors, while the original (n+1)th valid comparison vector is automatically moved to the first position, and the remaining vectors are sequentially shifted forward n positions. After the above displacement, a new comparison vector group with rearranged order is obtained, where the total number of vectors remains the same, but the relative order of the valid comparison vectors is cyclically changed.

[0055] S2. Compare the rearranged comparison vector group with the original reference vector group one by one according to their corresponding indices. Use the same similarity calculation method as the initial round (e.g., cosine similarity) to calculate the similarity of each pair of vectors. Compare each calculated similarity with the first threshold TH1 again. If the similarity is greater than the first threshold TH1, filter it out. If the similarity is less than or equal to the first threshold TH1, record the index of the reference vector corresponding to the similarity and the current round number for subsequent deduplication statistics and consistency determination.

[0056] Furthermore, based on the above embodiments of the invention, a second ratio is determined based on the deduplicated reference vector index. If the second ratio reaches a second threshold, the two objects are determined to meet the consistency requirement. If the second ratio still does not reach the second threshold after all valid comparison vectors have undergone one round of order rearrangement, the two objects are determined not to meet the consistency requirement, including: Merge the sequence number obtained in this round with the sequence numbers obtained in all previous rounds to remove duplicates, and count the total number of sequence numbers after deduplication; Determine the second ratio between the total number of duplicated serial numbers and the total number of valid reference vectors; If the second ratio reaches the second threshold, it is determined that the reference object and the comparison object meet the consistency requirements, and the subsequent process is terminated. If the second ratio does not reach the second threshold, it is determined whether all valid comparison vectors have completed one round of cyclic displacement. If one round has not been completed, the process returns to the step of performing cyclic displacement on the valid comparison vectors according to the preset displacement step size. If one round has been completed and the second ratio still does not reach the second threshold, it is determined that the reference object and the comparison object do not meet the consistency requirements.

[0057] In this embodiment of the invention, the process of multi-round cyclic comparison and consistency determination specifically includes: S1. The reference vector sequence number obtained by similarity calculation and filtering in the current round can be merged with the reference vector sequence numbers recorded in all previous completed rounds. Then, the merged sequence number set is deduplicated to obtain the total number of deduplicated sequence numbers N_total.

[0058] S2. Determine the ratio of the total number of deduplicated serial numbers N_total to the total number of valid reference vectors M, and obtain the second ratio P_total = N_total / M.

[0059] S3. If the second ratio P_total reaches the preset second threshold TH2, it is determined that the reference object and the comparison object meet the consistency requirements, and the comparison process is terminated.

[0060] S4. If the second ratio P_total does not reach the second threshold TH2, further determine whether all valid comparison vectors have completed one round of cyclic displacement. For example, the process of determining whether one round of cyclic displacement has been completed may include: assuming the total number of valid comparison vectors is L, and each cyclic displacement moves n vectors, then the number of displacements required to complete one round of complete cyclic displacement is L / gcd(L,n), where gcd() is the greatest common divisor function. The determination method is: after each cyclic displacement, the displacement count is accumulated. When the accumulated count reaches L / gcd(L,n), it is determined that all valid comparison vectors have returned to their original order, that is, one round covering all relative displacements has been completed. Alternatively, the set of valid comparison vectors that have appeared at the beginning (or have been moved) can be recorded. When all valid comparison vectors have been moved to the end (or the beginning) at least once, it is considered that one round has been completed.

[0061] S5.1 If it is determined that a round of comparison has not been completed, return to the step of performing a cyclic displacement on the valid comparison vector according to the preset displacement step size, generate a new comparison vector group, and start the next round of comparison.

[0062] S5.2 If it is determined that one cycle has been completed and the second ratio P_total has not yet reached the second threshold TH2, then it is finally determined that the reference object and the comparison object do not meet the consistency requirements, and the comparison process ends.

