CN110865992A - A retrieval library management method, retrieval method, device and medium - Google Patents

Abstract

The application discloses a search library management method, a search method, a device and a medium, wherein the search library management method comprises the following steps: acquiring attribute parameters of each thermal data under the condition that the total number of data in the thermal search library is greater than a first preset scale threshold; migrating the hot data with the attribute parameters meeting the hot data management conditions to a cold search library; calculating first matching degrees of each cold data and each hot data stored in a cold search library; and migrating cold data meeting cold data management conditions to a hot search library. According to the method and the device, the cold data and the hot data of the search library are separated, and the scale of the hot search library is ensured to be within a threshold range by triggering the cold and hot library upgrading and degrading logic, so that the effectiveness of the search data is improved, and the search reliability, accuracy and search efficiency are improved.

Description

A retrieval library management method, retrieval method, device and medium

technical field

The present application relates to the field of computer technology, and in particular, to a retrieval library management method, retrieval method, device and medium.

Background technique

At present, the application of artificial intelligence technology for identification and retrieval has been implemented and promoted in smart retail, smart community, smart finance, public security security and other application scenarios.

In the above application scenarios, in order to improve the retrieval efficiency, the retrieval library for identification is generally stored in a processor with better concurrency capability and better memory access speed. Taking a face retrieval scene as an example, the face database is usually stored in the video memory of an image processor (Graphics Processing Unit, GPU). However, in these application scenarios, with the addition of a large number of new identity users, it will increase the consumption of a large number of excellent processors and expensive (such as GPU) resources, greatly increasing the storage cost of retrieval. In addition, the retrieval reliability, accuracy and retrieval efficiency of the existing retrieval methods need to be further improved.

SUMMARY OF THE INVENTION

The present application provides a retrieval library management method, retrieval method, device and medium to solve at least one of the above technical problems.

In one aspect, the present application provides a retrieval database management method, the retrieval database includes a hot retrieval database and a cold retrieval database, and the method includes:

In the case that the total number of hot data stored in the hot search library is greater than a first preset scale threshold, acquiring attribute parameters of each hot data stored in the hot search library;

If it is determined that the attribute parameter of the hot data satisfies the hot data management condition, the hot data satisfying the hot data management condition is migrated to the cold retrieval library;

calculating a first degree of matching between each cold data stored in the cold retrieval library and each hot data in the hot retrieval library;

If it is determined that the first matching degree satisfies the cold data management condition, the cold data satisfying the cold data management condition is migrated to the hot retrieval library.

On the other hand, the present application also provides a retrieval method, which uses a retrieval database managed by data for retrieval, the retrieval database includes a hot retrieval database and a cold retrieval database, and the method includes:

Obtain the feature data to be retrieved of the object;

Retrieve the feature data to be retrieved in the hot retrieval database managed by the retrieval database to obtain retrieval results;

return the search result;

Wherein, the retrieval library performs data management through the following retrieval library management method, and the database management method includes:

In the case where the total number of hot data stored in the hot search library is greater than a first preset scale threshold, acquiring attribute parameters of each hot data in the hot search library;

If it is determined that the attribute parameter of the hot data satisfies the hot data management condition, the hot data satisfying the hot data management condition is migrated to the cold retrieval library;

calculating a first degree of matching between each cold data stored in the cold retrieval library and each hot data in the hot retrieval library;

If it is determined that the first matching degree satisfies the cold data management condition, the cold data satisfying the cold data management condition is migrated to the hot retrieval library.

On the other hand, the present application also provides a retrieval database management device, the retrieval database includes a hot retrieval database and a cold retrieval database, and the device includes:

an attribute acquisition module, configured to acquire attribute parameters of each hot data stored in the hot search library when the total number of hot data stored in the hot search library is greater than a first preset scale threshold;

a first migration module, configured to migrate the hot data satisfying the hot data management condition to the cold retrieval library if the attribute parameter of the hot data is determined to satisfy the hot data management condition;

a first calculation module, configured to calculate a first degree of matching between each cold data stored in the cold retrieval library and each hot data in the hot retrieval library;

The second migration module is configured to migrate the cold data satisfying the cold data management condition to the hot retrieval library if it is determined that the first matching degree satisfies the cold data management condition.

On the other hand, the present application also provides a retrieval device, the device includes:

The feature management module is used to obtain the feature data to be retrieved of the object;

The above-mentioned retrieval library management device is used for data management of the retrieval library, and the retrieval library includes a hot retrieval library and a cold retrieval library;

a retrieval module, used for retrieving the feature data to be retrieved in the hot retrieval database managed by data to obtain retrieval results;

The return module is used to return the retrieval result.

Another aspect also provides a retrieval library management device, the device includes a processor and a memory, the memory stores at least one instruction, at least a piece of program, a code set or an instruction set, the at least one instruction, the at least one A piece of program, the code set or the instruction set is loaded and executed by the processor to implement any one of the retrieval library management methods described above.

Another aspect also provides a retrieval device, the device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, the at least one instruction, the at least one program , the code set or the instruction set is loaded and executed by the processor to implement any one of the retrieval methods described above.

Another aspect also provides a computer-readable storage medium, where at least one instruction, at least one piece of program, code set or instruction set is stored in the storage medium, wherein the at least one instruction, at least one piece of program, code set or instruction set is composed of The processor loads and executes the retrieval library management method as described above.

Another aspect also provides a computer-readable storage medium, where at least one instruction, at least one piece of program, code set or instruction set is stored in the storage medium, wherein the at least one instruction, at least one piece of program, code set or instruction set is composed of The processor loads and executes the retrieval method as described in any of the above.

A retrieval library management method, retrieval method, device and medium provided by this application have the following technical effects:

The retrieval database in the embodiment of the present application includes a hot retrieval database and a cold retrieval database, and when the total number of hot data stored in the hot retrieval database is greater than a first preset scale threshold, obtain the data stored in the hot retrieval database. attribute parameters of each hot data; if it is determined that the attribute parameters of the hot data meet the hot data management conditions, migrate the hot data that meets the hot data management conditions to the cold retrieval library; calculate the respective cold data stored in the cold retrieval library The first matching degree with each hot data in the hot retrieval library; if it is determined that the first matching degree satisfies the cold data management condition, the cold data satisfying the cold data management condition is migrated to the hot retrieval library. Therefore, by separating the hot and cold data of the retrieval library, when the total number of hot data stored in the hot retrieval library is greater than the first preset scale threshold, the upgrading and upgrading logic of the hot and cold library is triggered, so as to ensure the scale of the hot retrieval library Within the threshold range, the validity of the hot data in the hot search library is improved, and the retrieval reliability, accuracy and search efficiency of the managed hot search library are improved.

