Picture duplicate checking method based on vector similarity search engine and related equipment

Abstract

The invention belongs to the field of artificial intelligence and big data, is applied to the field of financial science and technology, and relates to a picture duplicate checking method based on a vector similarity search engine. Performing feature extraction on the target picture to obtain a picture feature vector; performing vector segmentation and quantization on the picture feature vector through a vector similarity search engine algorithm to obtain a target query vector; searching a pre-stored historical vector sample set similar to the target query vector through the vector similarity search engine algorithm; and performing similarity comparison on the quantized target query vector and each historical vector sample in a historical vector sample set to obtain a duplicate checking result of the target picture. The invention further provides a picture duplicate checking device based on the vector similarity search engine, computer equipment and a storage medium. The stability of a picture duplicate checking system can be effectively improved, and the timeliness of picture duplicate checking can be improved.

Description

Technical Field

[0001] The present application relates to the fields of artificial intelligence and big data technology, and in particular to a method for checking for duplicate images based on a vector similarity search engine and related equipment. Background Art

[0002] In today's digital age, image duplication detection technology plays a vital role in agricultural insurance business in the financial field. In the process of agricultural insurance underwriting and claims settlement, accurate and efficient image duplication detection can effectively ensure business compliance, reduce risks, and improve service quality.

[0003] At present, in the agricultural insurance business process, the image duplication check interface is widely used in multiple scenarios. For example, on the mobile side, when the user submits the inspection data, it is necessary to display whether the image is repeated with the historical image, and it cannot hinder the business process; on the PC side, whether it is uploading attachments to submit inspection data, or supplementing the information, as well as in the underwriting process, it is necessary to check the image for duplicates. In addition, when the user uploads the inspection picture or attachment, the claims system and the claims middle platform will automatically trigger the interface call in real time to check the image for duplicates, and prompt whether there is duplication when viewing the attachment, which also cannot hinder the business process.

[0004] In the existing technology, deep learning models such as ResNet are usually used to extract image feature vectors, and then similarity comparison is performed through a vector database to achieve image duplication detection. Taking the commonly used vector database faiss as an example, it can complete the image duplication detection task well under a certain scale of data volume. However, with the continuous development of agricultural insurance business, there are hundreds of requests every day, each request contains dozens to hundreds of pictures, and historical business needs require full processing of historical data and real-time feedback of results. In such a business scenario, faiss gradually exposed the following problems:

[0005] 1. Insufficient system stability: As time goes by, the duplicate checking database continues to expand. The number of duplicate images found in the past year has reached 4,878, and the database will continue to grow. In this case, when Faiss processes large-scale data, the system stability gradually decreases, and it often issues alarms during the operation process and needs to restart the entire service, affecting the normal operation of agricultural insurance underwriting and claims settlement.

[0006] 2. Timeliness is difficult to guarantee: Under the premise of ensuring a certain timeliness, Faiss can only use the memory resident form to guarantee one concurrency. If a multi-concurrency solution is to be adopted, information interaction between memories is required, which requires technical experts to evaluate and is very difficult; and the use of asynchronous processing, queuing and other solutions requires the core of agricultural insurance to cooperate with scheduling solutions, which takes a long time. This makes it impossible to give results in time when facing a large number of image duplicate checking requests. Summary of the invention

[0007] The purpose of the embodiments of the present application is to propose a method and related equipment for image duplication checking based on a vector similarity search engine, which can effectively improve the stability of the image duplication checking system, avoid frequent alarms and service restarts, and at the same time improve the timeliness of image duplication checking, and meet the needs of rapid response to a large number of image duplication checking requests.

[0008] In order to solve the above technical problems, the embodiment of the present application provides a method for checking duplicate images based on a vector similarity search engine, which adopts the following technical solution:

[0009] Get the target image to be checked for duplicates;

[0010] Extracting features of the target image to obtain an image feature vector;

[0011] The image feature vector is segmented and quantized by a vector similarity search engine algorithm to obtain a target query vector;

[0012] Retrieving a pre-stored historical vector sample set similar to the target query vector through the vector similarity search engine algorithm; each historical vector sample in the historical vector sample set is pre-stored after being indexed by a picture feature vector extracted from a historical picture through the vector similarity search engine algorithm;

[0013] The quantized target query vector is compared with each historical vector sample in the historical vector sample set for similarity to obtain a duplicate checking result of the target image.

[0014] Furthermore, after segmenting and quantizing the image feature vector by the vector similarity search engine algorithm to obtain a target query vector, and before retrieving a pre-stored historical vector sample set similar to the target query vector by the vector similarity search engine algorithm, the method further includes:

[0015] The target query vector is stored as a new feature vector sample through a vector similarity search engine algorithm.

[0016] Furthermore, storing the target query vector as a new feature vector sample by using a vector similarity search engine algorithm includes:

[0017] Dividing each quantized target query vector into a preset number of slices to obtain vector slices;

[0018] Each of the vector slices is compressed, and each compressed vector slice is independently stored as a new feature vector sample.

[0019] Furthermore, the image feature vector is segmented and quantized by a vector similarity search engine algorithm to obtain a target query vector, including:

[0020] Segmenting the image feature vectors one by one to obtain feature vector segments;

[0021] Product quantization is performed on each of the feature vector segments, and the quantization center closest to each of the feature vector segments is found from a preset quantization center set to obtain the quantized feature vector segment, and the quantized feature vector segment is used as the target query vector.