[0063] This invention, through its multi-round cumulative deduplication and judgment mechanism, effectively overcomes the misjudgment problem caused by the shift in the order of the comparison vector groups. Even if there are few low-similarity matches in a single round of comparison, the accumulation and deduplication of matching results under different orders through multiple rounds of displacement can comprehensively capture the true similarity features between the reference object and the comparison object. Simultaneously, the "termination upon early attainment" strategy significantly reduces unnecessary computational overhead; and the final judgment condition of determining inconsistency if the target is not met after one round of displacement ensures the determinism and finite-step convergence of the judgment process. In summary, this scheme provides an efficient, accurate, and quantifiable judgment method for comparing the similarity of two objects in the same modality.

[0064] In a specific embodiment, taking the similarity comparison of two objects under the same text modality—reference text T1 and comparison text T2—as an example, the implementation process of this scheme is explained. The length of comparison text T2 is less than the length of reference text T1. The cyclic displacement step size n is 1 (i.e., only one valid comparison vector is shifted each time). The first threshold TH1 is 0.85, the second threshold TH2 is 0.6, and zero-vector padding is used. The specific process is as follows: S1. The BERT sentence vector algorithm is used to extract segmented features from the reference text T1, generating three reference vectors with a dimension of 768, denoted as Vt1, Vt2, and Vt3 respectively, forming a reference vector group V_ref=[Vt1 (sequence 1), Vt2 (sequence 2), Vt3 (sequence 3)], with 3 rows R1 and 3 valid sequence numbers M=3.

[0065] S2. Use the same BERT sentence vector algorithm as in step S1 to extract features from the comparison text T2. Since the text T2 is short, only two comparison vectors Vt4 and Vt5 with dimensions of 768 are generated to form a comparison vector group V_cmp=[Vt4,Vt5] with 2 rows R2.

[0066] S3. Since the two vector groups have different numbers of rows, the comparison vector group with the smaller number of rows is padded with a zero vector. That is, a 768-dimensional zero vector is added to the end of the comparison vector group, and the padded comparison vector group becomes V_cmp′=[Vt4,Vt5,zero vector], with 3 rows R2′. The zero vector is marked as an invalid vector and does not participate in the subsequent valid index counting.

[0067] S4. Calculate the cosine similarity according to the one-to-one correspondence of vector indices: Serial number 1: The similarity between Vt1 and Vt4 is 0.88 > 0.85 (first threshold TH1), so it is filtered out and not recorded; Sequence 2: The similarity between Vt2 and Vt5 is 0.72 < 0.85. Record: Round 1, Sequence 2, Similarity 0.72; Sequence number 3: The similarity between Vt3 and the zero vector is 0 < 0.85. Record: Round 1, Sequence number 3, Similarity 0.

[0068] The number of reference vector indices recorded after this round of screening is N=2.

[0069] S5. Calculate the first ratio P=N / M=2 / 3≈0.667. Since the first ratio P is greater than the second threshold TH2=0.6, it is determined that the consistency between the reference text T1 and the comparison text T2 is high, and the process ends directly.

[0070] In another specific embodiment, taking the similarity comparison of two objects under the same audio modality—reference audio A1 and comparison audio A2—as an example, the implementation process of this scheme is explained. The cyclic displacement step size n is 1 (i.e., only one valid comparison vector is shifted each time), the first threshold TH1 is 0.85, the second threshold TH2 is 0.7, and zero vector padding is not required. The specific process is as follows: S1. The MFCC feature extraction algorithm is used to vectorize the reference audio A1, generating two reference vectors, denoted as Va1 and Va2, which form a reference vector group V_ref=[Va1 (sequence number 1), Va2 (sequence number 2)], with the number of rows R1=2 and the number of valid sequence numbers M=2.

[0071] S2. Using the same MFCC algorithm as in step S1, the comparison audio A2 is vectorized to generate two comparison vectors Va3 and Va4, forming a comparison vector group V_cmp=[Va3,Va4] with 2 rows R2.

