An information retrieval method and system for comprehensive management of power data samples

Abstract

The embodiment of the present invention provides an information retrieval method and system for comprehensive management of power data samples, belonging to the field of Internet and computing technology. The information retrieval method includes: obtaining a search text and a search image input by a user; performing feature extraction on the search text to obtain a meta tag feature; searching according to the meta tag feature to obtain a first search result; performing feature extraction on the search image to obtain a scene feature; searching according to the scene feature to obtain a second search result; calculating the similarity between the first search result and the second search result; judging whether the similarity is greater than or equal to a preset threshold; and displaying the first search result and the second search result when judging that the similarity is greater than or equal to the threshold. The information retrieval method and system can improve the access efficiency to the power database.

Description

Technical Field

[0001] The present invention relates to the field of Internet and computing technology, and in particular to an information retrieval method and system for comprehensive management of power data samples. Background Art

[0002] Image data plays a vital role in the power system. They are widely used in equipment status monitoring, fault diagnosis and daily operation and maintenance, providing intuitive information support for ensuring the safe and efficient operation of the power grid. With the development of smart grid technology and the improvement of information level, the amount of power data samples representing the real-time status of power equipment has shown explosive growth. These images include but are not limited to line inspection photos of transmission lines, infrared thermal images of substation equipment, and photos of readings of various instruments.

[0003] Since a large number of electricity-related images and photos are stored, and since the similarity of power equipment icons is high, traditional image retrieval methods are difficult to achieve high-precision retrieval, thereby reducing the access efficiency of the power database. Summary of the invention

[0004] The purpose of the embodiments of the present invention is to provide an information retrieval method and system for comprehensive management of power data samples, which can improve the access efficiency to the power database.

[0005] In order to achieve the above object, an embodiment of the present invention provides an information retrieval method for comprehensive management of power data samples, comprising:

[0006] Get the search text and search image entered by the user;

[0007] Extracting features from the search text to obtain meta tag features;

[0008] Searching according to the meta tag feature to obtain a first search result;

[0009] Performing feature extraction on the search image to obtain scene features;

[0010] Searching according to the scene feature to obtain a second search result;

[0011] Calculating the similarity between the first search result and the second search result;

[0012] Determining whether the degree of similarity is greater than or equal to a preset threshold;

[0013] When it is determined that the degree of similarity is greater than or equal to the threshold, the first search result and the second search result are displayed.

[0014] Optionally, feature extraction is performed on the search text to obtain meta tag features, including:

[0015] Performing a preprocessing operation on the search text;

[0016] Obtaining a candidate keyword graph of the search text after preprocessing operation;

[0017] A search keyword is obtained from the candidate keyword graph.

[0018] Optionally, searching according to the meta tag feature to obtain a first search result includes:

[0019] The word2vec model is used to map the search keyword into a low-dimensional vector space to obtain a corresponding first low-dimensional vector;

[0020] The Doc2vec model is used to map the text to be searched into the low-dimensional vector space to obtain a corresponding second low-dimensional vector;

[0021] Calculating the distance between the first low-dimensional vector and the second low-dimensional vector;

[0022] Using the pagerank algorithm to calculate the weight of the search keyword in the candidate keyword graph;

[0023] The first search result is screened out from the text to be searched according to the weight and the distance.

[0024] Optionally, the weight of the search keyword is calculated in the candidate keyword graph using a pagerank algorithm, including:

[0025] The weight is calculated according to formula (1):

[0026] , (1)

[0027] in, For keywords The weight of is the damping coefficient, is the total number of nodes in the graph, For and represent keywords The node is the set of nodes connected to it. For and represent keywords The node is the set of nodes connected to it. For keywords The weight of For keywords and keywords The weight of the edge between For keywords and keywords The weight of the edge between .