[0022] Furthermore, the retrieving a pre-stored historical vector sample set similar to the target query vector by using the vector similarity search engine algorithm includes:

[0023] Using the inverted file index structure in the quantization-based vector retrieval algorithm in the vector similarity search engine algorithm to search, and finding the quantization center corresponding to the target query vector in the index;

[0024] According to the quantization center, a set of pre-stored historical vector samples similar to the target query vector is determined.

[0025] Furthermore, the quantized target query vector is compared with each historical vector sample in the historical vector sample set for similarity to obtain a duplicate checking result of the target image, including:

[0026] Performing cosine similarity calculation on the quantized target query vector and each historical vector sample in the historical vector sample set to obtain a similarity result;

[0027] Compare all the calculated similarity results with a preset similarity threshold to obtain a comparison result;

[0028] If the comparison result is that the similarity result is greater than the similarity threshold, it is determined that the target image is repeated with the corresponding historical image, and a duplicate alarm is triggered.

[0029] Furthermore, the step of extracting features from the target image to obtain an image feature vector includes:

[0030] The pre-trained ResNet50 model is used to extract features of the target image to obtain an image feature vector.

[0031] In order to solve the above technical problems, the embodiment of the present application also provides an image duplication checking device based on a vector similarity search engine, which adopts the following technical solution:

[0032] The acquisition module is used to obtain the target image to be checked for duplicates;

[0033] A feature extraction module is used to extract features from the target image to obtain a feature vector of the image;

[0034] An indexing module, used for segmenting and quantizing the image feature vector by a vector similarity search engine algorithm to obtain a target query vector;

[0035] A retrieval module is used to retrieve a set of pre-stored historical vector samples that are similar to the target query vector through the vector similarity search engine algorithm; each historical vector sample in the set of historical vector samples is pre-stored after being indexed by a picture feature vector extracted from a historical picture through the vector similarity search engine algorithm;

[0036] The comparison module is used to compare the similarity between the quantized target query vector and each historical vector sample in the historical vector sample set to obtain the duplicate checking result of the target image.

[0037] In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical solution:

[0038] It includes a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the steps of the image duplication checking method based on the vector similarity search engine described in the above embodiment are implemented.

[0039] In order to solve the above technical problems, the embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solution:

[0040] The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the steps of the method for checking for duplicate images based on a vector similarity search engine described in the above embodiment are implemented.

[0041] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0042] The embodiment of the present application obtains the target image to be checked for duplicates, performs feature extraction on it to obtain an image feature vector, then uses a vector similarity search engine algorithm to segment and quantize the vector to obtain a target query vector, and then uses the vector similarity search engine algorithm to retrieve vectors similar to the target query vector in a pre-stored historical vector sample set that is indexed by feature vectors extracted from historical images, and finally compares the target query vector with each vector in the historical vector sample set for similarity, thereby determining the duplicate check result of the target image. The embodiment of the present application uses an advanced vector similarity search engine algorithm for vector processing and retrieval, which can effectively improve the stability of the image duplicate check system in the agricultural insurance business, avoid frequent alarms and restart services, and improve timeliness, meet the needs of rapid response to a large number of image duplicate check requests, and ensure the smooth progress of business processes such as agricultural insurance underwriting and claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0044] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;

[0045] Figure 2 A flowchart of an embodiment of a method for checking for duplicate images based on a vector similarity search engine according to the present application;

[0046] Figure 3 It is a structural schematic diagram of an embodiment of an image duplication checking device based on a vector similarity search engine according to the present application;

[0047] Figure 4 It is a structural diagram of an embodiment of a computer device according to the present application. DETAILED DESCRIPTION

[0048] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0049] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0050] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0051] like Figure 1 As shown, the system architecture 100 may include a terminal device 101, a network 102 and a server 103. The terminal device 101 may be a laptop 1011, a tablet computer 1012 or a mobile phone 1013. The network 102 is used to provide a medium for a communication link between the terminal device 101 and the server 103. The network 102 may include various connection types, such as wired, wireless communication links or optical fiber cables.

[0052] The user can use the terminal device 101 to interact with the server 103 through the network 102 to receive or send messages, etc. Various communication client applications can be installed on the terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0053] The terminal device 101 can be any electronic device with a display screen and supporting web browsing. In addition to a laptop computer 1011, a tablet computer 1012 or a mobile phone 1013, the terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 (Moving Picture Experts Group Audio Layer IV), a laptop computer, a desktop computer, etc.

[0054] The server 103 may be a server that provides various services, such as a background server that provides support for web pages displayed on the terminal device 101 .

[0055] It should be noted that the image duplication checking method based on the vector similarity search engine provided in the embodiment of the present application is generally executed by a server, and accordingly, the image duplication checking device based on the vector similarity search engine is generally arranged in the server.

[0056] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0057] Continue to refer Figure 2 , shows a flow chart of an embodiment of a method for checking duplicate images based on a vector similarity search engine according to the present application. The method for checking duplicate images based on a vector similarity search engine comprises the following steps:

[0058] S10, obtaining the target image to be checked for duplication;

[0059] S11, extracting features from the target image to obtain an image feature vector;

[0060] S12, segmenting and quantizing the image feature vector by a vector similarity search engine algorithm to obtain a target query vector;

[0061] S13, retrieving a pre-stored historical vector sample set similar to the target query vector through the vector similarity search engine algorithm; each historical vector sample in the historical vector sample set is pre-stored after being indexed by a picture feature vector extracted from a historical picture through the vector similarity search engine algorithm;

[0062] S14, performing a similarity comparison between the quantized target query vector and each historical vector sample in the historical vector sample set to obtain a duplicate checking result of the target image.