[0072] S3. Since the number of rows in the two vector groups is equal, there is no need to fill the zero vectors. Proceed directly to the next step.

[0073] S4. Calculate the cosine similarity based on the one-to-one correspondence of vector indices: Serial number 1: Va1 and Va3 have a similarity of 0.86 > 0.85 (first threshold TH1), so they are filtered out and not recorded; Serial number 2: Va2 and Va4 have a similarity of 0.87 > 0.85, so they are filtered out and not recorded.

[0074] The number of reference vector indices recorded after this round of filtering is N=0.

[0075] S5. Calculate the first ratio P=N / M=0 / 2=0. Since the first ratio P is less than the second threshold TH2 (0.7) at this time, it is determined that the initial round consistency is insufficient and enters the cyclic displacement stage.

[0076] S6. Perform a circular shift on the comparison vector group V_cmp=[Va3,Va4], that is, move the first vector Va3 to the end and the second vector Va4 to the beginning, and obtain a new comparison vector group V_cmp′=[Va4,Va3] after the shift.

[0077] S7. Based on the new comparison vector group V_cmp′, recalculate the cosine similarity according to the vector index: Sequence 1: Va1 and Va4 have a similarity of 0.68 < 0.85. Record: Round 2, Sequence 1, Similarity 0.68; Sequence 2: The similarity between Va2 and Va3 is 0.75 < 0.85. Record: Round 2, Sequence 2, Similarity 0.75.

[0078] Summarize all reference vector indices recorded in rounds 1 and 2: round 1 has no records, round 2 records indices 1 and 2. After merging and deduplication, a unique set of indices {1,2} is obtained, and the total number of indices after deduplication is N_total=2. The second ratio P_total is calculated again, which is 2 / 2=1. Since the second ratio P_total is greater than the second threshold TH2 at this time, it is determined that the consistency between reference audio A1 and comparison audio A2 is high, and the process terminates. This embodiment avoids cross-modal interference and improves the targeting and accuracy of the comparison by limiting the comparison to objects of the same modality. Simultaneously, it employs vector group row alignment (such as zero-vector padding) to ensure consistent feature dimensions and avoid calculation errors caused by row mismatch. Based on this, it uses cosine similarity calculation and sets a filtering threshold of 0.85 to filter high-similarity data and focus on key features with low similarity, further improving filtering accuracy. When the initial judgment is insufficient, it effectively solves the problem of misjudgment caused by vector order offset by introducing a comparison vector group cyclic displacement, multi-round comparison, and deduplication sequence number statistics mechanism, enabling comprehensive capture of similar features. Furthermore, using the percentage of recorded sequence numbers to valid sequence numbers as a quantitative judgment standard and supporting dynamic threshold adjustment makes the judgment results more objective and adaptable to different scenarios. In addition, this method is applicable to multiple single modalities such as text, image, audio, and video; the vectorization algorithm and displacement step size can be flexibly replaced, exhibiting good compatibility and broad application prospects.

[0079] Figure 2 This is a schematic diagram of the structure of a multimodal object similarity comparison system provided in an embodiment of the present invention. Figure 2 As shown, the system includes: The vectorization processing module 21 is used to perform vectorization processing on the reference object and the comparison object of the same mode to obtain the reference vector group and the comparison vector group; The row number alignment module 22 is used to align the row numbers of the reference vector group and the comparison vector group; The similarity calculation and recording module 23 is used to determine the similarity between corresponding vectors in the reference vector group and the comparison vector group after row alignment, and to record the index of the reference vector with a similarity lower than the first threshold. The initial judgment module 24 is used to determine that two objects meet the consistency requirement if the first ratio of the number of serial numbers to the number of valid reference vectors in the reference vector group reaches the second threshold. The cyclic comparison module 25 is used to rearrange the order of the valid comparison vectors in the comparison vector group when the first ratio does not reach the second threshold, and repeat the above similarity calculation and sequence number recording steps based on the rearranged comparison vector group. The second ratio is determined based on the sequence number of the deduplicated reference vector. If the second ratio reaches the second threshold, it is determined that the two objects meet the consistency requirement. If the second ratio still does not reach the second threshold after all valid comparison vectors have completed one round of sequential rearrangement, it is determined that the two objects do not meet the consistency requirement. Among them, the effective reference vector refers to the original reference vector in the reference vector group before row alignment, and the effective comparison vector refers to the original comparison vector in the comparison vector group before row alignment.