[0028] Optionally, calculating the distance between the first low-dimensional vector and the second low-dimensional vector includes:

[0029] The distance is calculated according to formula (2):

[0030] , (2)

[0031] in, For the The first low-dimensional vector With The distance between the second lowest dimensional vector, is the dimension of the low-dimensional vector space, is the intermediate sequence variable, is the first low-dimensional vector The elements, is the first elements.

[0032] Optionally, feature extraction is performed on the search image to obtain scene features, including:

[0033] Performing an image block embedding operation on the search image to obtain a feature map obtained by combining multiple image blocks;

[0034] Adding a corresponding learnable embedding function to each of the image blocks;

[0035] Adding a corresponding position embedding function to each of the image blocks;

[0036] The feature map including the image block, the learnable embedding function, and the position embedding function is input into a preset feature extraction network to obtain the scene feature.

[0037] Optionally, the feature extraction network includes:

[0038] Multiple MSA units, including:

[0039] A multi-head attention mechanism for performing weighted operations on the image blocks, learnable embedding functions, and position embedding functions in the input feature map;

[0040] A layer normalization network, used to perform a normalization operation on the weighted feature map;

[0041] A skip connection network is used to extract features from the feature map after the normalization operation to obtain a corresponding feature vector;

[0042] Multiple MLP units, including:

[0043] A feedforward network, used for processing and transforming the feature vector to obtain corresponding extracted features;

[0044] A layer normalization unit, used for performing a normalization operation on the extracted features;

[0045] The classification head network is used to perform a classification operation on the extracted features after the normalization operation to obtain the scene features.

[0046] Optionally, calculating the similarity between the first search result and the second search result includes:

[0047] respectively using the LBP algorithm and the RLBP algorithm to extract corresponding feature histograms from the first search results;

[0048] Merging the two feature histograms to obtain a first feature histogram;

[0049] respectively using the LBP algorithm and the RLBP algorithm to extract corresponding feature histograms from the second search result;

[0050] Merging the two feature histograms to obtain a second feature histogram;

[0051] A cross distance between the first feature histogram and the second feature histogram is calculated to obtain the degree of similarity.

[0052] Optionally, calculating the cross distance between the first feature histogram and the second feature histogram to obtain the similarity includes:

[0053] The crossover distance is calculated according to formula (3):

[0054] , (3)

[0055] in, The first feature histogram And the second feature histogram The crossover distance between is the number of fused sub-images, is the length of the generated fusion feature vector, is the number of intervals divided in the feature histogram, is the intermediate sequence variable.

[0056] On the other hand, the present invention further provides an information retrieval system for integrated management of power data samples, the system comprising a processor, and the processor is configured to execute any of the methods described above.

[0057] Through the above technical solution, the present invention provides an information retrieval method and system for comprehensive management of power data samples, which realizes accurate retrieval positioning by using the results of keyword retrieval and image retrieval at the same time, and then combining the comparison of the two retrieval results. Compared with the single retrieval method in the prior art, the information retrieval method and system provided by the present invention improves the data access efficiency of the power database.

[0058] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific implementations, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings:

[0060] Figure 1 is a flow chart of an information retrieval method for comprehensive management of power data samples according to an embodiment of the present invention;

[0061] Figure 2 is a flow chart of a method for obtaining meta tag features according to an embodiment of the present invention;

[0062] Figure 3 is a flow chart of a method for searching based on meta tag features according to one embodiment of the present invention;

[0063] Figure 4 is a flow chart of a method for acquiring scene features according to an embodiment of the present invention;

[0064] Figure 5 is a schematic diagram of an image plate according to one embodiment of the present invention;

[0065] Figure 6 is an example block diagram of a feature extraction network according to one embodiment of the present invention;

[0066] Figure 7 is a flow chart of a method for calculating similarity according to one embodiment of the present invention;

[0067] Figure 8 is an exemplary diagram of a process of calculating a crossover distance according to one embodiment of the present invention;

[0068] Fig. 9 It is a flowchart of an information retrieval method for comprehensive management of power data samples according to an embodiment of the present invention. DETAILED DESCRIPTION

[0069] The specific implementation of the embodiment of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the embodiment of the present invention, and is not used to limit the embodiment of the present invention.