[0063] The embodiment of the present application obtains the target image to be checked for duplicates, performs feature extraction on it to obtain an image feature vector, then uses a vector similarity search engine algorithm to segment and quantize the vector to obtain a target query vector, and then uses the vector similarity search engine algorithm to retrieve vectors similar to the target query vector in a pre-stored historical vector sample set that is indexed by feature vectors extracted from historical images, and finally compares the target query vector with each vector in the historical vector sample set for similarity, thereby determining the duplicate check result of the target image. The embodiment of the present application uses an advanced vector similarity search engine algorithm for vector processing and retrieval, which can effectively improve the stability of the image duplicate check system in the agricultural insurance business, avoid frequent alarms and restart services, and improve timeliness, meet the needs of rapid response to a large number of image duplicate check requests, and ensure the smooth progress of business processes such as agricultural insurance underwriting and claims.

[0064] In some optional implementations, after step S12 and before step S13, the method further includes:

[0065] S121, storing the target query vector as a new feature vector sample through a vector similarity search engine algorithm.

[0066] After the present embodiment performs vector segmentation and quantization of the image feature vector by the vector similarity search engine algorithm to obtain the target query vector, the target query vector is further stored as a new feature vector sample by the vector similarity search engine algorithm, so that the powerful function of the vector similarity search engine can be fully utilized. In the process of checking for duplicates, not only historical data is retrieved, but also new query vectors are stored as samples in a timely manner, providing a richer data basis for subsequent checking for duplicates. Therefore, the present embodiment dynamically updates the feature vector sample library in the checking for duplicates process, so that the database can continuously adapt to new image data, and improve the accuracy and comprehensiveness of checking for duplicates. In addition, compared with the traditional checking for duplicates method, this method of storing new vectors in real time is more intelligent and efficient, and the updating of the sample library can be realized without additional complex operations. In summary, the present embodiment continuously expands the feature vector sample library, so that subsequent checking for duplicates can refer to more data, reduce the situation of missed checking and mischecking, and can further improve the accuracy and timeliness of checking for duplicates in images. At the same time, due to the high efficiency of the vector similarity search engine algorithm, even if new samples are continuously added, a faster retrieval speed can be maintained to ensure the efficient operation of the checking for duplicates system.

[0067] For example, in the agricultural insurance underwriting business scenario, when the insurance agent submits the target image in the verification data through the mobile terminal or PC terminal, the system first extracts features, vector segments and quantifies the target image to obtain the target query vector, and stores it as a new feature vector sample. Then, the system retrieves a pre-stored set of historical vector samples similar to the target query vector through the vector similarity search engine algorithm. For example, in the underwriting stage, if the agent uploads a new picture of crop damage, the system will not only compare it with the historical disaster-stricken pictures when checking for duplicates, but also store the feature vector of the new picture. In this way, as the business continues, the sample library becomes richer and richer, and the subsequent duplicate checking results will be more accurate. When a large number of similar disaster-stricken pictures appear, the system can promptly detect potential fraud risks and ensure the stability and fairness of agricultural insurance underwriting business.

[0068] Specifically, the step S121 includes:

[0069] S1210, dividing each quantized target query vector into a preset number of slices to obtain vector slices;

[0070] S1211, compress each of the vector slices, and independently store each compressed vector slice as a new feature vector sample.

[0071] In this embodiment, after the target image is subjected to feature extraction, vector segmentation and quantization to obtain the target query vector, it is first stored as a new feature vector sample and then a similar historical vector sample set is retrieved. The specific process is to divide the target query vector into a preset number of slices, then compress each vector slice, and independently store the compressed vector slices as new feature vector samples. This embodiment makes full use of the characteristics of the vector similarity search engine algorithm to timely update the sample library during the duplicate checking process, while improving storage efficiency and retrieval speed through sharding and compression. In addition, by adopting compression technology, the storage space is greatly reduced without significantly reducing the accuracy, adapting to the growing amount of data, ensuring that even if the amount of data increases, it can be quickly stored and retrieved, thereby improving system performance.

[0072] Specifically, after the target query vector is segmented and quantized by the vector similarity search engine algorithm, it is divided into a preset number of shards in order to further improve storage and retrieval efficiency. This preset number can be adjusted according to actual conditions, such as data scale, hardware resources, and requirements for retrieval speed.

[0073] Assume that the target query vector is V = (v 1 , v 2 , …, v 512 ), if the preset number of shards is 4, then it can be divided into 4 shards, namely V 1 =(v 1 , v 2 , …, v 128 ), V 2 =(v 129 , v 130 , …, v 256 ), V 3 =(v 257 , v 258 , …, v 384 ), V 4 =(v 385 , v 386 , …, v 512 ).

[0074] Each vector slice is compressed to reduce storage space and speed up storage and retrieval. A variety of compression algorithms can be used, such as product quantization (PQ) technology.

[0075] Taking PQ technology as an example, for each vector slice, find a set of quantization centers that are closest to it. Suppose a vector slice is V i =(v i，1 , v i，2 , …, v i，n ), the quantization center set is C = {c1 , c 2 , …, c k}, where each quantization center is also a vector with the same dimension as the vector slice. Calculate the distance between the vector slice and each quantization center, find the quantization center closest to it, and approximate the vector slice to the quantization center. This can greatly reduce the storage space, and at the same time, the approximate original vector can be restored for comparison by decompression and similarity calculation during retrieval.

