Feature processing method, device, electronic device and storage medium

Abstract

An embodiment of the present application provides a feature processing method, apparatus, electronic device, and storage medium, the method comprising: comparing feature data of an object to be identified with fused feature data of each candidate object in a feature database, respectively, to obtain a comparison score corresponding to each candidate object, wherein the feature database contains pre-entered original feature data and fused feature data of each candidate object; obtaining a maximum comparison score from all comparison scores, and in response to the maximum comparison score being greater than a target threshold, comparing the feature data of the object to be identified with the original feature data of a matching object in the feature database to obtain a fusion score, wherein the matching object is the candidate object corresponding to the maximum comparison score; and in response to the fusion score being greater than the target threshold, fusing the feature data of the object to be identified with the fused feature data of the matching object to obtain updated fused feature data of the matching object.

Description

Technical Field

[0001] The present application relates to the field of computer vision technology, and in particular to a feature processing method, device, electronic device and storage medium. Background Art

[0002] With the development of computer vision and target recognition technologies, the application of target recognition is increasing. For example, facial recognition is widely used in scenarios such as access and attendance. Currently, the target features in the scene map are compared with the base map features of each object in the database to identify the object that matches the target in the scene map. The base map features are usually just raw features extracted from a base map. However, if the scene map and the base map of the same object differ significantly, the target features in the scene map and the base map features of the same object may not be successfully compared, which affects the accuracy of the feature comparison and leads to a high rate of misidentification. Summary of the Invention

[0003] The present application provides a feature processing method, device, electronic device and storage medium, which can improve the accuracy of feature comparison and reduce the misrecognition rate.

[0004] The present invention provides a feature processing method, which includes:

[0005] Acquire an object to be identified and extract feature data of the object to be identified;

[0006] Comparing the feature data of the object to be identified with the fused feature data of each candidate object in a feature database, respectively, to obtain a comparison score corresponding to each candidate object, wherein the feature database contains the pre-entered original feature data and the fused feature data of each candidate object;

[0007] Obtaining a maximum comparison score from all the comparison scores, and in response to the maximum comparison score being greater than a target threshold, comparing the feature data of the object to be identified with the original feature data of the matching object in the feature database to obtain a fusion score, wherein the matching object is the candidate object corresponding to the maximum comparison score;

[0008] In response to the fusion score being greater than the target threshold, the feature data of the object to be identified is fused with the fused feature data of the matching object to obtain updated fused feature data of the matching object.

[0009] In the above embodiment, the feature data of the object to be identified is compared with the fused feature data of each candidate object to obtain a comparison score corresponding to each candidate object, which is used to determine whether the object to be identified and each candidate object are successfully matched. When the maximum comparison score is greater than the target threshold, the feature data of the object to be identified is further compared with the original feature data of the matching object to obtain a fusion score. When the fusion score is greater than the target threshold, the feature data of the object to be identified is fused with the fused feature data of the matching object to obtain updated fused feature data of the matching object. Accordingly, the fused feature data of the candidate object is not fixed, but is dynamically updated according to the feature fusion situation. It contains more effective feature information, which is conducive to improving the accuracy of feature comparison, reducing the false recognition rate, and thus improving the recognition ability.

[0010] In some possible implementations, the feature database further includes a feature fusion count for each candidate object; and after, in response to the fusion score being greater than the target threshold, fusing the feature data of the object to be identified with the fused feature data of the matching object to obtain updated fused feature data of the matching object, the method further includes:

[0011] The number of times the features of the matching objects in the feature database are fused is counted.

[0012] In the above implementation, the number of feature fusions of the matching objects in the feature database is counted, so as to quickly obtain the number of feature fusions of the matching objects.

[0013] In some possible implementations, before comparing the feature data of the object to be identified with the original feature data of the matching object in the feature database to obtain a fusion score in response to the maximum comparison score being greater than the comparison target threshold, the method further includes:

[0014] Obtaining the number of feature fusions of the matching object in the feature database;

[0015] The target threshold is obtained according to the number of feature fusions of the matching object and a predetermined mapping relationship, wherein the mapping relationship is used to characterize the corresponding relationship between the number of feature fusions and the comparison threshold.

[0016] In the above embodiment, the target threshold is related to the number of feature fusions of the matching object, so that the target threshold can dynamically adapt to the number of feature fusions, which is beneficial to improving the recognition pass rate while reducing the false recognition rate.

[0017] In some possible implementations, obtaining the target threshold according to the number of feature fusions of the matching object and a predetermined mapping relationship includes:

[0018] In response to the case where the number of feature fusion times of the matching object is less than or equal to a predetermined maximum number, obtaining, according to the number of feature fusion times of the matching object and the mapping relationship, a comparison threshold corresponding to the number of feature fusion times of the matching object as the target threshold;

[0019] In response to the situation that the number of feature fusions of the matching object is greater than the maximum number, a comparison threshold corresponding to the maximum number is obtained according to the maximum number and the mapping relationship as the target threshold.

[0020] In the above embodiment, the maximum number of times is introduced to represent the number of times the target threshold begins to stabilize. When the number of feature fusions of the matching object is greater than the maximum number of times, the target threshold remains at the comparison threshold corresponding to the maximum number and no longer changes. This is beneficial to avoid the negative impact of continuing to increase the target threshold on the recognition pass rate after the number of feature fusions reaches the maximum number.

[0021] In some possible implementations, the method for determining the mapping relationship includes: performing curve fitting based on a comparison threshold and a preset fusion number to obtain the mapping relationship.

[0022] In the above implementation, the correspondence between the number of feature fusion times and the comparison threshold is obtained by curve fitting, and the comparison threshold suitable for different number of feature fusion times can be quickly obtained according to the correspondence.

[0023] In some possible implementations, performing curve fitting based on the comparison threshold and the preset fusion number to obtain the mapping relationship includes:

[0024] Obtaining training sample feature data and test sample feature data;

[0025] Based on the training sample feature data and the comparison threshold used in the current round of training, the fused feature data of each sample object in the sample feature database is updated, and when the number of updates of the fused feature data of each sample object reaches the preset number of fusions in the current round of training, the latest fused feature data of each sample object is obtained;

[0026] Determining a comparison threshold for the next round of training based on the test sample feature data, the latest fused feature data of each sample object, and the target error recognition rate;

[0027] When the number of training rounds reaches the preset number of training rounds, curve fitting is performed based on the comparison threshold used in each round of training and the preset number of fusions in each round of training to obtain the mapping relationship.

[0028] In the above implementation, a comparison threshold that meets the target false recognition rate, as well as a correspondence between the number of feature fusions and the comparison threshold, are obtained through training and testing. In practical applications, a comparison threshold suitable for different numbers of feature fusions can be obtained based on this correspondence, which is beneficial to improving recognition accuracy while reducing the false recognition rate.