[0070] like Figure 1 FIG. 1 is a flowchart of an information retrieval method for comprehensive management of power data samples according to an embodiment of the present invention. Figure 1 In the method, the information retrieval method may include the following steps:

[0071] In step S10, the search text and search image input by the user are obtained;

[0072] In step S11, feature extraction is performed on the search text to obtain meta tag features;

[0073] In step S12, searching is performed according to the meta tag feature to obtain a first search result;

[0074] In step S13, feature extraction is performed on the search image to obtain scene features;

[0075] In step S14, searching is performed according to the scene feature to obtain a second search result;

[0076] In step S15, the similarity between the first search result and the second search result is calculated;

[0077] In step S16, it is determined whether the degree of similarity is greater than or equal to a preset threshold;

[0078] In step S17, when it is determined that the degree of similarity is greater than or equal to the threshold, the first search result and the second search result are displayed.

[0079] In this Figure 1 In the method shown, step S10 can be used to obtain the search text and search image input by the user. Step S11 can be used to extract features from the search text to obtain corresponding meta tag features. The method for extracting the meta tag can be in various forms known to those skilled in the art. In one example of the present invention, the specific method for obtaining the meta tag feature may include: Figure 2 In the steps shown in Figure 2 In the method for obtaining the meta tag feature, the method may include the following steps:

[0080] In step S20, the search text is preprocessed. Specifically, in this example, the preprocessing operation may be to first perform a word segmentation operation on the text and remove stop words, thereby screening out candidate keywords. More specifically, the preprocessing operation may be to convert the target document into Sentence segmentation is performed, that is , using the Chinese word segmentation tool Jieba to analyze the sentence ∈ Perform word segmentation and remove stop words by loading the stop word list, retaining the words that meet the conditions after segmentation, that is, ,in ∈ Candidate keywords.

[0081] In step S21, a candidate keyword graph of the search text after the preprocessing operation is obtained. Specifically, in this example, the keyword graph can be represented as ,in, is the set of vertices (keywords) in the graph, is the set of edges (co-occurrence relations) in the graph.

[0082] In step S22, the search keyword is acquired from the candidate keyword graph.

[0083] Step S12 may be used to search according to the meta tag feature to obtain the first search result. The specific method for searching according to the meta tag feature may be in various forms known to those skilled in the art. In one example of the present invention, the specific method for searching according to the meta tag feature may include: Figure 3 In the steps shown in Figure 3 In the specific method of searching according to the meta tag feature, the following steps may be included:

[0084] In step S30, the search keyword is mapped into a low-dimensional vector space using a word2vec model to obtain a corresponding first low-dimensional vector;

[0085] In step S31, the Doc2vec model is used to map the text to be searched into a low-dimensional vector space to obtain a corresponding second low-dimensional vector;

[0086] In step S32, the distance between the first low-dimensional vector and the second low-dimensional vector is calculated. Specifically, in this example, the distance can be calculated using the following formula (2):

[0087] , (2)

[0088] in, For the The first low-dimensional vector and the The distance between the second lowest dimensional vector, is the dimension of the low-dimensional vector space, is the intermediate sequence variable, is the first low-dimensional vector The elements, is the first elements, in this example, .

[0089] In step S33, the pagerank algorithm is used to calculate the weight of the search keyword in the candidate keyword graph. Specifically, in this example, the weight can be calculated using the following formula (1):

[0090] , (1)

[0091] in, For keywords The weight of is the damping coefficient, is the total number of nodes in the graph, For and represent keywords The node is the set of nodes connected to it. For and represent keywords The node is the set of nodes connected to it. For keywords The weight of For keywords and keywords The weight of the edge between For keywords and keywords The weight of the edge between them. Formula (1) calculates the weight of the current candidate keyword by introducing the weight of the associated keyword itself, thereby taking into account the contribution of each keyword and improving the accuracy of the weight.