[0076] Storing each compressed vector slice independently as a new feature vector sample means that each slice can be stored in a different location or storage device. The benefit of this is that it can improve storage reliability and parallel processing capabilities.

[0077] For example, different vector shards can be stored on different disk partitions, different nodes of a distributed storage system, or stored using different storage media. When searching, vector shards can be read from multiple locations at the same time, and similarity comparisons can be performed in parallel to improve search efficiency.

[0078] For example, in the agricultural insurance underwriting business scenario, suppose an insurance agent submits a photo of a damaged farmland as a target image to be checked for duplicates through a mobile terminal. The system first extracts features from the photo to obtain a 512-dimensional image feature vector. Then, the vector is segmented and quantized using a vector similarity search engine algorithm to obtain a target query vector. The target query vector is then divided into a preset number of slices, such as 4 slices. Assume that each slice is 128 dimensions, namely V 1 、V 2 、V 3 、V 4 . Compress each vector slice, for example, use PQ technology to find the quantization center of each slice, and approximately quantize the slice into a combination of quantization centers. Finally, each compressed vector slice is independently stored as a new feature vector sample. For example, V 1 Stored in disk partitions A, V 2 Stored in node B of the distributed storage system, V 3 Stored in a specific location in cloud storage C, V 4 The vector segments are stored in a specific folder D on the local disk. When new images need to be checked for duplicates, the system can quickly read the corresponding vector segments from various storage locations, decompress them, and compare their similarities. This not only improves the flexibility and reliability of storage, but also greatly improves the retrieval efficiency, ensuring the accuracy and timeliness of image duplicate checking in agricultural insurance underwriting business.

[0079] In some optional implementations, step S12 includes:

[0080] S125, segmenting the image feature vectors one by one, to obtain feature vector segments;

[0081] S126, product quantization is performed on each segment of the feature vector, and the quantization center closest to each segment of the feature vector is found from a preset quantization center set to obtain the quantized feature vector segment, and the quantized feature vector segment is used as the target query vector.

[0082] In this embodiment, by accurately segmenting and quantizing the image feature vector, the vector similarity search engine algorithm is used to convert it into a target query vector that is easier to store and retrieve, thereby providing strong support for efficient image duplication detection. In this embodiment, the strategy of segmenting the vectors one by one can fully consider the local characteristics of the image feature vector and achieve more refined processing. Compared with the traditional overall processing method, this segmentation method can better adapt to the feature differences in different regions and improve the expression ability of complex images. In addition, the product quantization process of this embodiment realizes efficient compression and approximate representation of high-dimensional vectors by finding the quantization center closest to each feature vector segment. This not only reduces the storage requirements and computing costs, but also provides the possibility for subsequent rapid retrieval. Therefore, this embodiment significantly improves the accuracy and efficiency of image duplication detection. Accurate vector representation and quantization make the identification of similar images more accurate and reduce the situation of misjudgment and missed judgment. At the same time, reducing storage requirements and computational complexity helps to speed up the duplication detection process and meet the high requirements for real-time and accuracy in agricultural insurance underwriting business.

[0083] For example, after obtaining the image feature vector, the vector segmentation operation is started one by one. Assume that the image feature vector is a high-dimensional vector, such as 512 dimensions. First, determine the segmentation strategy, which can adopt equal-length segmentation or unequal-length segmentation according to different requirements.

[0084] Equal-length segmentation: Divide the 512-dimensional vector into several segments, each with the same dimension. For example, divide it into 8 segments, each with 64 dimensions. For this method, you can use simple index partitioning, that is, partition according to fixed dimension intervals. For example, the first segment is the 1st to 64th elements of the vector, the second segment is the 65th to 128th elements, and so on.

[0085] Non-equal length segmentation: segment the image according to different characteristic regions. This requires a preliminary analysis of the image to determine the importance and characteristic differences of different regions. For example, for images of crop damage, the image can be divided into the main crop area, the background sky area, and the land area. Then different dimensions are assigned according to the importance of the characteristics of different regions. For example, the main crop area may be assigned more dimensions to better capture its detailed features. Assume that the main crop area is assigned 80 dimensions, the background sky area is assigned 40 dimensions, the land area is assigned 40 dimensions, and so on.

[0086] Determining the set of quantization centers is one of the key steps in product quantization. This set can be obtained by statistically analyzing and training a large number of image feature vectors in advance. Clustering algorithms, such as K-Means clustering, can be used to cluster a large number of vector samples into several clusters, and the center of each cluster is a quantization center. Assume that K quantization centers are obtained after clustering, and each quantization center is also a vector with the same dimension as the vector segment.

[0087] For each feature vector segment, use an appropriate distance metric, such as Euclidean distance, to calculate its distance from each quantization center in the quantization center set. The calculation formula is: Where A and B are the two vectors of the feature vector segment and the quantization center respectively. i and b i are the i-th elements of the above two vectors, and n is the dimension of the vector.

[0088] After finding the nearest quantization center, the feature vector segment is approximately represented as this quantization center. For example, if a segment of feature vector is closest to quantization center C, then the segment of feature vector is replaced by quantization center C. This can greatly reduce storage requirements while retaining the main features of the vector to a certain extent.

[0089] For example, in the agricultural insurance underwriting business, the insurance agent uploads a photo of damaged crops through the mobile terminal for duplication check. The system first extracts features from the photo and obtains a 512-dimensional image feature vector.