[0080] Furthermore, based on the above embodiments of the invention, the vectorization processing module 21 is specifically used for: Select the appropriate vectorization method according to the modality type; the modality type can be one of text, image, audio, or video. The reference object and the comparison object are vectorized separately according to the vectorization method to obtain the corresponding reference vector group and comparison vector group; wherein, each reference vector in the reference vector group and each comparison vector in the comparison vector group have the same dimension.

[0081] Furthermore, based on the above embodiments of the invention, the row number alignment module 22 is specifically used for: Count the number of rows in the first row of the reference vector group and the number of rows in the second row of the comparison vector group respectively; If the number of rows in the first row is not equal to the number of rows in the second row, then for the vector group with the smaller number of rows, fill the end with zero vectors with the same dimension as the original vectors, until the number of rows in the two vector groups is equal.

[0082] Furthermore, based on the above embodiments of the invention, the similarity calculation and recording module 23 is specifically used for: After aligning the rows, calculate the similarity between corresponding vectors in the reference vector group and the comparison vector group according to the vector index to obtain the similarity sequence. Each similarity in the similarity sequence is compared with the first threshold, and similarities greater than the first threshold are filtered out. Record the index of the reference vector corresponding to the retained similarity and the current round number.

[0083] Furthermore, based on the above embodiments of the invention, the initial determination module 24 is specifically used for: Count the total number of recorded serial numbers; The total number of valid reference vectors in the reference vector group is counted. Determine the first ratio between the total number of sequence numbers and the total number of valid reference vectors; If the first ratio reaches the second threshold, then the reference object and the comparison object are determined to meet the consistency requirements.

[0084] Furthermore, based on the above embodiments of the invention, the cyclic comparison module 25 is also specifically used for: If the first ratio does not reach the second threshold, then the valid comparison vectors are cyclically shifted according to the preset displacement step size, so as to move the valid comparison vectors that are currently at the front and whose number is equal to the preset displacement step size to the end of the comparison vector group. Based on the reference vector group and the rearranged comparison vector group, repeat the similarity calculation and sequence number recording steps to obtain the sequence number of the reference vector with a similarity lower than the first threshold in this round.

[0085] Furthermore, based on the above embodiments of the invention, the cyclic comparison module 25 is also specifically used for: Merge the sequence number obtained in this round with the sequence numbers obtained in all previous rounds to remove duplicates, and count the total number of sequence numbers after deduplication; Determine the second ratio between the total number of duplicated serial numbers and the total number of valid reference vectors; If the second ratio reaches the second threshold, it is determined that the reference object and the comparison object meet the consistency requirements, and the subsequent process is terminated. If the second ratio does not reach the second threshold, it is determined whether all valid comparison vectors have completed one round of cyclic displacement. If one round has not been completed, the process returns to the step of performing cyclic displacement on the valid comparison vectors according to the preset displacement step size. If one round has been completed and the second ratio still does not reach the second threshold, it is determined that the reference object and the comparison object do not meet the consistency requirements.

[0086] The multimodal object similarity comparison system provided in this embodiment of the invention can execute the multimodal object similarity comparison method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0087] Figure 3 A schematic diagram of an electronic device 30 that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0088] like Figure 3As shown, the electronic device 30 includes at least one processor 31 and a memory, such as a read-only memory (ROM) 32 or a random access memory (RAM) 33, communicatively connected to the at least one processor 31. The memory stores computer programs executable by the at least one processor. The processor 31 can perform various appropriate actions and processes based on the computer program stored in the ROM 32 or loaded from storage unit 38 into the RAM 33. The RAM 33 can also store various programs and data required for the operation of the electronic device 30. The processor 31, ROM 32, and RAM 33 are interconnected via a bus 34. An input / output (I / O) interface 35 is also connected to the bus 34.