[0092] In step S34, the first search result is screened out from the text to be searched according to the weight and the distance.

[0093] Step S13 is used to extract features from the searched image to obtain scene features. The specific method for obtaining the scene features may be in various forms known to those skilled in the art. In one example of the present invention, the method for obtaining the scene features may include: Figure 4 In this step. Figure 4 In the method for obtaining the scene feature, the method may include the following steps:

[0094] In step S40, the search image is subjected to an image block embedding operation to obtain a feature map obtained by combining multiple image blocks. Specifically, in this example, step S40 may be to first represent the search image as . Then reconstruct the retrieval image, and the resulting block sequence can be expressed as .in, Indicates the resolution of the retrieved image. Indicates the number of channels, represents the resolution of the image block, is the effective input sequence length. The corresponding example is shown in Figure 5 shown.

[0095] In step S41, a corresponding learnable embedding function is added to each image block. Specifically, step S41 may be to first perform a linear transformation operation on the image block, thereby obtaining a function represented as Then perform dimensionality reduction operation to obtain the representation as image blocks.

[0096] In step S42, a corresponding position embedding function is added to each image block;

[0097] In step S43, the feature map including the image block, the learnable embedding function, and the position embedding function is input into a preset feature extraction network to obtain scene features. The specific structure of the feature extraction network can be in various forms known to those skilled in the art. In one example of the present invention, Figure 6 As shown, the feature extraction network may include multiple MSA units 1 (Multi-head self-attention Block) and multiple MLP units (Multi-Layer Perceptron). Among them, the MSA unit 1 includes a multi-head attention mechanism 11, a layer normalization network 12, and a jump connection network 13. Among them, the multi-head attention mechanism 11 can be used to perform weighted operations on image blocks, learnable embedding functions, and position embedding functions in the input feature map. The layer normalization network 12 can be used to standardize the weighted feature map. The jump connection network 13 can be used to extract features from the feature map after the standardization operation to obtain the corresponding feature vector. In this example, the structure of the MSA unit can also be as follows Figure 6 As shown. And the processing process of the MSA unit can also be expressed by the following formula (4):

[0098] , (4)

[0099] in, For the The output of each MSA unit, represents a skip connection network, For the The output of each MSA unit, is the number of MSA units.

[0100] The MLP unit 2 may include a feedforward network 21, a layer normalization unit 22, and a classification head network 23. The feedforward network 21 may be used to process and transform the feature vector to obtain the corresponding extracted features. The layer normalization unit 22 may be used to perform a normalization operation on the extracted features. The classification head network 23 may be used to perform a classification operation on the extracted features after the normalization operation to obtain scene features. In this example, the MLP unit 2 may also be represented by the following formula (5):

[0101] , (5)

[0102] in, For the The output of the MLP unit, represents a skip connection network, For the The output of each MSA unit, is the number of MLP units.

[0103] Step S15 can be used to calculate the similarity between the first search result and the second search result. The specific calculation method of the similarity can be various forms known to those skilled in the art. In one example of the present invention, the similarity can be calculated using the following method: Figure 7 Specifically, in this Figure 7 In the method for calculating the similarity, the method may include the following steps:

[0104] In step S50, the LBP algorithm and the RLBP algorithm are respectively used to extract the corresponding feature histogram on the first search result;

[0105] In step S51, the two feature histograms are fused to obtain a first feature histogram;

[0106] In step S52, the LBP algorithm and the RLBP algorithm are respectively used to extract the corresponding feature histogram on the second search result;

[0107] In step S53, the two feature histograms are fused to obtain a second feature histogram;

[0108] In step S54, the cross distance between the first feature histogram and the second feature histogram is calculated to obtain the degree of similarity. Specifically, in an example of the present invention, an example diagram of the process of calculating the cross distance may also be as follows: Figure 8 In addition, the crossover distance can also be calculated using the following formula (3):

[0109] , (3)

[0110] in, The first feature histogram And the second feature histogram The crossover distance between is the number of fused sub-images, is the length of the generated fusion feature vector, is the number of intervals divided in the feature histogram, is the intermediate sequence variable.