[0029] In some possible implementations, determining the comparison threshold for the next round of training based on the test sample feature data, the latest fused feature data of each sample object, and the target error recognition rate includes:

[0030] Comparing the test sample feature data with the latest fused feature data of each sample object to obtain a corresponding test comparison score;

[0031] The error rate is calculated based on the test comparison score and the candidate comparison threshold. From all candidate comparison thresholds that can make the calculated error rate meet the target error rate, the lowest candidate comparison threshold is selected as the comparison threshold for the next round of training.

[0032] In the above implementation, the comparison threshold used in each round of training is constrained by the target false recognition rate, so that the comparison threshold obtained through training can optimize the false recognition rate caused by feature fusion, which is conducive to reducing the false recognition rate.

[0033] In some possible implementations, fusing the feature data of the object to be identified with the fused feature data of the matching object to obtain updated fused feature data of the matching object includes:

[0034] Determining a fusion weight of the feature data of the object to be identified according to the fusion score;

[0035] According to the fusion weight of the feature data of the object to be identified, the feature data of the object to be identified and the fused feature data of the matching object are weighted averaged to obtain updated fused feature data of the matching object.

[0036] In the above implementation, the fusion weight is determined according to the fusion score, and feature fusion is performed based on the fusion weight, which can more effectively integrate the feature data of the object to be identified into the fused feature data of the matching object, so that the fused features add new effective features without affecting the existing features, which is conducive to improving the success rate of subsequent feature comparisons.

[0037] In some possible implementations, the method further includes:

[0038] In response to the maximum comparison score being greater than the target threshold, determining the recognition result of the object to be recognized as the matching object;

[0039] In response to the maximum comparison score being less than or equal to the target threshold, determining that the recognition result of the object to be recognized is recognition failure.

[0040] In the above embodiment, the recognition result of the object to be recognized is determined by comparing the maximum comparison score with the target threshold, which is conducive to improving the recognition accuracy.

[0041] The present application also provides a feature processing device, comprising:

[0042] An extraction unit, configured to obtain an object to be identified and extract feature data of the object to be identified;

[0043] a first comparison unit, configured to compare the feature data of the object to be identified with the fused feature data of each candidate object in a feature database, respectively, to obtain a comparison score corresponding to each candidate object, wherein the feature database contains pre-entered original feature data and the fused feature data of each candidate object;

[0044] a second comparison unit, configured to obtain a maximum comparison score from all the comparison scores, and in response to the maximum comparison score being greater than a target threshold, compare the feature data of the object to be identified with the original feature data of the matching object in the feature database to obtain a fusion score, wherein the matching object is the candidate object corresponding to the maximum comparison score;

[0045] A fusion unit is configured to, in response to the fusion score being greater than the target threshold, fuse the feature data of the object to be identified with the fused feature data of the matching object to obtain updated fused feature data of the matching object.

[0046] In some possible implementations, the feature database also includes the number of feature fusions for each candidate object; the device also includes a counting unit for, in response to the fusion score being greater than the target threshold, fusing the feature data of the object to be identified with the fused feature data of the matching object, and after obtaining the updated fused feature data of the matching object, counting the number of feature fusions of the matching object in the feature database.

[0047] In some possible embodiments, the device further includes an acquisition unit for comparing the feature data of the object to be identified with the original feature data of the matching object in the feature database in response to the maximum comparison score being greater than the target threshold, and obtaining the number of feature fusions of the matching object in the feature database before obtaining the fusion score, and obtaining the target threshold based on the number of feature fusions of the matching object and a predetermined mapping relationship, wherein the mapping relationship is used to characterize the correspondence between the number of feature fusions and the comparison threshold.

[0048] In some possible embodiments, when the acquisition unit obtains the target threshold based on the number of feature fusions of the matching object and a predetermined mapping relationship, it is specifically used to: in response to the situation that the number of feature fusions of the matching object is less than or equal to the predetermined maximum number, obtain the comparison threshold corresponding to the number of feature fusions of the matching object as the target threshold based on the number of feature fusions of the matching object and the mapping relationship; in response to the situation that the number of feature fusions of the matching object is greater than the maximum number, obtain the comparison threshold corresponding to the maximum number as the target threshold based on the maximum number and the mapping relationship.

[0049] In some possible implementations, the device further includes a determining unit configured to perform curve fitting based on a comparison threshold and a preset fusion number to obtain the mapping relationship.

[0050] In some possible implementations, when the determination unit performs curve fitting based on the comparison threshold and the preset fusion number to obtain the mapping relationship, it is specifically configured to:

[0051] Obtaining training sample feature data and test sample feature data;

[0052] Based on the training sample feature data and the comparison threshold used in the current round of training, the fused feature data of each sample object in the sample feature database is updated, and when the number of updates of the fused feature data of each sample object reaches the preset number of fusions in the current round of training, the latest fused feature data of each sample object is obtained;

[0053] Determining a comparison threshold for the next round of training based on the test sample feature data, the latest fused feature data of each sample object, and the target error recognition rate;

[0054] When the number of training rounds reaches the preset number of training rounds, curve fitting is performed based on the comparison threshold used in each round of training and the preset number of fusions in each round of training to obtain the mapping relationship.

[0055] In some possible implementations, when determining the comparison threshold for the next round of training based on the test sample feature data, the latest fused feature data of each sample object, and the target error recognition rate, the determination unit is specifically configured to:

[0056] Comparing the test sample feature data with the latest fused feature data of each sample object to obtain a corresponding test comparison score;

[0057] The error rate is calculated based on the test comparison score and the candidate comparison threshold. From all candidate comparison thresholds that can make the calculated error rate meet the target error rate, the lowest candidate comparison threshold is selected as the comparison threshold for the next round of training.

[0058] In some possible implementations, when the fusion unit fuses the feature data of the object to be identified with the fused feature data of the matching object to obtain the updated fused feature data of the matching object, it is specifically used to: determine the fusion weight of the feature data of the object to be identified based on the fusion score; and perform weighted averaging of the feature data of the object to be identified and the fused feature data of the matching object based on the fusion weight of the feature data of the object to be identified to obtain the updated fused feature data of the matching object.

[0059] In some possible embodiments, the device further includes an identification unit configured to: in response to a situation where the maximum comparison score is greater than the target threshold, determine that the identification result of the object to be identified is the matching object; and in response to a situation where the maximum comparison score is less than or equal to the target threshold, determine that the identification result of the object to be identified is an identification failure.

[0060] An embodiment of the present application further provides an electronic device, comprising: a processor and a memory, wherein the memory is used to store computer program code, the computer program code comprising computer instructions, and when the processor executes the computer instructions, the electronic device executes the above method.

[0061] An embodiment of the present application also provides an electronic device, comprising: a processor, a sending device, an input device, an output device and a memory, wherein the memory is used to store computer program code, and the computer program code includes computer instructions. When the processor executes the computer instructions, the electronic device executes the above method.

[0062] An embodiment of the present application further provides a computer-readable storage medium, in which a computer program is stored. The computer program includes program instructions, and when the program instructions are executed by a processor, the processor is caused to execute the above method.