Description of drawings

In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present application or in the prior art, the following briefly introduces the accompanying drawings used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description The drawings are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings without any creative effort.

1 is a schematic diagram of an implementation environment provided by an embodiment of the present application;

2 is a schematic flowchart of a retrieval library management method provided by an embodiment of the present application;

3 is a schematic flowchart of a step of migrating hot data to a cold retrieval library provided by an embodiment of the present application;

4 is a schematic flowchart for migrating hot data to a cold retrieval library provided by an embodiment of the present application;

5 is a schematic flowchart of a step of performing data management on a hot search library provided by an embodiment of the present application;

6 is a schematic flowchart of a warehouse clearing step of a cold retrieval warehouse provided by an embodiment of the present application;

7 is a schematic flowchart of a step of performing data management on a cold retrieval database provided by an embodiment of the present application;

8 is a structural block diagram of a retrieval library management apparatus provided by an embodiment of the present application;

9 is a schematic flowchart of a retrieval method provided by an embodiment of the present application;

10 is a schematic diagram of a framework for implementing a retrieval method based on a face retrieval scenario in the present application;

11 is a structural block diagram of a retrieval apparatus provided by an embodiment of the present application;

FIG. 12 is a schematic diagram of the hardware structure of a device provided by the present application for implementing the method provided by the embodiment of the present application.

Detailed ways

Machine Learning (ML) is a multi-field interdisciplinary subject involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It specializes in how computers simulate or realize human learning behaviors to acquire new knowledge or skills, and to reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent, and its applications are in all fields of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, teaching learning and other technologies.

In order to make those skilled in the art better understand the solutions of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only The embodiments are part of the present application, but not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the scope of protection of the present application.

First, briefly describe the technical terms that may be involved in the present invention:

Face recognition: refers to the process of extracting face features and performing similarity comparison. According to different application scenarios, face recognition is mainly divided into 1:1 face verification (Face Verification), 1:N face recognition (Face Recognition) and face retrieval (Face Retrieval) three categories.

1:N face recognition: Find one or more faces with the highest similarity to the face to be retrieved in a large-scale face database.

SDK: A software development kit (Software Development Kit, SDK) is generally a collection of development tools used by software engineers to build application software for specific software packages, software frameworks, hardware platforms, business systems, and the like.

CPU/GPU: The CPU is the central processing unit, and the GPU is the image processor; both are computing resources for video multimedia.

At present, identity recognition and retrieval have been implemented and promoted in smart retail, smart community, smart finance, public security security and other application scenarios. In these application scenarios, in order to improve the retrieval efficiency, the retrieval library for identification is generally stored in a processor with better concurrency capability and better memory access speed. Existing retrieval methods are mainly suitable for application scenarios with fixed retrieval bases, such as scenarios where retrieval bases are registered in advance (such as access control, attendance, etc.); and for scenarios where retrieval bases are not fixed (such as smart retail, smart communities, etc.) , with the addition of a large number of users with new identities, the scale of the retrieval base continues to increase, which brings huge challenges to the existing retrieval methods.

Taking a face retrieval scene as an example, the face database is usually stored in the video memory of an image processor (Graphics Processing Unit, GPU). However, in these application scenarios, with a large number of new users joining, it will consume a lot of excellent processor (such as GPU) resources and greatly increase the storage cost of retrieval. In addition, the retrieval reliability, accuracy and retrieval efficiency of the existing retrieval methods need to be further improved.

In addition, the inventor also found that: in the process of identity recognition, a retrieval algorithm is usually used to perform identity retrieval in the retrieval database, however, the retrieval accuracy based on the retrieval algorithm depends on the size of the retrieval database. If the scale of the retrieval database is larger, the model accuracy and computing performance of the retrieval algorithm are required to be higher. For example, taking the face retrieval scenario as an example, as the number of user identities in the retrieval database increases, the accuracy of general face recognition algorithms will drop significantly. As shown in Table 1 below, it is the same retrieval algorithm model, under the one-thousandth error rate, the first hit rate of faces of different retrieval database scales.

Table 1 Relation between retrieval accuracy and retrieval database size

Search library size/number of people 100 500 1000 Accuracy/% 100 99.2 78.4

In order to solve at least one of the above technical problems, the present application provides a retrieval library management method, retrieval method, device and medium.

In order to make the objectives, technical solutions and advantages of the present application clearer, the embodiments of the present application will be further described in detail below with reference to the accompanying drawings.

Please refer to FIG. 1 , which shows a schematic diagram of an implementation environment provided by an embodiment of the present application. The implementation environment may include: a terminal 10 and a server 20 connected to the terminal 10 through a network.

The terminal 10 may specifically include software running in the physical device, such as applications installed on the device, etc., and may also include types such as smart phones, desktop computers, tablet computers, notebook computers, digital assistants, and smart wearable devices installed with applications. at least one of the physical devices. Specifically, an operating system runs on the terminal 10, and the operating system may be a Windows (Windows) operating system, a Linux operating system, a desktop operating system such as a Mac OS (Apple desktop operating system), or an iOS (Apple mobile terminal operating system) ) or a mobile operating system such as the Android operating system.

The server 20 may be an independent server, or a server cluster composed of multiple independent servers, or a distributed server, or a cloud server that provides basic cloud computing services such as cloud computing servers, cloud databases, and cloud storage. The distributed server may specifically be a block chain (Block Chain) structure, and any node in the block chain structure can execute or participate in executing the retrieval library management method or retrieval method.

It should be understood that the implementation environment shown in FIG. 1 is only an application environment of the solution of the present application, and does not constitute a limitation to the application environment of the solution of the present application. Other application environments may also include more or less than those shown in the figure. computer equipment, or the network connection relationship of computer equipment.

A specific embodiment of a retrieval database management method of the present application is introduced below. FIG. 2 is a schematic flowchart of a retrieval database management method provided by an embodiment of the present application. The present application provides the operation steps of the method as described in the embodiment or the flowchart. , but based on routine or non-creative work may include more or fewer operating steps. The sequence of steps enumerated in the embodiments is only one of the execution sequences of many steps, and does not represent the only execution sequence. As shown in FIG. 2 , the execution body of the method can be a server in the above application environment, the retrieval library includes a hot retrieval library and a cold retrieval library, and the method can include:

S201: In the case that the total number of hot data stored in the hot search library is greater than a first preset scale threshold, acquire attribute parameters of each hot data stored in the hot search library.