[0090] Assuming that a non-equal length segmentation strategy is adopted, the image is divided into the main crop area (80 dimensions), the background sky area (40 dimensions), and the land area (40 dimensions) according to the image analysis. For the 80-dimensional vector segment of the main crop area, calculate its distance to each quantization center in the quantization center set. After calculation, it is found that the distance to the quantization center D is the shortest, so the segment is approximately represented as the quantization center D. Similarly, the vector segments of the background sky area and the land area are processed similarly.

[0091] Finally, the quantized feature vector segments are obtained, and these segments are combined to form the target query vector. When the similarity is subsequently compared with the pre-stored historical vector sample set, the system can quickly find similar historical pictures. If a high degree of similarity with the historical picture is found, the system can issue an alarm in time to remind the underwriter to further verify whether there is a duplicate claim. In this way, through precise vector segmentation and quantification, the efficiency and accuracy of image duplication checking in agricultural insurance underwriting business are improved, providing a strong guarantee for the smooth progress of the business.

[0092] In some optional implementations, retrieving a pre-stored historical vector sample set similar to the target query vector by using the vector similarity search engine algorithm includes:

[0093] Using the inverted file index structure in the quantization-based vector retrieval algorithm in the vector similarity search engine algorithm to search, and finding the quantization center corresponding to the target query vector in the index;

[0094] According to the quantization center, a set of pre-stored historical vector samples similar to the target query vector is determined.

[0095] This embodiment uses the inverted file index structure of the quantization-based vector retrieval algorithm in the vector similarity search engine algorithm to quickly and accurately retrieve a set of historical vector samples similar to the target query vector to achieve efficient image duplication detection. Among them, this embodiment adopts the inverted file index structure in the quantization-based vector retrieval (IVF_PQ) algorithm, which can efficiently locate a set of vectors that may be similar to the target query vector, greatly reducing the time and computing cost of retrieval. Compared with traditional retrieval methods, this method is more intelligent and efficient, and can adapt to the processing requirements of large-scale data. Therefore, this embodiment reduces unnecessary calculations and speeds up the duplication detection process by quickly locating a set of similar vectors. At the same time, accurate retrieval results help to better identify duplicate images and ensure the accuracy and reliability of the business.

[0096] Specifically, we first need to understand the inverted file index structure in the IVF_PQ algorithm. The IVF_PQ algorithm combines the inverted file index (IVF) and product quantization (PQ) technology. In the inverted file index structure, the data is divided into multiple clusters, each with a center. When searching, for the target query vector, first calculate its quantized result. Then, find the corresponding cluster center in the index based on the quantization result. Since similar vectors are usually assigned to similar clusters, by finding the cluster center corresponding to the target query vector, the range that may contain similar vectors can be quickly determined. Specifically, let the target query vector be V=(v 1 , v 2 , …, vn ). After vector segmentation and quantization, the quantized target query vector is obtained. In the index, each cluster center has an identifier. By calculating the distance between the target query vector and each cluster center, the nearest cluster center (i.e., the quantized center mentioned in the context) can be found. This distance can be calculated using metrics such as Euclidean distance. Once the nearest cluster center is determined, the range of the pre-stored historical vector sample set similar to the target query vector can be determined.

[0097] For example, in the agricultural insurance underwriting business, when the insurance agent uploads a new farmland photo as the target image. The system first performs feature extraction, vector segmentation and quantization on the photo to obtain the target query vector. Then, the inverted file index structure of the IVF_PQ algorithm in the vector similarity search engine algorithm is used for retrieval. For example, suppose that a large number of feature vectors of farmland photos uploaded in the previous underwriting process are stored in the historical vector sample set. These vectors have been processed by the IVF_PQ algorithm and assigned to different clusters. When the system processes a new target query vector, the corresponding cluster center is found by calculation. If the target query vector is very close to a cluster center, it can be determined that the historical vector sample set that may contain similar vectors is within this cluster range. The underwriter can determine whether the newly uploaded photo is repeated with the historical photo based on the retrieval results. If a duplication is found, the system can issue an alarm in time to remind the underwriter to further verify, avoid the occurrence of repeated claims, and ensure the fairness and rationality of the agricultural insurance underwriting business.

[0098] In some optional implementations, step S14 includes:

[0099] S140, performing cosine similarity calculation on the quantized target query vector and each historical vector sample in the historical vector sample set to obtain a similarity result;

[0100] S141, comparing all the calculated similarity results with a preset similarity threshold to obtain a comparison result;

[0101] S142: If the comparison result shows that the similarity result is greater than the similarity threshold, it is determined that the target image is repeated with the corresponding historical image, and a duplicate alarm is triggered.

[0102] This embodiment performs an accurate similarity comparison between the quantized target query vector and the historical vector sample set, thereby accurately determining whether the target image is a duplicate of the historical image, thereby achieving efficient image duplication detection. Among them, this embodiment adopts cosine similarity calculation, which can effectively measure the similarity between vectors and has a good effect on the comparison of high-dimensional vectors; in addition, this embodiment sets a preset similarity threshold, so that the duplication detection result is more objective and adjustable, meeting the needs of different business scenarios. Therefore, this embodiment can accurately identify duplicate images through accurate similarity calculation and threshold comparison, reducing misjudgment and missed judgment. At the same time, triggering the duplication detection alarm function can promptly remind relevant personnel to handle it, thereby improving the efficiency and security of business processing.