[0089] Multiple components in electronic device 30 are connected to I / O interface 35, including: input unit 36, such as keyboard, mouse, etc.; output unit 37, such as various types of monitors, speakers, etc.; storage unit 38, such as disk, optical disk, etc.; and communication unit 39, such as network card, modem, wireless transceiver, etc. Communication unit 39 allows electronic device 30 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0090] Processor 31 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 31 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 31 performs the various methods and processes described above, such as multimodal object similarity comparison methods.

[0091] In some embodiments, the multimodal object similarity comparison method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 38. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 30 via ROM 32 and / or communication unit 39. When the computer program is loaded into RAM 33 and executed by processor 31, one or more steps of the multimodal object similarity comparison method described above may be performed. Alternatively, in other embodiments, processor 31 may be configured to perform the multimodal object similarity comparison method by any other suitable means (e.g., by means of firmware).

[0092] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0093] In some embodiments, the multimodal object similarity comparison method may be implemented as a computer program, which is implicitly included in a computer program product. When executed by a processor, the computer program implements the multimodal object similarity comparison method of the present invention. The computer program product can be understood as a software product that primarily implements its solution through a computer program. The computer program used to implement the method of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer program may be executed entirely on a machine, partially on a machine, partially on a remote machine as a standalone software package, or entirely on a remote machine or server.

[0094] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0095] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0096] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0097] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0098] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0099] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for comparing the similarity of multimodal objects, characterized in that, The method includes: Vectorize the reference object and the comparison object of the same mode to obtain the reference vector group and the comparison vector group; Align the reference vector group and the comparison vector group by row number; Determine the similarity between the corresponding vectors in the reference vector group and the comparison vector group after row alignment, and record the index of the reference vector whose similarity is lower than the first threshold; If the first ratio of the number of the sequence numbers to the number of valid reference vectors in the reference vector group reaches the second threshold, then it is determined that the two objects meet the consistency requirement. Otherwise, the valid comparison vectors in the comparison vector group are rearranged in order, and the above similarity calculation and sequence number recording steps are repeated based on the rearranged comparison vector group. The second ratio is determined based on the deduplicated reference vector sequence number. If the second ratio reaches the second threshold, it is determined that the two objects meet the consistency requirement. If the second ratio still does not reach the second threshold after all the valid comparison vectors have been rearranged in order, it is determined that the two objects do not meet the consistency requirement. Wherein, the effective reference vector refers to the original reference vector in the reference vector group before row alignment, and the effective comparison vector refers to the original comparison vector in the comparison vector group before row alignment.

2. The method according to claim 1, characterized in that, The vectorization process of the reference object and comparison object of the same mode to obtain the reference vector group and comparison vector group includes: Select the corresponding vectorization method according to the type of the modality; the type of the modality is one of text, image, audio, or video; The reference object and the comparison object are vectorized according to the vectorization method to obtain the corresponding reference vector group and the comparison vector group; wherein each reference vector in the reference vector group and each comparison vector in the comparison vector group have the same dimension.

3. The method according to claim 1, characterized in that, The row alignment of the reference vector group and the comparison vector group includes: Count the number of the first row of the reference vector group and the number of the second row of the comparison vector group respectively; If the number of rows in the first vector group is not equal to the number of rows in the second vector group, then the vector group with the smaller number of rows is filled with zero vectors with the same dimension as the original vectors until the number of rows in the two vector groups is equal.