In this embodiment, the hot search library is a search base library used for searching. A large amount of hot data for retrieval and comparison stored in the hot retrieval library includes, but is not limited to, the user's identity and basic data of the user. The basic data of the user includes object data, such as a face image, a face feature image, a voiceprint, an iris, and the like.

As an example, taking the scene of face detection as an example, if user A enters a smart retail mall, the collection device collects the face image of user A, and then the hot search library obtains the face image of user A, and based on the Whether the feature of the face image has a corresponding face image feature in the hot search database, if it exists, it means that the user A has visited before; if it does not exist, it means that the user A is a new visiting user, then it needs to The user registers an identification ID, and associates the registered identification ID with the corresponding face image feature as newly added hot data, and stores them in the hot retrieval database.

The cold retrieval library is used to store cold data to be managed. Manage the cold data in the cold retrieval library and the hot data in the hot retrieval library under the pre-set management conditions, including but not limited to data migration, data deletion, data merging, and data transfer to disk , data synchronization, etc.

In smart retail, smart community, smart finance, public security security and other application scenarios, there will be a large number of users visiting constantly. To identify the users who visit constantly and in large numbers, the retrieval algorithm is usually used to search in the database. As the inventors discovered, the retrieval algorithm is related to the size of the database. Therefore, in order to improve the retrieval accuracy, when the retrieval base exceeds the first preset size threshold, the number of retrieval bases has exceeded the computing power of the retrieval algorithm, and there is a problem that the accuracy of retrieval results is low. Retrieve the management policy of the library. The first preset scale threshold is adaptively adjusted according to the theoretical accuracy rate of the retrieval algorithm, the theoretical accuracy of the retrieval model, and the theoretical computing capability.

When the total number of hot data stored in the hot search library is greater than the first preset scale threshold, it represents a triggering condition for the management of the hot search library. At this time, it is necessary to obtain the attribute parameters of each hot data stored in the hot retrieval library, so as to facilitate the screening of hot data that needs to be managed. The attribute parameter of the hot data includes at least one of a recording time of the hot data and a user activity associated with the hot data. For example, the recorded time may be the visit time of the user associated with the recorded hot data, and the activity may be based on the number of times the user visits within a preset time period (eg, one week, one month, half a year, or one year, etc.).

It should be noted that the value of the first preset scale threshold is not fixed, and can be adaptively configured according to the retrieval algorithm, so as to match different retrieval algorithms and improve retrieval performance.

In practical applications, taking the scene of face detection as an example, if the specified database size corresponding to the retrieval accuracy requirement of the face retrieval algorithm is 10,000, the first preset size threshold can be configured to be 10,000. After that, when it is detected that the total number of hot data stored in the current hot search library is greater than the first preset scale threshold, database management of the hot search library is triggered. During database management, the attribute parameters of each hot data stored in the hot search database (such as the storage time, Activeness), and then manage the hot data that meets the predetermined conditions based on this attribute parameter, so as to avoid affecting the normal face feature addition, deletion, modification, and query business during working hours, and make full use of machine night resources.

In other embodiments, when it is determined that the total number of hot data stored in the hot search library is less than the first preset scale threshold k1, the deep data management of the hot search library may not be triggered first; Whether the total number of stored hot data is greater than the third preset scale threshold k3 (k3 is less than k1), if the judgment result is yes, then trigger the data management of the hot search database again. Therefore, by reducing the total number of data in the hot search library, the search efficiency and accuracy rate based on the hot search library are further improved.

S203: If it is determined that the attribute parameter of the hot data meets the hot data management condition, migrate the hot data that meets the hot data management condition to the cold retrieval library.

After triggering the data management of the hot search library and obtaining the attribute parameters of each hot data in the hot search library, it is necessary to filter out the hot data that needs to be managed from all the hot data, and then perform corresponding data on the filtered hot data. manage.

The attribute parameter of the hot data The attribute parameter of the hot data includes at least one of a timestamp of the last recording of the hot data and a user activity associated with the hot data.

In this embodiment, as shown in FIGS. 3 and 4 , the attribute parameters of the hot data include a timestamp of the latest recording of the hot data and a user activity associated with the hot data. If it is determined that the attribute parameter of the hot data meets the hot data management condition, the hot data that meets the hot data management condition is migrated to the cold retrieval library, including:

S301: If it is determined that the time interval between the last recorded timestamp of the hot data and the first preset time is greater than or equal to a first preset threshold, and it is determined that the user activity associated with the hot data is less than a second preset threshold, determine The attribute parameter of the thermal data satisfies the thermal data management condition.

The last recorded timestamp is the time recorded by the last visit of the user identity associated with the hot data. The first preset time includes, but is not limited to, the current management time, some other fixed time, and the like. The first preset time may be configured according to the user's visit situation in an actual application scenario. For example, in a scenario with frequent visits, such as in a smart community scenario, since the user needs to enter and leave the community frequently, the first preset time can be set to a smaller value, such as half a month, a month, quarterly or Wait for half a year. In scenarios with infrequent visit times, such as in smart retail scenarios, since the user occasionally enters the retail mall, the first preset time can be set to a larger value, such as quarterly, half-yearly, one-year, etc.

The object activity is an activity index representing the user corresponding to the hot data in the second preset time period. Specifically, the object activity degree may be determined according to the number of times the user is active in the preset time period and the second preset time period. The active number of times may be the cumulative number of visits, the cumulative number of uses, and the like. The second preset time and the number of active times may be configured according to user visits in an actual application scenario. For example, in a scenario with frequent visits, such as in a smart community scenario, since the user needs to enter and leave the community frequently, the second preset time can be configured to be a smaller value and the number of activities to be a larger value, such as configuring the second preset time. The preset time is nearly three days, nearly a week, nearly half a month, etc.; the corresponding active times can be determined according to the situation, for example, it can be set to 5 to 20 times. In scenarios with infrequent visits, such as in smart retail scenarios, since the user occasionally enters the retail mall, the second preset time can be set to a larger value, such as nearly half a month, nearly a month, etc.; correspondingly The number of active times can be determined according to the situation, for example, it can be set to 3 to 8 times.

For example, in a smart retail scenario, within a preset management period, the last recorded time stamp of the hot data base corresponding to the user ID of user B in the hot data base is t1. In the past month, the cumulative number of times user B visited the smart retail mall was found to be m times. If the interval between t1 and the current query time is greater than the first preset threshold (eg, one year), and m is less than the second preset threshold (eg, 5 times), it is determined that the attribute parameter of the hot data corresponding to user B satisfies the hot data management condition.