[0063] An embodiment of the present application further provides a computer program product, which includes a computer program or instructions. When the computer program or instructions are run on a computer, the computer is caused to execute the above method.

[0064] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and are not intended to limit the technical solutions of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the background technology, the drawings required for use in the embodiments of the present application or the background technology will be described below.

[0066] The drawings herein are incorporated into and constitute a part of the specification. These drawings illustrate embodiments consistent with the present application and, together with the specification, are used to illustrate the technical solutions of the present application.

[0067] Figure 1 This is a flow chart of a feature processing method provided in an embodiment of the present application;

[0068] Figure 2 This is a flowchart of a method for determining a mapping relationship provided in an embodiment of the present application;

[0069] Figure 3 This is a flow chart of another feature processing method provided in an embodiment of the present application;

[0070] Figure 4 This is a structural diagram of a feature processing device provided in an embodiment of the present application;

[0071] Figure 5 is a structural diagram of another feature processing device provided in an embodiment of the present application;

[0072] Figure 6 This is a hardware structure diagram of a feature processing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0073] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0074] The terms "first," "second," and the like in the specification and claims of this application and the accompanying drawings are used to distinguish between different objects, not to describe a particular order. Furthermore, the terms "including," "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or elements is not limited to the listed steps or elements but may optionally include steps or elements not listed, or may optionally include other steps or elements inherent to the process, method, product, or apparatus.

[0075] It should be understood that in this application, "at least one (item)" refers to one or more, "more than one" refers to two or more, and "at least two (items)" refers to two or three and more than three. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " can indicate that the previous and next associated objects are in an "or" relationship, referring to any combination of these items, including any combination of single or plural items. For example, at least one item (item) of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0076] References to "embodiments" in this application mean that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0077] With the development of computer vision and target recognition technologies, the application of target recognition is increasing. For example, facial recognition is widely used in scenarios such as access and attendance. Currently, the target features in the scene map are compared with the base map features of each object in the database to identify the object that matches the target in the scene map. The base map features are usually just raw features extracted from a base map. However, if the scene map and base map of the same object differ significantly, the target features in the scene map and the base map features of the same object may not be successfully compared, thus affecting the feature comparison accuracy and resulting in a high misidentification rate.

[0078] For example, in a scenario where attendance is recorded based on facial recognition, a facial image is collected for each employee in advance as a base image and stored in a base database. After the on-site image is collected, the facial features of the on-site image are compared with the base image features of each employee in the base database to identify the employee whose face matches the on-site image. However, in actual applications, due to changes in employee appearance or clothing, for example, the same employee does not wear a hat (or a mask, or makeup) in the base image, but wears a hat (or a mask, or makeup) in the on-site image, the facial features of the employee's on-site image and the base image features may not be successfully matched, thereby affecting the accuracy of feature matching and, in turn, the accuracy of facial recognition.

[0079] Based on this, an embodiment of the present application provides a feature processing method to improve the accuracy of feature comparison and reduce the misrecognition rate.

[0080] The execution subject of the feature processing method may be a feature processing device. For example, the feature processing method may be executed by a terminal device, a server, or other processing device, wherein the terminal device may be a user equipment (UE), a mobile device, a user terminal, a terminal, a cellular phone, a cordless phone, a personal digital assistant (PDA), a handheld device, a computing device, an in-vehicle device, a wearable device, etc. In some possible implementations, the feature processing method may be implemented by a processor calling computer-readable instructions stored in a memory.

[0081] See also Figure 1 , Figure 1 : is a flow chart of a feature processing method provided in an embodiment of the present application, the feature processing method comprising:

[0082] S101: Acquire an object to be identified and extract feature data of the object to be identified.

[0083] The object to be identified refers to the object that needs to be identified. Specifically, an image to be identified can be obtained, the image to be identified including the object to be identified, and the object to be identified and its feature data can be extracted from the image to be identified. For example, in the face recognition scenario, the image to be identified is a facial image captured on-site, the object to be identified is a face, and the feature data of the object to be identified are facial features.

[0084] For example, existing feature extraction methods can be used to extract feature data of the object to be identified from the image to be identified. For example, a feature extraction network in an existing face recognition model can be used to extract facial features from a face image.

[0085] S102 , comparing the feature data of the object to be identified with the fused feature data of each candidate object in the feature database to obtain a comparison score corresponding to each candidate object, wherein the feature database contains pre-entered original feature data and fused feature data of each candidate object.

[0086] Candidate objects are objects that need to be compared with the object to be identified. This comparison is achieved by comparing the feature data of the object to be identified with the fused feature data of the candidate object. For example, in a facial recognition-based attendance system, the feature data of the object to be identified are the facial features in the facial images captured on-site by the company's attendance machine. The candidate objects are all company employees. Feature comparison is used to identify which employee matches the facial image captured on-site.

[0087] The fused feature data of a candidate object refers to the features used to compare with the feature data of the object to be identified to determine whether the candidate object and the object to be identified are a successful match. For any candidate object, the corresponding comparison score is the comparison score between the feature data of the object to be identified and the fused feature data of the candidate object. The comparison score can be understood as the similarity between the feature data of the object to be identified and the fused feature data of the candidate object. It is used to determine whether the feature data of the object to be identified and the fused feature data of the candidate object have been successfully compared, and therefore whether the object to be identified and the candidate object are a successful match.

[0088] For example, an existing feature comparison method (such as a cosine similarity calculation method) may be used to calculate the similarity between feature data to obtain a comparison score, which is in the range of 0 to 1.

[0089] The original feature data of a candidate object refers to the features extracted from the candidate object's base map, that is, the features that have not been fused. It should be noted that the original feature data of a candidate object is fixed, while the fused feature data of a candidate object is dynamic and is updated with each feature fusion.

[0090] The fused feature data of each candidate object corresponds to the original feature data one by one. Optionally, the fused feature data and the original feature data of each candidate object in the feature database are associated with an identifier (eg, id) of the candidate object.

[0091] S103, obtaining a maximum comparison score from all comparison scores, and in response to the maximum comparison score being greater than a target threshold, comparing the feature data of the object to be identified with the original feature data of the matching object in the feature database to obtain a fusion score, wherein the matching object is the candidate object corresponding to the maximum comparison score.

[0092] The larger the comparison score corresponding to the candidate object, the higher the similarity between the feature data of the object to be identified and the fused feature data of the candidate object, and the greater the probability that the object to be identified is successfully matched with the candidate object; the smaller the comparison score corresponding to the candidate object, the lower the similarity between the feature data of the object to be identified and the fused feature data of the candidate object, and the smaller the probability that the object to be identified is successfully matched with the candidate object.

[0093] Each candidate object corresponds to a comparison score. The maximum comparison score is selected from all comparison scores, and the candidate object corresponding to the maximum comparison score is determined as the matching object. The matching object can be understood as the candidate object with the highest probability of successfully matching the object to be identified among all candidate objects.