[0111] Step S16 can be used to determine whether the calculated similarity is greater than or equal to a preset threshold. If it is determined that the similarity is greater than or equal to the threshold, it means that the match is successful, and the first search result and the second search result can be directly displayed. On the contrary, if the similarity is less than the threshold, it means that the search is wrong, and the process can return to step S11, that is, re-perform feature search, so as to obtain new first search result and second search result.

[0112] In addition, considering that after the first match fails, if the previous search logic is still executed, the previous search process may be repeated. Therefore, in one embodiment of the present invention, when obtaining the first search result, step S12 can generate multiple first search results at the same time, and arrange them in descending order according to the matching degree of the search sample. Similarly, step S14 can also generate multiple second search results, and arrange them in descending order according to the matching degree.

[0113] In the case where the similarity is less than the threshold value in step S16, the first search result with a lower matching degree and the second search result can be selected to calculate the similarity. If the match is unsuccessful again, the previous first search result and the second search result with a lower matching degree can be used to calculate the similarity. This process is repeated until the first search result and the second search result that meet the conditions are found. Specifically, the information retrieval method can also include the following steps: Fig. 9 In the steps shown in Fig. 9 In the method, the information retrieval method may include the following steps:

[0114] In step S60, the search text and search image input by the user are obtained;

[0115] In step S61, feature extraction is performed on the search text to obtain meta tag features;

[0116] In step S62, a search is performed according to the meta tag feature to obtain a plurality of first search results, and the first search results are sorted in descending order according to the matching degree;

[0117] In step S63, feature extraction is performed on the search image to obtain scene features;

[0118] In step S64, a search is performed according to the scene features to obtain a plurality of second search results, and the second search results are sorted in descending order according to the matching degree;

[0119] In step S65, the first search result and the second search result are selected in descending order of matching degree;

[0120] In step S66, the similarity between the first search result and the second search result is calculated;

[0121] In step S67, it is determined whether the degree of similarity is greater than or equal to a preset threshold;

[0122] In step S68, if it is determined that the similarity is greater than or equal to the threshold, the first search result and the second search result are displayed;

[0123] In step S69, when it is determined that the similarity is less than the threshold, it is determined whether the current value of the number of iterations is an odd number or an even number;

[0124] In step S70, when it is determined that the value of the iteration number is an odd number, the first search result with a lower matching degree and the current second search result are selected, and the process returns to the step of calculating the similarity between the first search result and the second search result, that is, returns to step S66;

[0125] In step S71, when it is determined that the value of the iteration number is an even number, the second search result with a lower matching degree and the current first search result are selected, and the process returns to the step of calculating the similarity between the first search result and the second search result, that is, returns to step S66.

[0126] On the other hand, the present invention further provides an information retrieval system for integrated management of power data samples, the system comprising a processor, and the processor is configured to execute any of the methods described above.

[0127] Through the above technical solution, the present invention provides an information retrieval method and system for comprehensive management of power data samples, which realizes accurate retrieval positioning by using the results of keyword retrieval and image retrieval at the same time, and then combining the comparison of the two retrieval results. Compared with the single retrieval method in the prior art, the information retrieval method and system provided by the present invention improves the data access efficiency of the power database.

[0128] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0129] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0130] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0131] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0132] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0133] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0134] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0135] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0136] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. An information retrieval method for comprehensive management of power data samples, characterized in that: The information retrieval method comprises: Get the search text and search image entered by the user; Performing feature extraction on the search text to obtain meta tag features; Searching according to the meta tag feature to obtain a first search result; Performing feature extraction on the search image to obtain scene features; Searching according to the scene feature to obtain a second search result; Calculating the similarity between the first search result and the second search result; Determining whether the degree of similarity is greater than or equal to a preset threshold; If it is determined that the similarity is greater than or equal to the threshold, displaying the first search result and the second search result; Calculating the similarity between the first search result and the second search result includes: respectively using the LBP algorithm and the RLBP algorithm to extract corresponding feature histograms from the first search results; Merging the two feature histograms to obtain a first feature histogram; respectively using the LBP algorithm and the RLBP algorithm to extract corresponding feature histograms from the second search result; Merging the two feature histograms to obtain a second feature histogram; A cross distance between the first feature histogram and the second feature histogram is calculated to obtain the degree of similarity.