In other embodiments, the attribute parameter of the hot data may only include a timestamp of the latest recording of the hot data or a user activity associated with the hot data. For example, in a security scenario, the attribute parameter of the thermal data may only include the timestamp of the latest recording.

The hot retrieval library of the present application performs the identity screening corresponding to the hot data through the last recorded time stamp of the hot data and/or the activity of the associated user, which ensures the timeliness and reliability of the retrieval results.

S303: Determine the hot data that meets the hot data management condition as the hot data to be managed.

S305: Migrate the hot data to be managed to the cold retrieval library.

When the preset management period is reached, the hot data to be managed is migrated to the cold retrieval library. The preset management period includes, but is not limited to, one day, one week, one month, and the like.

In one embodiment, the migrating the hot data to be managed to the cold retrieval library includes:

S3051: Calculate a second degree of matching between the hot data to be managed and each cold data in the cold retrieval library.

The second matching degree is used to represent the degree of similarity between the hot data to be managed and the cold data in the cold retrieval library. The number of the second matching degrees is multiple, which can be determined by calculating the similarity between the hot data to be managed and all the cold data respectively. The similarity includes, but is not limited to, cosine similarity.

S3053: Among the multiple second matching degrees corresponding to the thermal data to be managed, determine whether the maximum second matching degree is greater than or equal to a third preset threshold.

S3055: If the determination result is yes, perform data merging of the hot data to be managed and the cold data whose maximum second matching degree is greater than or equal to the third preset threshold;

S3057: If the determination result is no, store the hot data to be managed whose maximum second matching degree is less than the third preset threshold into the cold retrieval library.

Wherein, the third preset threshold is a preset value, and its value can be configured but not limited to be any value from 0.8 to 1. Data merging refers to merging the hot data to be managed into the corresponding cold data storage locations. Among the plurality of second matching degrees corresponding to the hot data to be managed, if it is determined that the maximum second matching degree is greater than or equal to the third preset threshold, it indicates that the cold retrieval library has already stored the matching degree corresponding to the maximum second matching degree. For cold data that is similar to the hot data to be managed, the corresponding hot data to be managed and the cold data need to be merged, and the hot data to be managed in the hot search database is deleted. Correspondingly, if it is determined that the maximum second matching degree is less than the third preset threshold, it indicates that the hot data to be managed corresponding to the maximum second matching degree is new cold data, and it is directly stored in the cold retrieval library.

The present application triggers the hot storage downgrade strategy through the first preset scale threshold of the hot retrieval database, and sinks the identities in the database with poor timeliness to the cold storage, thereby ensuring the latest size and identity of the hot retrieval database, and improving the retrieval accuracy and speed. The hot library downgrade strategy ensures that the hot data of the retrieval base library is the hot library identity after the time activity has decreased, so that the returned retrieval results are all the latest object information corresponding to the query identity.

S205: Calculate a first degree of matching between each cold data stored in the cold retrieval library and each hot data in the hot retrieval library.

The first matching degree is used to represent the degree of similarity between each cold data stored in the cold retrieval library and each hot data in the hot retrieval library. The number of the first matching degrees can be multiple, which can be determined by calculating the similarity between all the cold data and all the hot data. The similarity includes, but is not limited to, cosine similarity.

S207: If it is determined that the first matching degree satisfies the cold data management condition, migrate the cold data that meets the cold data management condition to the hot retrieval library.

Using the identity in the cold retrieval database to retrieve the hot retrieval database, the first matching degree of all the cold data in the cold retrieval database and all the hot data in the hot retrieval database is calculated. Cold data is upgraded and managed.

In this embodiment, as shown in FIG. 5 , if it is determined that the first matching degree satisfies the cold data management condition, the cold data that meets the cold data management condition is migrated to the hot retrieval library, including:

S401: Among the multiple first matching degrees corresponding to each cold data, if it is determined that the maximum first matching degree is greater than or equal to a fourth preset threshold, determine that the first matching degree satisfies the cold data management condition.

S403: Determine the cold data that meets the cold data management condition as the cold data to be managed.

S405: Migrate the cold data to be managed to the hot retrieval library.

Wherein, the fourth preset threshold may be a preset value, and its value can be configured but not limited to be any value from 0.8 to 1. The hot search library is retrieved using the identity in the cold search library. Among the multiple first matching degrees corresponding to each cold data, if it is determined that the maximum first matching degree is greater than or equal to the fourth preset threshold, it indicates that there are already stored in the hot search library. Hot data that is similar to the cold data corresponding to the maximum first matching degree, that is to say, the user corresponding to the identity of the cold data has recently visited again, so the identity information of the user is registered in the hot retrieval database, which will satisfy the cold data. The cold data corresponding to the data management condition is determined as the cold data to be managed, and the data is migrated to the hot retrieval database. For example, in one case, the user identifier of the cold retrieval library may be directly overwritten with the user identifier of the hot retrieval library, and the cold data corresponding to the identity of the cold retrieval library may be deleted. In another case, the cold data corresponding to the identity of the cold retrieval repository can be deleted directly.

Correspondingly, if it is determined that the maximum first matching degree is less than the fourth preset threshold, it means that the user corresponding to the identity of the cold data has not re-visited recently. At this time, the cold data will not be migrated, and the cold data will still be Stored in cold retrieval library.

The retrieval database in the embodiment of the present application includes a hot retrieval database and a cold retrieval database, and when the total number of hot data stored in the hot retrieval database is greater than a first preset scale threshold, obtain the data stored in the hot retrieval database. attribute parameters of each hot data; if it is determined that the attribute parameters of the hot data meet the hot data management conditions, the hot data that meets the hot data management conditions are migrated to the cold retrieval library; calculate the respective cold data stored in the cold retrieval library The first matching degree with each hot data in the hot retrieval library; if it is determined that the first matching degree satisfies the cold data management condition, the cold data satisfying the cold data management condition is migrated to the hot retrieval library. Therefore, by separating the hot and cold data of the retrieval library, when the total number of hot data stored in the hot retrieval library is greater than the first preset scale threshold, the upgrading and upgrading logic of the hot and cold library is triggered to ensure the scale of the hot retrieval library. Within the threshold range, the validity of the hot data in the hot search library is improved, and it is also beneficial to improve the search reliability, accuracy and search efficiency of the managed hot search library.