[0094] In an exemplary case, if at least two candidate objects have the same maximum matching score, the at least two candidate objects are recorded as candidate matching objects. The feature data of the object to be identified is compared with the original feature data of each candidate matching object to obtain a comparison score between the feature data of the object to be identified and the original feature data of each candidate matching object. This score is recorded as the secondary matching score corresponding to each candidate matching object. Each candidate matching object is associated with a secondary matching score, and the largest secondary matching score is selected. The candidate matching object corresponding to the largest secondary matching score is selected as the matching object.

[0095] The target threshold can be understood as the minimum match score required to indicate a successful feature match. In one example, the target threshold can be a preset fixed threshold. In another example, the target threshold can be a threshold related to the number of feature fusions, with different numbers of feature fusions corresponding to different thresholds. The target threshold can be determined based on the number of feature fusions for the matching object.

[0096] By comparing the matching score corresponding to the matching object with the target threshold, it is determined whether the feature data of the object to be identified and the fused feature data of the matching object have been successfully matched, and then whether the object to be identified and the matching object have been successfully matched. Specifically, if the matching score corresponding to the matching object is greater than the target threshold, it is determined that the feature data of the object to be identified and the fused feature data of the matching object have been successfully matched, and thus the object to be identified and the matching object have been successfully matched. If the matching score corresponding to the matching object is less than or equal to the target threshold, it is determined that the feature data of the object to be identified and the fused feature data of the matching object have failed to be matched, and thus the object to be identified and the matching object have failed to be matched, that is, there is no candidate object in the feature database that successfully matches the object to be identified.

[0097] The fusion score refers to the comparison score between the feature data of the object to be identified and the original feature data of the matching object. The fusion score can be understood as the similarity between the feature data of the object to be identified and the original feature data of the matching object. It is used to determine whether the feature data of the object to be identified and the original feature data of the matching object are successfully compared, and then determine whether to fuse the feature data of the object to be identified with the fused feature data of the matching object.

[0098] S104 , in response to the fusion score being greater than the target threshold, fusing the feature data of the object to be identified with the fused feature data of the matching object to obtain updated fused feature data of the matching object.

[0099] The larger the fusion score, the higher the similarity between the feature data of the object to be identified and the original feature data of the matching object, and the more conducive it is to improving the success rate of subsequent feature comparison after fusing the feature data of the object to be identified into the fused feature data of the matching object; the smaller the fusion score, the lower the similarity between the feature data of the object to be identified and the original feature data of the matching object, and the less conducive it is to improving the success rate of subsequent feature comparison after fusing the feature data of the object to be identified into the fused feature data of the matching object.

[0100] By comparing the fusion score with the target threshold, it is determined whether the feature data of the object to be identified and the original feature data of the matching object are successfully matched, and then it is determined whether the feature data of the object to be identified and the fused feature data of the matching object are fused. Specifically, if the fusion score is greater than the target threshold, it is determined that the feature data of the object to be identified and the original feature data of the matching object are successfully matched, and the feature data of the object to be identified and the fused feature data of the matching object are fused to obtain the updated fused feature data of the matching object. After obtaining the updated fused feature data of the matching object, the fused feature data of the matching object in the feature database is synchronously updated. Specifically, the updated fused feature data of the matching object replaces the current fused feature data of the matching object in the feature database and is used for the next comparison with the feature data of the object to be identified to determine whether the matching object and the object to be identified are successfully matched.

[0101] If the fusion score is less than or equal to the target threshold, it is determined that the feature data of the object to be identified and the original feature data of the matching object have failed to be compared, so there is no need to fuse the feature data of the object to be identified with the fused feature data of the matching object, that is, the fused feature data of the matching object is not updated. It should be understood that the fused feature data of the candidate object in the feature database is initially the same as the original feature data of the candidate object. When the candidate object does not undergo any feature fusion, the fused feature data of the candidate object remains the same as the original feature data of the candidate object; after the candidate object undergoes the first feature fusion, the fused feature data of the candidate object is updated to the feature data obtained by fusing the feature data of the object to be identified that was successfully matched for the first time and the original feature data of the candidate object, which is recorded as the first fused feature data; after the candidate object undergoes the second feature fusion, the fused feature data of the candidate object is updated to the feature data obtained by fusing the feature data of the object to be identified that was successfully matched for the second time and the first fused feature data of the candidate object, which is recorded as the second fused feature data; after the candidate object undergoes the third feature fusion, the fused feature data of the candidate object is updated to the feature data obtained by fusing the feature data of the object to be identified that was successfully matched for the third time and the second fused feature data of the candidate object; and so on, to achieve dynamic update of the fused feature data of the candidate object.

[0102] In the above embodiment, the feature data of the object to be identified is compared with the fused feature data of each candidate object to obtain a comparison score corresponding to each candidate object, which is used to determine whether the object to be identified and each candidate object are successfully matched. When the maximum comparison score is greater than the target threshold, the feature data of the object to be identified is further compared with the original feature data of the matching object to obtain a fusion score. When the fusion score is greater than the target threshold, the feature data of the object to be identified is fused with the fused feature data of the matching object to obtain updated fused feature data of the matching object. Accordingly, the fused feature data of the candidate object is not fixed, but is dynamically updated according to the feature fusion situation. It contains more effective feature information, which is conducive to improving the accuracy of feature comparison, reducing the false recognition rate, and thus improving the recognition ability.

[0103] It should be noted that in the above embodiment, the same threshold (i.e., the target threshold) is used to evaluate the comparison score and the fusion score to determine whether the feature comparison is successful. This can reduce the complexity of obtaining the target threshold without affecting the accuracy of the judgment and save computing resources.

[0104] In other embodiments, different thresholds (e.g., a first threshold and a second threshold) may be used to evaluate the comparison score and the fusion score to determine whether the feature comparison is successful. Specifically, in response to the comparison score corresponding to the matching object being greater than the first threshold, it is determined that the feature data of the object to be identified and the fused feature data of the matching object are successfully compared; in response to the fusion score being greater than the second threshold, it is determined that the feature data of the object to be identified and the original feature data of the matching object are successfully compared.

[0105] It should be understood that the comparison between the feature data of the object to be identified and the original feature data of the matching object is a one-to-one comparison. The impact of adding a comparison step to the recognition process on recognition efficiency is negligible. Furthermore, the entire feature fusion process is completed on the machine side, and the information of the object to be identified is not transferred back, thus protecting user privacy.

[0106] In a possible implementation, the feature database also includes the number of feature fusions for each candidate object; in response to the fusion score being greater than the target threshold, the feature data of the object to be identified is fused with the fused feature data of the matching object to obtain the updated fused feature data of the matching object, and the following step is also included: counting the number of feature fusions of the matching object in the feature database.

[0107] The number of feature fusions performed on a candidate object refers to the number of times the candidate object has undergone feature fusion, and can be used to reflect whether the candidate object's fused feature data has been fully fused. For example, a greater number of feature fusions performed on a candidate object indicates a more fully fused fused feature data; a smaller number of feature fusions performed on a candidate object indicates a less fully fused feature data.