[0103] Specifically, the cosine similarity is calculated first. Let the target query vector be Q = (q 1 ,q 2 ,…,q n ), a historical vector sample in the historical vector sample set is H = (h 1 ,h 2 ,…,h n ), then their cosine similarity calculation formula is This calculation process will calculate the quantized target query vector and each historical vector sample in the historical vector sample set one by one to obtain multiple similarity results. Then, all the calculated similarity results are compared with the preset similarity threshold. The similarity threshold is a value set according to specific business needs and experience, for example, it can be set to 0.8. If a similarity result is greater than this threshold, it is considered that the corresponding target image has a high similarity with the historical image. Finally, if the comparison result is that the similarity result is greater than the similarity threshold, it is determined that the target image is repeated with the corresponding historical image, and a duplicate check alarm is triggered. Alarms can be issued in a variety of ways, such as sending emails, popping up prompt windows, etc., so that relevant personnel can handle them in time.

[0104] For example, in the agricultural insurance underwriting business, when the insurance agent uploads a new crop photo as the target image. The system first performs feature extraction, vector segmentation and quantization on the photo to obtain the target query vector. Then, the cosine similarity between this target query vector and the historical vector sample in the historical vector sample set is calculated. Assume that the historical vector sample set stores a large number of feature vectors of crop photos uploaded during the previous underwriting process. After calculation, if the cosine similarity between the target query vector of the target image and the historical vector sample of a historical image is greater than the preset similarity threshold, such as 0.8. The system will determine that the two images are duplicated and trigger a duplicate check alarm. After receiving the alarm, the underwriter can further verify whether there is a duplicate claim. For example, if it is found that the target image and the historical image are photos of the same farmland at different times, but a claim has been made, the underwriter can refuse the duplicate claim to ensure the fairness and rationality of the agricultural insurance business.

[0105] In some optional implementations, the step S11 includes:

[0106] The pre-trained ResNet50 model is used to extract features of the target image to obtain an image feature vector.

[0107] This embodiment uses the powerful feature extraction capability of the pre-trained ResNet50 model to process the target image, obtain a feature vector that can accurately characterize the content of the image, and provide high-quality input data for subsequent image duplication checking. Among them, ResNet50 has been pre-trained with a large amount of data and has good generalization ability and adaptability to different types of images. Compared with other possible feature extraction methods, it can extract feature vectors with more discriminativeness and robustness. Therefore, the feature vector extracted by ResNet50 in this embodiment can better capture the essential characteristics of the image, making similar images closer in the feature space, so that they are more easily accurately identified as duplicate images.

[0108] Specifically, ResNet50 is a deep residual network composed of multiple residual blocks. Each residual block contains a convolution layer, a batch normalization layer, and an activation function layer. When extracting features from a target image, the target image is first adjusted to a size suitable for ResNet50 input, usually 224x224 pixels. Then, the image enters the ResNet50 network. In the network, the image passes through a series of convolutional layers and pooling layers to gradually extract features from low to high levels. Low-level features mainly include information such as the edge and texture of the image. As the network level deepens, more abstract high-level features such as the shape and structure of the object are gradually extracted. Finally, the features are mapped to a vector of fixed dimension, namely the image feature vector, through a fully connected layer. For example, suppose the target image is a photo of crops damaged by disasters. After being processed by ResNet50, the shape of the crops in the image, the features of the damaged parts, etc. are extracted and converted into a 512-dimensional vector, which can accurately represent the unique features of this image.

[0109] For example, in the agricultural insurance underwriting business, the insurance agent uploads a photo of farmland as the target image through the mobile terminal. The system first uses the pre-trained ResNet50 model to extract features from the photo. If the photo is of a farmland soaked by floods, ResNet50 will extract features such as the shape of the farmland, the distribution of floods, and the state of crops, and convert them into image feature vectors. This feature vector will be used as input for the subsequent duplication check step. In the underwriting stage, the system compares this feature vector with the historical vector sample set. If there are similar photos of farmland soaked by floods in the historical photos, and their feature vectors are very close in the feature space, then the system can determine that the newly uploaded photo may be a duplicate of the historical photo. The underwriter can conduct further verification based on the duplication check results to ensure the accuracy and fairness of agricultural insurance claims and avoid duplicate claims.

[0110] It is understandable that the input target image may be standardized in advance, including resizing, color normalization, etc., to ensure uniformity of feature extraction.

[0111] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0112] The basic technologies of artificial intelligence generally include sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics and other technologies. Artificial intelligence software technology mainly includes computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning. This application can be applied in the field of financial technology.

[0113] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through computer-readable instructions, and the computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0114] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0115] Further references Figure 3 , as a response to the above Figure 2 The present application provides an embodiment of a device for checking duplicate images based on a vector similarity search engine. Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0116] like Figure 3 As shown, the image duplicate checking device 400 based on the vector similarity search engine described in this embodiment includes: an acquisition module 401, a feature extraction module 402, an indexing module 403, a retrieval module and a comparison module 405. Among them:

[0117] Acquisition module 401, used to acquire the target image to be checked for duplication;

[0118] A feature extraction module 402 is used to extract features from the target image to obtain an image feature vector;

[0119] An indexing module 403 is used to segment and quantize the image feature vector by a vector similarity search engine algorithm to obtain a target query vector;

[0120] A retrieval module 404 is used to retrieve a set of historical vector samples that are similar to the target query vector and are pre-stored through the vector similarity search engine algorithm; each historical vector sample in the set of historical vector samples is pre-stored after being indexed by a picture feature vector extracted from a historical picture through the vector similarity search engine algorithm;

[0121] The comparison module 405 is used to compare the quantized target query vector with each historical vector sample in the historical vector sample set to obtain a duplicate checking result of the target image.