In some embodiments, as shown in FIG. 6 , the method may further include:

S501: Detect the total number of cold data stored in the cold retrieval library;

S503: If the total number of detected cold data is greater than the second preset size threshold, use the difference between the total number of detected cold data and the second preset size threshold as the number N to be dumped, where N is a positive integer;

S505: Sort all the cold data in the cold retrieval library according to the order of the latest recording time;

S507: Dump the top N cold data to the disk.

The second preset scale threshold may be determined according to the storage performance of the cold retrieval library. Exemplarily, the second preset scale threshold includes, but is not limited to, 100,000, 500,000, 1,000,000, and the like.

Specifically, after completing the above process of upgrading the cold storage and heating, the total number of all cold data in the current cold retrieval library is determined, and if the second preset scale threshold (configurable, such as 1 million) is reached, the cold storage cleaning module is triggered. The identities of the cold retrieval database are sorted according to the last visit time, and the data of the cold retrieval database is cleaned. The main task is to write the cold data with the longest access time (such as TOP N) in the cold retrieval library from memory to disk and delete them from memory to ensure that the scale of the cold retrieval library is controllable. The cold storage information written to the disk is mainly used for backup, and the business selects the required ID according to time or other dimensions for offline processing or targeted reactivation and re-entering the hot storage.

In some embodiments, the hot retrieval library is stored in a memory corresponding to an image processor or a sound processor for matching objects to be retrieved, and the cold retrieval library is stored in a memory corresponding to a central processing unit;

The object to be retrieved includes, but is not limited to, at least one of the following: human face, audio and iris.

As an example, taking the scene of retrieving faces as an example, if there are N face identities in the retrieval database, M face identities that meet the aging requirements can be stored in the video memory of the image processor GPU, and the remaining N-M face identities can be used as Cold data is stored in the memory of the central processing unit CPU. Therefore, by storing the hot and cold data separately, the dependence and demand on the high-performance GPU is greatly reduced, and the storage cost of the retrieval library is reduced.

The present application separates the retrieval database from hot and cold, without affecting each other, and does not affect online requests during the database management process, making full use of time resources to retrieve the hot retrieval database online during the day, and use the cold retrieval database to retrieve the hot retrieval database at night. At the same time, the hot search library with high query heat is stored in GPU memory, and the cold search library with inactive identity is stored in CPU memory, and the scale of hot library and cold library is guaranteed to be within the specified threshold, and the hot library scale is triggered when the upper limit reaches the threshold. Cold and hot storage upgrade and upgrade process; when the upper limit of the cold storage scale reaches the threshold, the storage clearance logic is triggered to ensure maximum utilization of machine CPU, GPU and disk resources, reducing the demand for expensive GPUs and reducing retrieval costs.

The following are apparatus embodiments of the present application, which can be used to execute the method embodiments of the present application. For details not disclosed in the device embodiments of the present application, please refer to the method embodiments of the present application.

Please refer to FIG. 8 , which shows a structural block diagram of a retrieval library management apparatus provided by an embodiment of the present application. The apparatus has the function of realizing the server side in the above method example, and the function may be realized by hardware or by executing corresponding software by the hardware. The retrieval library includes a hot retrieval library and a cold retrieval library, and the apparatus 60 may include:

An attribute acquisition module 601, configured to acquire attribute parameters of each hot data stored in the hot search library when the total number of hot data stored in the hot search library is greater than a first preset scale threshold;

A first migration module 602, configured to migrate the hot data satisfying the hot data management condition to the cold retrieval library if it is determined that the attribute parameter of the hot data satisfies the hot data management condition;

a first calculation module 603, configured to calculate the first matching degree of each cold data stored in the cold search library and each hot data in the hot search library;

The second migration module 604 is configured to migrate the cold data satisfying the cold data management condition to the hot retrieval library if it is determined that the first matching degree satisfies the cold data management condition.

In some embodiments, the attribute parameter of the hot data includes at least one of a timestamp of the last recording of the hot data and a user activity associated with the hot data.

In some embodiments, in the case that the attribute parameter of the hot data includes the last recorded time stamp of the hot data and the user activity level associated with the hot data, the first migration module includes:

The first determination unit is used for determining that the time interval between the last recorded timestamp of the hot data and the first preset time is greater than or equal to a first preset threshold, and determining that the user activity associated with the hot data is less than the second Presetting a threshold value to determine that the attribute parameter of the thermal data satisfies the thermal data management condition;

a second determining unit, configured to determine thermal data that satisfies the thermal data management condition as thermal data to be managed;

A first migration unit, configured to migrate the hot data to be managed to the cold retrieval library.

In some embodiments, the first migration unit includes:

a calculation subunit, configured to calculate the second degree of matching between the hot data to be managed and each cold data in the cold retrieval library;

a determining unit, configured to determine whether the maximum second matching degree is greater than or equal to a third preset threshold among the plurality of second matching degrees corresponding to the thermal data to be managed;

a data merging subunit, configured to perform data merging of hot data to be managed and cold data whose maximum second matching degree is greater than or equal to a third preset threshold if the determination result is yes;

The storage subunit is configured to store the hot data to be managed whose maximum second matching degree is less than the third preset threshold in the cold retrieval library if the determination result is negative.

In some embodiments, the second migration module includes:

a third determining unit, configured to, among the multiple first matching degrees corresponding to each cold data, determine that the first matching degree satisfies the cold data management condition if it is determined that the maximum first matching degree is greater than or equal to a fourth preset threshold;

a fourth determining unit, configured to determine cold data that meets the cold data management condition as cold data to be managed;

A second migration unit, configured to migrate the cold data to be managed to the hot retrieval library.

In some embodiments, the apparatus further includes:

a detection module for detecting the total number of cold data stored in the cold retrieval library;

A quantity determination module, configured to use the difference between the total number of detected cold data and the second preset size threshold as the number N to be dumped if the total number of detected cold data is greater than the second preset size threshold, where N is positive integer;

a sorting module, used for sorting all the cold data in the cold retrieval library according to the order of the latest record time;

The dump module is used to dump the top N cold data to the disk.

In some embodiments, the hot retrieval library is stored in a memory corresponding to an image processor or a sound processor for matching objects to be retrieved, and the cold retrieval library is stored in a memory corresponding to a central processing unit;

The object to be retrieved includes at least one of the following: human face, audio and iris.

An embodiment of the present application provides a retrieval library management device, the device may include a processor and a memory, and the memory stores at least one instruction, at least one piece of program, code set or instruction set, the at least one instruction, the at least one piece of program , the code set or the instruction set is loaded and executed by the processor to implement the retrieval library management method provided by the above method embodiments.