[0108] Optionally, the number of feature fusions of each candidate object in the feature database is associated with the identifier (eg, ID) of the candidate object, so that the information associated with the identifier of each candidate object in the feature database includes: the original feature data, fused feature data, and the number of feature fusions of the candidate object.

[0109] Specifically, after the feature data of the object to be identified is fused with the fused feature data of the matching object to obtain the updated fused feature data for the matching object, the feature fusion count for the matching object in the feature database is incremented by one. It should be understood that the initial value of the feature fusion count for the matching object in the feature database is zero. If no feature fusion is performed on the matching object, the feature fusion count for the matching object remains unchanged at the initial value. Thereafter, each time feature fusion is performed on the matching object, the feature fusion count for the matching object in the feature database is incremented by one.

[0110] In the above implementation, the number of feature fusions of the matching objects in the feature database is counted, so as to quickly obtain the number of feature fusions of the matching objects.

[0111] In a possible embodiment, in response to the maximum comparison score being greater than the target threshold, the feature data of the object to be identified is compared with the original feature data of the matching object in the feature database, and before obtaining the fusion score, the following steps are also included: obtaining the number of feature fusions of the matching object in the feature database; obtaining the target threshold based on the number of feature fusions of the matching object and a predetermined mapping relationship, wherein the mapping relationship is used to characterize the correspondence between the number of feature fusions and the comparison threshold.

[0112] While feature fusion improves the recognition pass rate, the associated problem is that the false positive rate can also increase. Raising the target threshold can help address this issue. For features that have been fully fused, selecting a higher target threshold can maintain a high recognition pass rate while reducing the false positive rate. However, for features that have been fused less frequently, raising the target threshold can reduce the recognition pass rate.

[0113] To effectively mitigate the aforementioned drawbacks of feature fusion, this implementation considers both newly added features or features that have not yet been fully fused, as well as features that have already been fully fused, and determines the target threshold based on the number of feature fusions performed on the matching object. The number of feature fusions performed on a matching object refers to the number of times the matching object has undergone feature fusion. If no feature fusion has been performed on a matching object, the number of feature fusions performed on the matching object is zero. The number of feature fusions performed on the matching object is incremented by one with each subsequent feature fusion.

[0114] The mapping relationship represents the correspondence between the number of feature fusions and the comparison threshold. If the number of feature fusions is known, the corresponding comparison threshold can be obtained through this mapping relationship. After obtaining the number of feature fusions for the matching object, the comparison threshold corresponding to the number of feature fusions and the mapping relationship can be obtained, thereby obtaining the target threshold.

[0115] In the above embodiment, the target threshold is related to the number of feature fusions of the matching object, so that the target threshold can dynamically adapt to the number of feature fusions, which is beneficial to improving the recognition pass rate while reducing the false recognition rate.

[0116] In one possible implementation, a target threshold is obtained based on the number of feature fusions of the matching object and a predetermined mapping relationship, specifically including the following steps: in response to the situation where the number of feature fusions of the matching object is less than or equal to a predetermined maximum number, obtaining a comparison threshold corresponding to the number of feature fusions of the matching object as the target threshold based on the number of feature fusions of the matching object and the mapping relationship; in response to the situation where the number of feature fusions of the matching object is greater than the maximum number, obtaining a comparison threshold corresponding to the maximum number as the target threshold based on the maximum number and the mapping relationship.

[0117] When the number of feature fusions reaches a certain value, the corresponding target threshold stabilizes. The maximum number of times can be understood as the number of times the target threshold begins to stabilize. First, compare the number of feature fusions of the matching object with the maximum number of times, and then determine the target threshold based on the comparison result and the mapping relationship.

[0118] If the number of feature fusions of the matching object is less than or equal to the maximum number, the comparison threshold corresponding to the number of feature fusions is obtained based on the number of feature fusions and the mapping relationship, and is used as the target threshold. If the number of feature fusions of the matching object is greater than the maximum number, the comparison threshold corresponding to the maximum number is obtained based on the maximum number and the mapping relationship, and is used as the target threshold.

[0119] In the above embodiment, the maximum number of times is introduced to represent the number of times the target threshold begins to stabilize. When the number of feature fusions of the matching object is greater than the maximum number of times, the target threshold remains at the comparison threshold corresponding to the maximum number and no longer changes. This is beneficial to avoid the negative impact of continuing to increase the target threshold on the recognition pass rate after the number of feature fusions reaches the maximum number.

[0120] In a possible implementation, the method for determining the mapping relationship includes: performing curve fitting based on a comparison threshold and a preset fusion number to obtain the mapping relationship.

[0121] Based on the comparison threshold and the preset fusion times, at least two data pairs can be obtained, each data pair includes a comparison threshold and a feature fusion time, and the comparison threshold and the feature fusion time in each data pair correspond one to one.

[0122] Exemplarily, a scatter plot is drawn with the comparison threshold in each data pair as the y-axis data and the number of feature fusions in each data pair as the x-axis data. Each point in the scatter plot represents a data pair. The scatter plot is curve fitted to obtain a curve showing the change of the comparison threshold with the number of feature fusions, thereby obtaining the corresponding relationship between the number of feature fusions and the comparison threshold, that is, the mapping relationship.

[0123] In the above implementation, the correspondence between the number of feature fusion times and the comparison threshold is obtained by curve fitting, and the comparison threshold suitable for different number of feature fusion times can be quickly obtained according to the correspondence.

[0124] See also Figure 2 , Figure 2 2 is a flow chart of a method for determining a mapping relationship provided in an embodiment of the present application. The method for determining a mapping relationship includes the following steps S201 to S204.

[0125] S201, obtaining training sample feature data and test sample feature data.

[0126] Training sample feature data refers to feature data extracted from training sample images of sample objects, while test sample feature data refers to feature data extracted from test sample images of sample objects. Sample objects refer to the objects used for training to obtain mapping relationships. Training sample images refer to sample images used during training. Test sample images refer to sample images used during testing. Taking face recognition as an example, the sample objects are sample people. Both training sample images and test sample images are facial images of sample people. Therefore, both training sample feature data and test sample feature data are facial feature data.

[0127] S202, based on the training sample feature data and the comparison threshold used in the current round of training, the fused feature data of each sample object in the sample feature database is updated. When the number of updates of the fused feature data of each sample object reaches the preset number of fusions in the current round of training, the latest fused feature data of each sample object is obtained.

[0128] The sample feature database contains the pre-entered original feature data and fused feature data of each sample object. Taking the face recognition scenario as an example, the original feature data and fused feature data of each sample object are all face feature data. Before starting training, the initial comparison threshold, target error rate, number of training rounds, and number of fusions in each round of training (that is, the number of feature fusions required for each sample object in each round of training) are pre-set. The initial comparison threshold is used as the comparison threshold for the first round of training. After each round of training, the fused feature data of each sample object will be updated. The updated fused feature data of each sample object after each round of training is tested using the test sample feature data to obtain the error rate. The comparison threshold is updated with the error rate obtained not exceeding the target error rate as the test target for the next round of training.