4. The method according to claim 1, characterized in that, The process of determining the similarity between corresponding vectors in the reference vector group and the comparison vector group after row alignment, and recording the index of the reference vector with a similarity lower than a first threshold, includes: After aligning the rows, the similarity between corresponding vectors in the reference vector group and the comparison vector group is calculated according to the vector index to obtain a similarity sequence. Each similarity in the similarity sequence is compared with the first threshold, and similarities greater than the first threshold are filtered out. Record the index of the reference vector corresponding to the retained similarity and the current round number.

5. The method according to claim 1, characterized in that, If the ratio of the number of the sequence numbers to the number of valid reference vectors in the reference vector group reaches a second threshold, then it is determined that the two objects meet the consistency requirement, including: Count the total number of the recorded serial numbers; Count the total number of valid reference vectors in the reference vector group; Determine a first ratio between the total number of the sequence numbers and the total number of the valid reference vectors; If the first ratio reaches the second threshold, then it is determined that the reference object and the comparison object meet the consistency requirement.

6. The method according to claim 1, characterized in that, The step of rearranging the order of the valid comparison vectors in the comparison vector group and repeating the above similarity calculation and sequence number recording steps based on the rearranged comparison vector group includes: If the first ratio does not reach the second threshold, then the effective comparison vector is cyclically shifted according to the preset displacement step size, so as to move the effective comparison vectors that are currently at the front and whose number is equal to the preset displacement step size to the end of the comparison vector group; Based on the reference vector group and the rearranged comparison vector group, the similarity calculation and sequence number recording steps are repeated to obtain the sequence number of the reference vector with a similarity lower than the first threshold in this round.

7. The method according to claim 6, characterized in that, The second ratio is determined based on the deduplicated reference vector index. If the second ratio reaches the second threshold, the two objects are determined to meet the consistency requirement. If the second ratio still does not reach the second threshold after all valid comparison vectors have undergone one round of order rearrangement, the two objects are determined not to meet the consistency requirement, including: The sequence number obtained in this round is merged with the sequence numbers obtained in all previous rounds to remove duplicates, and the total number of sequence numbers after deduplication is counted. Determine a second ratio between the total number of deduplicated sequence numbers and the total number of valid reference vectors; If the second ratio reaches the second threshold, it is determined that the reference object and the comparison object meet the consistency requirement, and the subsequent process is terminated. If the second ratio does not reach the second threshold, it is determined whether all valid comparison vectors have completed one round of cyclic displacement. If one round has not been completed, the process returns to the step of performing cyclic displacement on the valid comparison vectors according to the preset displacement step size. If one round has been completed and the second ratio still does not reach the second threshold, it is determined that the reference object and the comparison object do not meet the consistency requirement.

8. A multimodal object similarity comparison system, characterized in that, The system includes: The vectorization processing module is used to vectorize the reference object and the comparison object of the same mode to obtain the reference vector group and the comparison vector group; A row alignment module is used to align the row numbers of the reference vector group and the comparison vector group. The similarity calculation and recording module is used to determine the similarity between the corresponding vectors in the reference vector group and the comparison vector group after row alignment, and to record the index of the reference vector whose similarity is lower than the first threshold. The initial determination module is used to determine that two objects meet the consistency requirement if the first ratio of the number of the sequence numbers to the number of valid reference vectors in the reference vector group reaches a second threshold. The cyclic comparison module is used to rearrange the valid comparison vectors in the comparison vector group when the first ratio does not reach the second threshold, and repeat the above similarity calculation and sequence number recording steps based on the rearranged comparison vector group. The second ratio is determined based on the sequence number of the deduplicated reference vector. If the second ratio reaches the second threshold, it is determined that the two objects meet the consistency requirement. If the second ratio still does not reach the second threshold after all the valid comparison vectors have completed one round of sequential rearrangement, it is determined that the two objects do not meet the consistency requirement. Wherein, the effective reference vector refers to the original reference vector in the reference vector group before row alignment, and the effective comparison vector refers to the original comparison vector in the comparison vector group before row alignment.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the multimodal object similarity comparison method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the multimodal object similarity comparison method according to any one of claims 1-7.

Images