[0122] The embodiment of the present application obtains the target image to be checked for duplicates, performs feature extraction on it to obtain an image feature vector, then uses a vector similarity search engine algorithm to segment and quantize the vector to obtain a target query vector, and then uses the vector similarity search engine algorithm to retrieve vectors similar to the target query vector in a pre-stored historical vector sample set that is indexed by feature vectors extracted from historical images, and finally compares the target query vector with each vector in the historical vector sample set for similarity, thereby determining the duplicate check result of the target image. The embodiment of the present application uses an advanced vector similarity search engine algorithm for vector processing and retrieval, which can effectively improve the stability of the image duplicate check system in the agricultural insurance business, avoid frequent alarms and restart services, and improve timeliness, meet the needs of rapid response to a large number of image duplicate check requests, and ensure the smooth progress of business processes such as agricultural insurance underwriting and claims.

[0123] In some optional implementations of this embodiment, the image duplication checking device 400 based on a vector similarity search engine further includes a storage module, which is used to store the target query vector as a new feature vector sample through a vector similarity search engine algorithm.

[0124] In some optional implementations of this embodiment, the storage module includes:

[0125] A slicing submodule, used to divide each quantized target query vector into a preset number of slicings to obtain vector slicings;

[0126] The compression submodule is used to compress each of the vector slices and independently store each compressed vector slice as a new feature vector sample.

[0127] In some optional implementations of this embodiment, the index module includes:

[0128] A segmentation submodule, used to segment the image feature vectors one by one to obtain feature vector segments;

[0129] The quantization submodule is used to perform product quantization on each feature vector segment, find the quantization center closest to each feature vector segment from a preset quantization center set, obtain the quantized feature vector segment, and use the quantized feature vector segment as the target query vector.

[0130] In some optional implementations of this embodiment, the retrieval module is specifically used to:

[0131] Using the inverted file index structure in the IVF_PQ algorithm in the vector similarity search engine algorithm to search, find the quantization center corresponding to the target query vector in the index;

[0132] According to the quantization center, a set of pre-stored historical vector samples similar to the target query vector is determined.

[0133] In some optional implementations of this embodiment, the comparison module includes:

[0134] A similarity calculation submodule is used to perform cosine similarity calculation on the quantized target query vector and each historical vector sample in the historical vector sample set to obtain a similarity result;

[0135] A comparison submodule, used for comparing all the similarity results calculated with a preset similarity threshold to obtain a comparison result;

[0136] The judgment submodule is used to determine that the target image is repeated with the corresponding historical image and trigger a duplicate alarm if the comparison result is that the similarity result is greater than the similarity threshold.

[0137] In some optional implementations of this embodiment, the feature extraction module is specifically used to:

[0138] The pre-trained ResNet50 model is used to extract features of the target image to obtain an image feature vector.

[0139] To solve the above technical problems, the present application also provides a computer device. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0140] The computer device 6 includes a memory 61, a processor 62, and a network interface 63 that are interconnected and communicated through a system bus. It should be noted that the figure only shows a computer device 6 with a memory 61, a processor 62, and a network interface 63, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (Application Specific Integrated Circuit, ASIC), programmable gate arrays (Field-Programmable Gate Array, FPGA), digital processors (Digital Signal Processor, DSP), embedded devices, etc.

[0141] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.

[0142] The memory 61 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 61 can be an internal storage unit of the computer device 6, such as a hard disk or memory of the computer device 6. In other embodiments, the memory 61 can also be an external storage device of the computer device 6, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (FlashCard), etc. equipped on the computer device 6. Of course, the memory 61 can also include both the internal storage unit of the computer device 6 and its external storage device. In this embodiment, the memory 61 is generally used to store the operating system and various application software installed on the computer device 6, such as computer-readable instructions of a method for checking duplicate images based on a vector similarity search engine, etc. In addition, the memory 61 can also be used to temporarily store various types of data that have been output or are to be output.

[0143] The processor 62 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips in some embodiments. The processor 62 is generally used to control the overall operation of the computer device 6. In this embodiment, the processor 62 is used to run computer-readable instructions or process data stored in the memory 61, such as computer-readable instructions for running the image duplication detection method based on the vector similarity search engine.

[0144] The network interface 63 may include a wireless network interface or a wired network interface. The network interface 63 is generally used to establish a communication connection between the computer device 6 and other electronic devices.

[0145] The embodiment of the present application obtains the target image to be checked for duplicates, performs feature extraction on it to obtain an image feature vector, then uses a vector similarity search engine algorithm to segment and quantize the vector to obtain a target query vector, and then uses the vector similarity search engine algorithm to retrieve vectors similar to the target query vector in a pre-stored historical vector sample set that is indexed by feature vectors extracted from historical images, and finally compares the target query vector with each vector in the historical vector sample set for similarity, thereby determining the duplicate check result of the target image. The embodiment of the present application uses an advanced vector similarity search engine algorithm for vector processing and retrieval, which can effectively improve the stability of the image duplicate check system in the agricultural insurance business, avoid frequent alarms and restart services, and improve timeliness, meet the needs of rapid response to a large number of image duplicate check requests, and ensure the smooth progress of business processes such as agricultural insurance underwriting and claims.