[0129] It is known which sample object the training sample feature data actually corresponds to. In each round of training, for each piece of training sample feature data, if the comparison score of the training sample feature data with the fused feature data of the corresponding sample object, and the comparison score with the original feature data of the corresponding sample object, are both greater than the comparison threshold used in the current round of training, the training sample feature data is incorporated into the fused feature data of the corresponding sample object, completing a feature fusion of the corresponding sample object and completing an update of the fused feature data of the corresponding sample object. It should be understood that the number of updates to the fused feature data of a sample object is equal to the number of feature fusions for that sample object. After each feature fusion of a sample object is completed, the number of feature fusions for that sample object is counted and recorded.

[0130] Each round of training ends when the number of feature fusions for each sample object reaches the preset number of fusions for the current round of training. For example, if the preset number of fusions for each round of training is n, then after the end of round i, each sample object has undergone feature fusion i*n times, and the latest fused feature data is the fused feature data after being updated i*n times, where 1≤i≤m, where m represents the preset number of training rounds.

[0131] It should be understood that in other examples, the preset number of fusions for different rounds of training may also be different. For example, considering that the feature fusion that each sample object has experienced is more complete as time goes by, the preset number of fusions for later rounds of training may be reduced, thereby shortening the training time.

[0132] S203 : Determine a comparison threshold for the next round of training based on the test sample feature data, the latest fused feature data of each sample object, and the target error recognition rate.

[0133] It is known which sample object the test sample feature data actually corresponds to. After each round of training, for each test sample feature data, the sample object to which the test sample feature data matches is determined based on the comparison score between the test sample feature data and the latest fused feature data of each sample object. The error rate is calculated based on the sample object matched by each test sample feature data and the actual corresponding sample object. Under the constraint that the calculated error rate does not exceed the target error rate, the corresponding comparison threshold is calculated and used as the comparison threshold for the next round of training. The comparison threshold used for each round of training is recorded.

[0134] S204 , when the number of training rounds reaches a preset number of training rounds, curve fitting is performed based on the comparison threshold used in each training round and the preset number of fusions in each training round to obtain a mapping relationship.

[0135] The termination condition of the entire training is that the number of training rounds reaches the preset number of training rounds. When the number of training rounds reaches the preset number of training rounds, the training is terminated. The comparison threshold used in each round of training is counted (recorded as the comparison threshold for each round), and the number of feature fusions corresponding to the comparison threshold for each round is counted according to the preset number of fusions for each round of training. Specifically, the number of feature fusions corresponding to the comparison threshold for the i-th round is the sum of the preset number of fusions for the 1st to the i-1th rounds. For example, if the preset number of fusions for each round of training is n, the number of feature fusions corresponding to the comparison threshold for the i-th round is (i-1)*n, where 1≤i≤m, and m represents the preset number of training rounds.

[0136] For example, the comparison threshold used in each round of training is used as the y-axis data, and the number of fusions corresponding to each round of comparison threshold is used as the x-axis data. A scatter plot is drawn, and a curve is fitted on the scatter plot to obtain a curve showing that the comparison threshold changes with the number of feature fusions, thereby obtaining the corresponding relationship between the number of feature fusions and the comparison threshold.

[0137] When the number of feature fusion times reaches a certain value, the curve of the comparison threshold changing with the number of feature fusion times becomes flat and tends to be stable. The number of times when the comparison threshold begins to tend to be stable can be obtained from the curve. Optionally, the number of times when the comparison threshold begins to tend to be stable obtained through the curve is determined as the maximum number in the previous embodiment.

[0138] In the above implementation, a comparison threshold that meets the target false recognition rate, as well as a correspondence between the number of feature fusions and the comparison threshold, are obtained through training and testing. In practical applications, a comparison threshold suitable for different numbers of feature fusions can be obtained based on this correspondence, which is beneficial to improving recognition accuracy while reducing the false recognition rate.

[0139] In one possible implementation, based on the test sample feature data, the latest fused feature data of each sample object, and the target error rate, the comparison threshold used in the next round of training is determined, specifically including the following steps: comparing the test sample feature data with the latest fused feature data of each sample object respectively to obtain a corresponding test comparison score; calculating the error rate based on the test comparison score and the candidate comparison threshold, and selecting the lowest candidate comparison threshold from all candidate comparison thresholds that can make the calculated error rate meet the target error rate as the comparison threshold used in the next round of training.

[0140] It is known which sample object the test sample feature data actually corresponds to. After each round of training, for each test sample feature data, the comparison score between the test sample feature data and the latest fused feature data of each sample object is recorded as the test comparison score corresponding to each sample object. The test comparison score corresponding to each sample object is compared with the candidate comparison threshold. The sample object corresponding to the test comparison score greater than the candidate comparison threshold is determined to be the sample object matched by the test sample feature data. If the sample object matched by the test sample feature data is different from the sample object actually corresponding to the test sample feature data, it is recorded as a false match. Based on this, the false recognition rate is calculated. The false recognition rate can be understood as the ratio of the number of false matches to the total number of comparisons.

[0141] If the calculated error rate is greater than the target error rate, the calculated error rate is determined to not meet the target error rate. If the calculated error rate is less than or equal to the target error rate, the calculated error rate is determined to meet the target error rate. Using the constraint that the calculated error rate is less than or equal to the target error rate as the constraint, candidate comparison thresholds that meet this constraint are calculated, and the lowest candidate comparison threshold is selected as the comparison threshold for the next round of training.

[0142] In the above implementation, the comparison threshold used in each round of training is constrained by the target false recognition rate, so that the comparison threshold obtained through training can optimize the false recognition rate caused by feature fusion, which is conducive to reducing the false recognition rate.

[0143] In a possible implementation, the feature data of the object to be identified is fused with the fused feature data of the matching object to obtain updated fused feature data of the matching object. Specifically, the following steps may be included: determining the fusion weight of the feature data of the object to be identified based on the fusion score; and performing weighted averaging of the feature data of the object to be identified and the fused feature data of the matching object based on the fusion weight of the feature data of the object to be identified to obtain updated fused feature data of the matching object.

[0144] The fusion weight of the feature data of the object to be identified can be understood as the importance of the feature data of the object to be identified in the feature fusion. The fusion weight is determined based on the fusion score. The higher the fusion score, the higher the corresponding fusion weight; the lower the fusion score, the lower the corresponding fusion weight.

[0145] Exemplarily, when the fusion score is greater than the target threshold, the fusion score is in the range of t to 1, where t represents the target threshold, for example, t = 0.9. Optionally, the closer the fusion score is to 1, the closer the corresponding fusion weight is to the first weight; the closer the fusion score is to the target threshold, the closer the corresponding fusion weight is to the second weight. The first weight can be understood as the maximum fusion weight of the feature data of the object to be identified during the fusion process, and the second weight can be understood as the minimum fusion weight of the feature data of the object to be identified during the fusion process. For example, the first weight is set to 15% and the second weight is set to 0.