Embodiments of the present application further provide a computer-readable storage medium, where at least one instruction, at least one piece of program, code set or instruction set is stored in the storage medium, the at least one instruction, at least one piece of program, code set or instruction set The set is loaded by the processor and executes any of the retrieval library management methods described above.

A specific embodiment of a retrieval method of the present application is introduced below. FIG. 9 is a schematic flowchart of a retrieval method provided by an embodiment of the present application. Or non-creative work may involve more or fewer operational steps. The sequence of steps enumerated in the embodiments is only one of the execution sequences of many steps, and does not represent the only execution sequence. As shown in FIG. 9 , the execution body of the method can be a server in the above application environment, and retrieval is performed by using a retrieval database managed by data, and the retrieval database includes a hot retrieval database and a cold retrieval database. The method can include:

S701: Acquire feature data to be retrieved of the object.

In this embodiment, the objects here may be the user's face, audio, and iris.

For example, taking face retrieval as an example, a user's face image is collected by a collection device, and then image processing is performed on the face image to obtain feature data to be retrieved.

The collection device includes, but is not limited to, capturing a face photo for a camera, a self-portrait of an ID card, and an original video frame. The image processing includes, but is not limited to, image preprocessing, face detection and registration, and face feature extraction. Among them, the image preprocessing includes image noise reduction processing and the like. Face detection and registration are used to obtain face frames and face registration to obtain five-point face information. The facial feature extraction can be based on the facial feature extraction SDK that calculates the features according to the five points of the face, and obtains the feature data to be retrieved.

S703: Retrieve the feature data to be retrieved in the data-managed hot retrieval database to obtain a retrieval result.

S705: Return the retrieval result.

Wherein, the retrieval library performs data management through the above at least one retrieval library management method.

In some embodiments, the retrieval method may further include the step of managing the database, and the managing the database may include:

In the case where the total number of hot data stored in the hot search library is greater than a first preset scale threshold, acquiring attribute parameters of each hot data in the hot search library;

If it is determined that the attribute parameter of the hot data satisfies the hot data management condition, the hot data satisfying the hot data management condition is migrated to the cold retrieval library;

calculating a first degree of matching between each cold data stored in the cold retrieval library and each hot data in the hot retrieval library;

If it is determined that the first matching degree satisfies the cold data management condition, the cold data satisfying the cold data management condition is migrated to the hot retrieval library.

It should be noted that the specific content and beneficial effects of the database management here are similar to those of the above-mentioned embodiment, and are not repeated here.

For example, after determining the feature data to be retrieved, a conventional retrieval algorithm model based on machine learning is used to perform similarity retrieval on the feature data to be retrieved in the hot retrieval database managed by data, if the similarity is retrieved. When the hot data reaches the preset threshold, the corresponding hot data is used as the retrieval result and returned. If no hot data whose similarity reaches the preset threshold is retrieved, it means that the user corresponding to the feature data to be retrieved is a new identity user, and a new identity and its corresponding feature data are inserted into the hot search database for the new identity user. and add the new identity to the library.

It should be noted that the data management of the retrieval database by using any of the retrieval database management methods described above is performed periodically.

When the size of the hot search library reaches the set threshold, the data management strategy is triggered to complete the sinking of the hot data in the hot search library, and the scale of the hot search library is controllable and the latest hot data in the library is kept. The cold retrieval library receives the hot data sinking from the hot retrieval library, and searches the hot retrieval library regularly (configurable) every day, and completes the combination with the high retrieval threshold of the hot retrieval library. Logic to ensure that the scale of the cold retrieval library is controllable.

During the retrieval process of the present application, the retrieval is carried out by using the hot retrieval database managed by the data instead of the cold retrieval database, which can greatly improve the retrieval efficiency. As a transition library, the cold retrieval library is stored in the same retrieval library as the hot retrieval library, which facilitates data reading and writing between the two, and is also conducive to efficient management of the hot data in the hot retrieval library. In the database management process, by separating the hot and cold data of the retrieval library, when the total number of hot data stored in the hot retrieval library is greater than the first preset scale threshold, the upgrade and upgrade logic of the hot and cold library is triggered to ensure that the The scale of the hot search database for the retrieval base is within the threshold range, which not only improves the validity of the hot data in the hot search library, but also improves the retrieval reliability, accuracy and retrieval efficiency of the managed hot search library.

The retrieval method of the present application is specifically described below by taking a face retrieval scenario as an example. FIG. 10 is a schematic diagram of the framework of the present application for implementing a retrieval method based on a face retrieval scenario. As shown in Figure 10, the framework can include three parts: face input, face preprocessing and face retrieval.

Among them, 1) the face input part, the face picture can be obtained through the acquisition module. The collection module includes, but is not limited to, a face photo captured by a camera, a self-portrait of an ID card, and an original video frame.

2) The face preprocessing part can be realized by a face image preprocessing module, a face detection/registration module and a face feature extraction module. The face image preprocessing module is used to preprocess the acquired face image, and the preprocessing includes but is not limited to noise reduction processing. The face detection/registration module is used for detecting the preprocessed face picture to obtain a face frame, and performing face registration based on the face frame to obtain five-point face information. The face feature extraction module is used for calculating face features according to the acquired five-point information of the face. In practical applications, face preprocessing can be performed by running a software development kit (Software Development Kit, SDK).

3) The face retrieval part can include a retrieval library, the retrieval library includes a hot retrieval library and a cold retrieval library, wherein the hot retrieval library is stored in the video memory of the GPU, and the hot retrieval library can include a face feature management module, a retrieval module and a hot retrieval library. Search the library management module. The cold search library is stored in the memory of the CPU, and the cold search library may include a cold search library management module and a cold search library cleaning module. The hot search library management module, the cold search library management module and the cold search library cleaning module are independent of each other and can be performed concurrently and synchronously, so that the face search service can make full use of the computing resources of GPU and CPU for deep learning calculations, which greatly improves the overall performance of the device. computing power.

The face feature management module is used to perform corresponding feature management operations (for example, operations of adding, deleting, modifying, and checking features) on the retrieval database. The hot retrieval library is used as a retrieval base, which stores the hot data for face retrieval. The hot search library management module is used to manage the hot data stored in the hot search library, and when the above hot data management conditions are met, the hot data is migrated to the cold search library to realize the strategy of reducing the hot library to the cold library. The retrieval module is used for retrieving hot data similar to the face features to be retrieved in the hot retrieval database, and returns the retrieval results based on the retrieval similarity results.