[0146] The present application also provides another implementation, namely, providing a computer-readable storage medium, which stores computer-readable instructions, and the computer-readable instructions can be executed by at least one processor to enable the at least one processor to perform the steps of the above-mentioned image duplication checking method based on the vector similarity search engine.

[0147] The embodiment of the present application obtains the target image to be checked for duplicates, performs feature extraction on it to obtain an image feature vector, then uses a vector similarity search engine algorithm to segment and quantize the vector to obtain a target query vector, and then uses the vector similarity search engine algorithm to retrieve vectors similar to the target query vector in a pre-stored historical vector sample set that is indexed by feature vectors extracted from historical images, and finally compares the target query vector with each vector in the historical vector sample set for similarity, thereby determining the duplicate check result of the target image. The embodiment of the present application uses an advanced vector similarity search engine algorithm for vector processing and retrieval, which can effectively improve the stability of the image duplicate check system in the agricultural insurance business, avoid frequent alarms and restart services, and improve timeliness, meet the needs of rapid response to a large number of image duplicate check requests, and ensure the smooth progress of business processes such as agricultural insurance underwriting and claims.

[0148] The non-Company software tools or components appearing in the embodiments of this application are merely examples and do not represent actual use.

[0149] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0150] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. A method for checking duplicate images based on a vector similarity search engine, characterized in that: The steps include: Get the target image to be checked for duplicates; Extracting features of the target image to obtain an image feature vector; The image feature vector is segmented and quantized by a vector similarity search engine algorithm to obtain a target query vector; Retrieving a pre-stored historical vector sample set similar to the target query vector through the vector similarity search engine algorithm; each historical vector sample in the historical vector sample set is pre-stored after being indexed by a picture feature vector extracted from a historical picture through the vector similarity search engine algorithm; The quantized target query vector is compared with each historical vector sample in the historical vector sample set for similarity to obtain a duplicate checking result of the target image.

2. The method for checking duplicate images based on a vector similarity search engine according to claim 1, characterized in that: After the image feature vector is segmented and quantized by the vector similarity search engine algorithm to obtain a target query vector, and before the vector similarity search engine algorithm is used to retrieve a set of pre-stored historical vector samples that are similar to the target query vector, the method further includes: The target query vector is stored as a new feature vector sample through a vector similarity search engine algorithm.

3. The method for checking duplicate images based on a vector similarity search engine according to claim 2, characterized in that: The storing the target query vector as a new feature vector sample by using a vector similarity search engine algorithm includes: Dividing each quantized target query vector into a preset number of slices to obtain vector slices; Each of the vector slices is compressed, and each compressed vector slice is independently stored as a new feature vector sample.

4. The method for checking duplicate images based on a vector similarity search engine according to claim 1, characterized in that: The step of segmenting and quantizing the image feature vector by a vector similarity search engine algorithm to obtain a target query vector includes: Segmenting the image feature vectors one by one to obtain feature vector segments; Product quantization is performed on each of the feature vector segments, and the quantization center closest to each of the feature vector segments is found from a preset quantization center set to obtain the quantized feature vector segment, and the quantized feature vector segment is used as the target query vector.

5. The method for checking duplicate images based on a vector similarity search engine according to claim 4, characterized in that: The retrieving a pre-stored historical vector sample set similar to the target query vector by using the vector similarity search engine algorithm includes: Using the inverted file index structure in the quantization-based vector retrieval algorithm in the vector similarity search engine algorithm to search, and finding the quantization center corresponding to the target query vector in the index; According to the quantization center, a set of pre-stored historical vector samples similar to the target query vector is determined.

6. The method for checking duplicate images based on a vector similarity search engine according to claim 1, characterized in that: The step of comparing the quantized target query vector with each historical vector sample in the historical vector sample set to obtain a duplicate checking result of the target image includes: Performing cosine similarity calculation on the quantized target query vector and each historical vector sample in the historical vector sample set to obtain a similarity result; Compare all the calculated similarity results with a preset similarity threshold to obtain a comparison result; If the comparison result is that the similarity result is greater than the similarity threshold, it is determined that the target image is repeated with the corresponding historical image, and a duplicate alarm is triggered.

7. The method for checking duplicate images based on a vector similarity search engine according to claim 1, characterized in that: The step of extracting features from the target image to obtain an image feature vector includes: The pre-trained ResNet50 model is used to extract features of the target image to obtain an image feature vector.

8. An image duplication checking device based on a vector similarity search engine, characterized in that: include: The acquisition module is used to obtain the target image to be checked for duplicates; A feature extraction module is used to extract features from the target image to obtain a feature vector of the image; An indexing module, used for segmenting and quantizing the image feature vector by a vector similarity search engine algorithm to obtain a target query vector; A retrieval module is used to retrieve a set of pre-stored historical vector samples that are similar to the target query vector through the vector similarity search engine algorithm; each historical vector sample in the set of historical vector samples is pre-stored after being indexed by a picture feature vector extracted from a historical picture through the vector similarity search engine algorithm; The comparison module is used to compare the similarity between the quantized target query vector and each historical vector sample in the historical vector sample set to obtain the duplicate checking result of the target image.

9. A computer device, characterized in that: It includes a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the steps of the image duplication checking method based on a vector similarity search engine as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the steps of the image duplication checking method based on a vector similarity search engine as described in any one of claims 1 to 7 are implemented.

Images