[0146] Exemplarily, the feature data of the object to be identified and the fused feature data of the matching object are weighted averaged, which is achieved by the following formula: F2 = a*F+(1-a)*F1, where F represents the feature data of the object to be identified, F1 represents the fused feature data of the matching object, F2 represents the updated fused feature data of the matching object, a represents the fusion weight of the feature data of the object to be identified, and 1-a represents the fusion weight of the fused feature data of the matching object.

[0147] In the above implementation, the fusion weight is determined according to the fusion score, and feature fusion is performed based on the fusion weight, which can more effectively integrate the feature data of the object to be identified into the fused feature data of the matching object, so that the fused features add new effective features without affecting the existing features, which is conducive to improving the success rate of subsequent feature comparisons.

[0148] In one possible implementation, in response to the maximum comparison score being greater than a target threshold, the recognition result of the object to be identified is determined to be a matching object; in response to the maximum comparison score being less than or equal to the target threshold, the recognition result of the object to be identified is determined to be a recognition failure.

[0149] The recognition result of the object to be identified is used to indicate whether there are candidate objects in the feature database that successfully match the object to be identified. If the maximum comparison score is greater than the target threshold, the recognition is successful and a candidate object that successfully matches the object to be identified exists in the feature database. The candidate object that successfully matches the object to be identified is the candidate object corresponding to the maximum comparison score, i.e., the matching object. If the maximum comparison score is less than or equal to the target threshold, the recognition fails and no candidate object successfully matches the object to be identified exists in the feature database.

[0150] After obtaining the recognition result of the object to be recognized, corresponding subsequent processing can be performed based on the recognition result. For example, in a scenario based on facial recognition access, the object to be recognized is the face of a person waiting to pass, and the candidate objects are people allowed to pass. If there is a person who successfully matches the person to pass among the allowed persons, the person to pass is allowed to pass; if there is no person who successfully matches the person to pass among the allowed persons, the person to pass is blocked.

[0151] In the above embodiment, the recognition result of the object to be recognized is determined by comparing the maximum comparison score with the target threshold, which is conducive to improving the recognition accuracy.

[0152] See also Figure 3 , Figure 3 It is a flow chart of a feature processing method provided in an embodiment of the present application, and the feature processing method includes the following steps S301 to S311.

[0153] S301: Acquire an object to be identified and extract feature data of the object to be identified.

[0154] S302 : Compare the feature data of the object to be identified with the fused feature data of each candidate object in the feature database to obtain a comparison score corresponding to each candidate object.

[0155] S303: Obtain a maximum matching score from all matching scores, and determine the candidate object corresponding to the maximum matching score as a matching object.

[0156] S304: Obtain the number of feature fusions of the matching object in the feature database, and obtain a target threshold value according to the number of feature fusions of the matching object and a predetermined mapping relationship.

[0157] S305: Determine whether the comparison score corresponding to the matching object is greater than the target threshold. If so, proceed to step S306; if not, proceed to step S311.

[0158] S306: The identification is successful, and the matching object is determined to be an object that successfully matches the object to be identified.

[0159] S307 , comparing the feature data of the object to be identified with the original feature data of the matching object in the feature database to obtain a fusion score.

[0160] S308: Determine whether the fusion score is greater than the target threshold. If so, proceed to step S309; ​​if not, end the process.

[0161] S309, determining the fusion weight of the feature data of the object to be identified according to the fusion score, and performing weighted averaging on the feature data of the object to be identified and the fused feature data of the matching object according to the fusion weight of the feature data of the object to be identified to obtain updated fused feature data of the matching object.

[0162] S310: Update the fusion feature data of the matching object in the feature database, and increase the number of feature fusion times of the matching object in the feature database by one.

[0163] S311: Recognition fails, and it is determined that there is no object in the feature database that successfully matches the object to be recognized.

[0164] For the detailed description and beneficial effects of steps S301 to S311 , please refer to the previous embodiments and will not be repeated here.

[0165] Those skilled in the art will understand that in the above-mentioned method of the specific implementation method, the writing order of each step does not mean a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of each step should be determined by its function and possible internal logic.

[0166] The above description of the various embodiments tends to emphasize the differences between the various embodiments. The same or similar aspects can be referenced with each other and will not be repeated herein for the sake of brevity.

[0167] See Figure 4 , Figure 4 This is a schematic diagram of the structure of a feature processing device provided in an embodiment of the present application. The feature processing device 400 includes: an extraction unit 401, a first comparison unit 402, a second comparison unit 403, and a fusion unit 404, wherein:

[0168] The extraction unit 401 is used to obtain the object to be identified and extract feature data of the object to be identified;

[0169] A first comparison unit 402 is configured to compare the feature data of the object to be identified with the fused feature data of each candidate object in the feature database, respectively, to obtain a comparison score corresponding to each candidate object, wherein the feature database contains pre-entered original feature data and fused feature data of each candidate object;

[0170] A second comparison unit 403 is configured to obtain a maximum comparison score from all the comparison scores, and in response to the maximum comparison score being greater than a target threshold, compare the feature data of the object to be identified with the original feature data of the matching object in the feature database to obtain a fusion score, wherein the matching object is the candidate object corresponding to the maximum comparison score;

[0171] The fusion unit 404 is configured to fuse the feature data of the object to be identified with the fused feature data of the matching object in response to the fusion score being greater than the target threshold, to obtain updated fused feature data of the matching object.

[0172] In some possible implementations, the feature database also includes the number of feature fusions for each candidate object; see Figure 5 The device 400 also includes a counting unit 405, which is used to, in response to the fusion score being greater than the target threshold, fuse the feature data of the object to be identified with the fused feature data of the matching object, and after obtaining the updated fused feature data of the matching object, count the number of times the features of the matching object in the feature database are fused.

[0173] For some possible implementations, see Figure 5 The device 400 also includes an acquisition unit 406, which is used to compare the feature data of the object to be identified with the original feature data of the matching object in the feature database in response to the maximum comparison score being greater than the target threshold in the second comparison unit 403, and before obtaining the fusion score, obtain the number of feature fusions of the matching object in the feature database, and obtain the target threshold according to the number of feature fusions of the matching object and a predetermined mapping relationship, wherein the mapping relationship is used to characterize the correspondence between the number of feature fusions and the comparison threshold.

[0174] In some possible implementations, when the acquisition unit 406 obtains the target threshold based on the number of feature fusions of the matching object and a predetermined mapping relationship, it is specifically used to: in response to the situation where the number of feature fusions of the matching object is less than or equal to the predetermined maximum number, obtain the comparison threshold corresponding to the number of feature fusions of the matching object according to the number of feature fusions of the matching object and the mapping relationship as the target threshold; in response to the situation where the number of feature fusions of the matching object is greater than the maximum number, obtain the comparison threshold corresponding to the maximum number according to the maximum number and the mapping relationship as the target threshold.

[0175] For some possible implementations, see Figure 5 The device 400 further includes a determining unit 407 for performing curve fitting based on a comparison threshold and a preset fusion number to obtain a mapping relationship.