The cold retrieval library management module is used to manage the cold data stored in the cold retrieval library, and when the above cold data management conditions are met, the cold data is migrated to the hot retrieval library to realize the strategy of upgrading the cold storage to the hot library. The cold retrieval library cleaning module is used to perform data cleaning on the cold data stored in the cold retrieval library, and transfer the cold data to the disk under the condition that the above data cleaning conditions are met.

The retrieval method of this application can be applied to the mainstream 1:N face retrieval framework in the field of computer vision CV, and plays an important role in identity filing, identity retrieval, identity merging, etc. in scenarios such as smart retail, smart community, and security. The timeliness and retrieval performance of the retrieval identity are guaranteed.

The retrieval method based on the upgrading and upgrading strategy of retrieval cold and hot storages proposed in the present application solves a series of problems caused by the excessively large scale of retrieval bases in current retrieval. The strategy mainly includes the time-based thermal storage sinking cold storage strategy and the retrieval thermal storage-based cold storage rising thermal storage strategy. The two strategies are synchronized according to their respective rules, which ensures the controllable size of the thermal retrieval database and the timeliness of the information in the storage. . The new retrieval base designed based on this strategy helps to improve the performance of retrieval algorithm and resource utilization.

The following are apparatus embodiments of the present application, which can be used to execute the method embodiments of the present application. For details not disclosed in the device embodiments of the present application, please refer to the method embodiments of the present application.

Please refer to FIG. 11 , which shows a structural block diagram of a retrieval apparatus provided by an embodiment of the present application. The apparatus has the function of realizing the server side in the above method example, and the function may be realized by hardware or by executing corresponding software by the hardware. The apparatus 80 may include:

A feature management module 801, configured to acquire feature data to be retrieved of an object;

A retrieval library management device 802, configured to perform data management on the retrieval library, the retrieval library includes a hot retrieval library and a cold retrieval library;

The retrieval module 803 is used for retrieving the characteristic data to be retrieved in the hot retrieval database managed by the data to obtain retrieval results;

Returning module 804, configured to return the retrieval result.

In one embodiment, the retrieval library management apparatus is any one of the retrieval library management apparatuses described above, and the retrieval library management apparatus may include:

an attribute acquisition module, configured to acquire attribute parameters of each hot data stored in the hot search library when the total number of hot data stored in the hot search library is greater than a first preset scale threshold;

a first migration module, configured to migrate the hot data satisfying the hot data management condition to the cold retrieval library if the attribute parameter of the hot data is determined to satisfy the hot data management condition;

a first calculation module, configured to calculate a first degree of matching between each cold data stored in the cold retrieval library and each hot data in the hot retrieval library;

The second migration module is configured to migrate the cold data satisfying the cold data management condition to the hot retrieval library if it is determined that the first matching degree satisfies the cold data management condition.

In some embodiments, the retrieval library management device includes a hot retrieval library management module and a cold retrieval library management module,

The hot search library management module includes the attribute acquisition module and the first migration module;

The cold retrieval library management module includes the first calculation module and the second migration module.

In a feasible embodiment, the retrieval module can also be used to calculate the first matching degree between the cold data stored in the cold retrieval library and the hot data in the hot retrieval library, and to calculate the relationship between the hot data to be managed and the cold retrieval library. The second matching degree of each cold data in . In this way, the above-mentioned calculation subunit and the first calculation module in the retrieval library management device can be omitted, the functions of the retrieval module are fully utilized, and additional feature matching function modules need not be additionally configured, thereby reducing the retrieval cost.

In some embodiments, the hot retrieval library is stored in an image processor or a sound processor for matching objects to be retrieved, and the cold retrieval library is stored in a central processing unit;

The object to be retrieved includes at least one of the following: human face, audio and iris.

An embodiment of the present application provides a retrieval device, the device may include a processor and a memory, the memory stores at least one instruction, at least a piece of program, code set or instruction set, the at least one instruction, the at least one piece of program, the at least one piece of program, the The code set or instruction set is loaded and executed by the processor to implement the retrieval method provided by the above method embodiments.

Embodiments of the present application further provide a computer-readable storage medium, where at least one instruction, at least one piece of program, code set or instruction set is stored in the storage medium, the at least one instruction, at least one piece of program, code set or instruction set The set is loaded by the processor and executes any of the retrieval methods described above.

Further, FIG. 12 shows a schematic diagram of the hardware structure of a device for implementing the retrieval library management method or retrieval method provided by the embodiment of the present application. The device may be a computer terminal, a mobile terminal, or other devices. The device may also participate in forming or including the apparatus provided by the embodiments of the present application. As shown in FIG. 12, the computer terminal 10 may include one or more processors 102 (102a, 102b, . processing means of a logic device FPGA or the like), a memory 104 for storing data, and a transmission means 106 for communication functions. In addition, may also include: display, input/output interface (I/O interface), universal serial bus (USB) port (may be included as one of the ports of the I/O interface), network interface, power supply and/or camera. Those of ordinary skill in the art can understand that the structure shown in FIG. 12 is only for illustration, and does not limit the structure of the above electronic device. For example, the computer terminal 10 may also include more or fewer components than those shown in FIG. 12 , or have a different configuration than that shown in FIG. 12 .

It should be noted that the one or more processors 102 and/or other data processing circuits described above may generally be referred to herein as "data processing circuits." The data processing circuit may be embodied in whole or in part as software, hardware, firmware or any other combination. Furthermore, the data processing circuitry may be a single stand-alone processing module, or incorporated in whole or in part into any of the other elements in the computer terminal 10 (or mobile device). As referred to in the embodiments of the present application, the data processing circuit acts as a kind of processor control (eg, selection of a variable resistance termination path connected to an interface).

The memory 104 can be used to store software programs and modules of application software, such as program instructions/data storage devices corresponding to the methods described in the embodiments of the present application, the processor 102 executes the software programs and modules stored in the memory 104 by running the software programs and modules. Various functional applications and data processing are to implement the above-mentioned neural network processing method. Memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, memory 104 may further include memory located remotely from processor 102, which may be connected to computer terminal 10 through a network. Examples of such networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

Transmission means 106 are used to receive or transmit data via a network. A specific example of the above-mentioned network may include a wireless network provided by a communication provider of the computer terminal 10 . In one example, the transmission device 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 may be a radio frequency (Radio Frequency, RF) module, which is used for wirelessly communicating with the Internet.

The display may be, for example, a touch screen type liquid crystal display (LCD) that enables a user to interact with the user interface of the computer terminal 10 (or mobile device).