2. The information retrieval method according to claim 1, characterized in that: Performing feature extraction on the search text to obtain meta tag features, including: Performing a preprocessing operation on the search text; Obtaining a candidate keyword graph of the search text after preprocessing operation; A search keyword is obtained from the candidate keyword graph.

3. The information retrieval method according to claim 2, characterized in that: Searching according to the meta tag feature to obtain a first search result includes: The word2vec model is used to map the search keyword into a low-dimensional vector space to obtain a corresponding first low-dimensional vector; The Doc2vec model is used to map the text to be searched into the low-dimensional vector space to obtain a corresponding second low-dimensional vector; Calculating the distance between the first low-dimensional vector and the second low-dimensional vector; Using the pagerank algorithm to calculate the weight of the search keyword in the candidate keyword graph; The first search result is screened out from the text to be searched according to the weight and the distance.

4. The information retrieval method according to claim 3, characterized in that: The weight of the search keyword is calculated in the candidate keyword graph using the pagerank algorithm, including: The weight is calculated according to formula (1): ，（1） in, For keywords The weight of is the damping coefficient, is the total number of nodes in the graph, For and represent keywords The node is the set of nodes connected to it. For and represent keywords The node is the set of nodes connected to it. For keywords The weight of For keywords and keywords The weight of the edge between For keywords and keywords The weight of the edge between .

5. The information retrieval method according to claim 3, characterized in that: Calculating the distance between the first low-dimensional vector and the second low-dimensional vector includes: The distance is calculated according to formula (2): ，（2） in, For the The first low-dimensional vector With The distance between the second lowest dimensional vectors, is the dimension of the low-dimensional vector space, is the intermediate sequence variable, is the first low-dimensional vector The elements, is the first elements.

6. The information retrieval method according to claim 1, characterized in that: Performing feature extraction on the search image to obtain scene features includes: Performing an image block embedding operation on the search image to obtain a feature map obtained by combining multiple image blocks; Adding a corresponding learnable embedding function to each of the image blocks; Adding a corresponding position embedding function to each of the image blocks; The feature map including the image block, the learnable embedding function, and the position embedding function is input into a preset feature extraction network to obtain the scene feature.

7. The information retrieval method according to claim 6, characterized in that: The feature extraction network includes: Multiple MSA units, including: A multi-head attention mechanism for performing weighted operations on the image blocks, learnable embedding functions, and position embedding functions in the input feature map; A layer normalization network, used to perform a normalization operation on the weighted feature map; A skip connection network is used to extract features from the feature map after the normalization operation to obtain a corresponding feature vector; Multiple MLP units, including: A feedforward network, used for processing and transforming the feature vector to obtain corresponding extracted features; A layer normalization unit, used for performing a normalization operation on the extracted features; The classification head network is used to perform a classification operation on the extracted features after the normalization operation to obtain the scene features.

8. The information retrieval method according to claim 1, characterized in that: Calculating a cross distance between the first feature histogram and the second feature histogram to obtain the similarity degree includes: The crossover distance is calculated according to formula (3): ，（3） in, The first feature histogram And the second feature histogram The crossover distance between is the number of fused sub-images, is the length of the generated fusion feature vector, is the number of intervals divided in the feature histogram, is the intermediate sequence variable.

9. An information retrieval system for comprehensive management of power data samples, characterized in that: The system comprises a processor configured to execute the method according to any one of claims 1 to 8.

Images