It should be noted that: the above-mentioned order of the embodiments of the present application is only for description, and does not represent the advantages and disadvantages of the embodiments. And the above describes specific embodiments of the present application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that in the embodiments and still achieve desirable results. Additionally, the processes depicted in the figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Each embodiment in this application is described in a progressive manner, and the same and similar parts between the various embodiments may be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the apparatus and server embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and for related parts, please refer to the partial descriptions of the method embodiments.

Those of ordinary skill in the art can understand that all or part of the steps of implementing the above embodiments can be completed by hardware, or can be completed by instructing relevant hardware through a program, and the program can be stored in a computer-readable storage medium. The storage medium mentioned may be a read-only memory, a magnetic disk or an optical disk, etc.

The above descriptions are only preferred embodiments of the present application, and are not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the protection of the present application. within the range.

Claims (10)

1. A retrieval library management method, characterized in that, the retrieval library includes a hot retrieval library and a cold retrieval library, and the method comprises: In the case where the total number of hot data stored in the hot search library is greater than a first preset scale threshold, acquiring attribute parameters of each hot data in the hot search library; If it is determined that the attribute parameter of the hot data satisfies the hot data management condition, the hot data satisfying the hot data management condition is migrated to the cold retrieval library; calculating a first degree of matching between each cold data stored in the cold retrieval library and each hot data in the hot retrieval library; If it is determined that the first matching degree satisfies the cold data management condition, the cold data satisfying the cold data management condition is migrated to the hot retrieval library. 2. The method according to claim 1, wherein the attribute parameter of the hot data comprises at least one of a timestamp of the latest recording of the hot data and a user activity associated with the hot data; In the case that the attribute parameter of the hot data includes the last recorded time stamp of the hot data and the user activity associated with the hot data, if it is determined that the attribute parameter of the hot data satisfies the hot data management condition, the hot data management condition will be satisfied Migrate the hot data to the cold retrieval library, including: If it is determined that the time interval between the last recorded timestamp of the hot data and the first preset time is greater than or equal to a first preset threshold, and it is determined that the user activity associated with the hot data is less than a second preset threshold, determine the The attribute parameters of the thermal data meet the thermal data management conditions; Determine the thermal data that meets the thermal data management conditions as the thermal data to be managed; The hot data to be managed is migrated to the cold retrieval library. 3. The method according to claim 2, wherein the migrating the hot data to be managed to the cold retrieval repository comprises: calculating the second degree of matching between the hot data to be managed and the cold data in the cold retrieval library; Among the plurality of second matching degrees corresponding to the thermal data to be managed, determine whether the largest second matching degree is greater than or equal to a third preset threshold; If the determination result is yes, perform data merging of the hot data to be managed and the cold data whose maximum second matching degree is greater than or equal to the third preset threshold; If the determination result is no, the hot data to be managed whose maximum second matching degree is smaller than the third preset threshold is stored in the cold retrieval library. 4 . The method according to claim 1 , wherein, if it is determined that the first matching degree satisfies the cold data management condition, migrating the cold data that satisfies the cold data management condition to the hot retrieval library comprises: 5 . Among the multiple first matching degrees corresponding to each cold data, if it is determined that the maximum first matching degree is greater than or equal to a fourth preset threshold, it is determined that the first matching degree satisfies the cold data management condition; Determine the cold data that meets the cold data management conditions as the cold data to be managed; Migrate the cold data to be managed to the hot retrieval repository. 5. The method according to claim 1, wherein the method further comprises: detecting the total number of cold data stored in the cold retrieval repository; If the total number of detected cold data is greater than the second preset size threshold, the difference between the total number of detected cold data and the second preset size threshold is taken as the number N to be dumped, where N is a positive integer; Sort all cold data in the cold retrieval library according to the order of the latest recording time; Dump the top N cold data to disk. 6. The method according to claim 1, wherein the hot retrieval library is stored in a memory corresponding to an image processor or a sound processor for matching the object to be retrieved, and the cold retrieval library is stored in a central processing unit in the memory corresponding to the device; The object to be retrieved includes at least one of the following: human face, audio and iris. 7. A retrieval method, characterized in that the retrieval is carried out by using a retrieval database managed by data, and the retrieval database includes a hot retrieval database and a cold retrieval database, and the method comprises: Obtain the feature data to be retrieved of the object; Retrieve the feature data to be retrieved in the hot retrieval database managed by the retrieval database to obtain retrieval results; return the search result; Wherein, the retrieval library performs data management through the following retrieval library management method, and the database management method includes: In the case where the total number of hot data stored in the hot search library is greater than a first preset scale threshold, acquiring attribute parameters of each hot data in the hot search library; If it is determined that the attribute parameter of the hot data satisfies the hot data management condition, the hot data satisfying the hot data management condition is migrated to the cold retrieval library; calculating a first degree of matching between each cold data stored in the cold retrieval library and each hot data in the hot retrieval library; If it is determined that the first matching degree satisfies the cold data management condition, the cold data satisfying the cold data management condition is migrated to the hot retrieval library. 8. A retrieval library management device, wherein the retrieval library includes a hot retrieval library and a cold retrieval library, and the device comprises: an attribute acquisition module, configured to acquire attribute parameters of each hot data stored in the hot search library when the total number of hot data stored in the hot search library is greater than a first preset scale threshold; a first migration module, configured to migrate the hot data satisfying the hot data management condition to the cold retrieval library if the attribute parameter of the hot data is determined to satisfy the hot data management condition; a first calculation module, configured to calculate a first degree of matching between each cold data stored in the cold retrieval library and each hot data in the hot retrieval library; The second migration module is configured to migrate the cold data satisfying the cold data management condition to the hot retrieval library if it is determined that the first matching degree satisfies the cold data management condition. 9. A retrieval device, characterized in that a retrieval database managed by data is used for retrieval, the retrieval database includes a hot retrieval database and a cold retrieval database, and the device comprises: The feature management module is used to obtain the feature data to be retrieved of the object; The retrieval library management device according to claim 8, which is used for performing data management on the retrieval library, and the retrieval library includes a hot retrieval library and a cold retrieval library; a retrieval module, used for retrieving the feature data to be retrieved in the hot retrieval database managed by the data to obtain retrieval results; The return module is used to return the retrieval result. 10. A computer-readable storage medium, wherein the storage medium stores at least one instruction, at least one program, code set or instruction set, the at least one instruction, at least one program, code set or instruction set The retrieval library management method according to any one of claims 1 to 7 or the retrieval method according to claim 8 is loaded and executed by the processor.

Images