[0176] In some possible implementations, when the determination unit 407 performs curve fitting based on the comparison threshold and the preset number of fusions to obtain a mapping relationship, it is specifically used to: obtain training sample feature data and test sample feature data; based on the training sample feature data and the comparison threshold used in the current round of training, update the fusion feature data of each sample object in the sample feature database, and obtain the latest fusion feature data of each sample object when the number of updates of the fusion feature data of each sample object reaches the preset number of fusions in the current round of training; determine the comparison threshold used for the next round of training based on the test sample feature data, the latest fusion feature data of each sample object, and the target error rate; when the number of training rounds reaches the preset number of training rounds, perform curve fitting based on the comparison threshold used in each round of training and the preset number of fusions in each round of training to obtain a mapping relationship.

[0177] In some possible implementations, when determining the comparison threshold to be used in the next round of training based on the test sample feature data, the latest fused feature data of each sample object, and the target error rate, the determination unit 407 is specifically used to: compare the test sample feature data with the latest fused feature data of each sample object respectively to obtain a corresponding test comparison score; calculate the error rate based on the test comparison score and the candidate comparison threshold, and select the lowest candidate comparison threshold from all candidate comparison thresholds that can make the calculated error rate meet the target error rate as the comparison threshold to be used in the next round of training.

[0178] In some possible implementations, when the fusion unit 404 fuses the feature data of the object to be identified with the fused feature data of the matching object to obtain updated fused feature data of the matching object, it is specifically used to: determine the fusion weight of the feature data of the object to be identified based on the fusion score; and perform weighted averaging of the feature data of the object to be identified and the fused feature data of the matching object based on the fusion weight of the feature data of the object to be identified to obtain updated fused feature data of the matching object.

[0179] For some possible implementations, see Figure 5 The device 400 further includes an identification unit 408, which is used to: in response to a situation where the maximum comparison score is greater than a target threshold, determine that the identification result of the object to be identified is a matching object; in response to a situation where the maximum comparison score is less than or equal to the target threshold, determine that the identification result of the object to be identified is an identification failure.

[0180] For the specific definition of the feature processing device, please refer to the definition of the feature processing method above and will not be repeated here. Each unit in the above-mentioned feature processing device can be implemented in whole or in part by software, hardware, or a combination thereof. Each of the above-mentioned units can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above-mentioned units.

[0181] See also Figure 6 , Figure 6 A hardware structure diagram of a feature processing device provided in an embodiment of the present application may include:

[0182] Processor 601, memory 602 and transceiver 603. The processor 601, memory 602 and transceiver 603 are connected via a bus 604. The memory 602 is used to store instructions, and the processor 601 is used to execute the instructions stored in the memory 602 to implement the steps in the above method.

[0183] The processor 601 is configured to execute instructions stored in the memory 602 to control the transceiver 603 to receive and send signals, thereby completing the steps in the above method. The memory 602 may be integrated into the processor 601 or may be provided separately from the processor 601.

[0184] As an implementation method, the functions of the transceiver 603 may be implemented by a transceiver circuit or a dedicated transceiver chip. The processor 601 may be implemented by a dedicated processing chip, a processing circuit, a processor, or a general-purpose chip.

[0185] As another implementation method, it is possible to use a general-purpose computer to implement the image processing device provided in the embodiment of the present application. That is, the program code that implements the functions of the processor 601 and the transceiver 603 is stored in the memory 602, and the general-purpose processor implements the functions of the processor 601 and the transceiver 603 by executing the code in the memory 602.

[0186] For the concepts, explanations, detailed descriptions and other steps involved in the device and related to the technical solutions provided in the embodiments of the present application, please refer to the description of the method steps performed by the device in the aforementioned method or other embodiments, which will not be repeated here.

[0187] An embodiment of the present application also provides an electronic device, including: a processor and a memory, the memory is used to store computer program code, the computer program code includes computer instructions, and when the processor executes the computer instructions, the electronic device executes the method in the above method embodiment.

[0188] An embodiment of the present application also provides an electronic device, including: a processor, a sending device, an input device, an output device and a memory, the memory is used to store computer program code, the computer program code includes computer instructions, and when the processor executes the computer instructions, the electronic device executes the method in the above method embodiment.

[0189] An embodiment of the present application further provides a computer-readable storage medium, in which a computer program is stored. The computer program includes program instructions. When the program instructions are executed by a processor, the processor executes the method in the above method embodiment.

[0190] An embodiment of the present application further provides a computer program product, which includes a computer program or instructions. When the computer program or instructions are run on a computer, the computer is enabled to execute the method in the above method embodiment.

[0191] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0192] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0193] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A feature processing method, characterized in that: The method comprises: Acquire an image to be identified, wherein the image to be identified includes an object to be identified, and extract the object to be identified and feature data of the object to be identified from the image to be identified; Comparing the feature data of the object to be identified with the fused feature data of each candidate object in a feature database, respectively, to obtain a comparison score corresponding to each candidate object, wherein the feature database contains the pre-entered original feature data and the fused feature data of each candidate object; Obtaining a maximum comparison score from all the comparison scores, and in response to the maximum comparison score being greater than a target threshold, comparing the feature data of the object to be identified with the original feature data of the matching object in the feature database to obtain a fusion score, wherein the matching object is the candidate object corresponding to the maximum comparison score; In response to the fusion score being greater than the target threshold, fusing the feature data of the object to be identified with the fused feature data of the matching object to obtain updated fused feature data of the matching object; The target threshold is related to the number of feature fusions of the matching object.

2. The method according to claim 1, characterized in that The feature database further includes a feature fusion count for each candidate object; in response to the fusion score being greater than the target threshold, after fusing the feature data of the object to be identified with the fused feature data of the matching object to obtain updated fused feature data of the matching object, the method further includes: The number of times the features of the matching objects in the feature database are fused is counted.

3. The method according to claim 2, characterized in that In response to the maximum comparison score being greater than a target threshold, before comparing the feature data of the object to be identified with the original feature data of the matching object in the feature database to obtain a fusion score, the method further includes: Obtaining the number of feature fusions of the matching object in the feature database; The target threshold is obtained according to the number of feature fusions of the matching object and a predetermined mapping relationship, wherein the mapping relationship is used to characterize the corresponding relationship between the number of feature fusions and the target threshold.

4. The method according to claim 3, characterized in that The obtaining of the target threshold according to the number of feature fusions of the matching object and a predetermined mapping relationship includes: In response to the case where the number of feature fusion times of the matching object is less than or equal to a predetermined maximum number, obtaining, according to the number of feature fusion times of the matching object and the mapping relationship, a comparison threshold corresponding to the number of feature fusion times of the matching object as the target threshold; In response to the situation that the number of feature fusions of the matching object is greater than the maximum number, a comparison threshold corresponding to the maximum number is obtained according to the maximum number and the mapping relationship as the target threshold.

5. The method according to claim 3, characterized in that The method for determining the mapping relationship includes: Based on the comparison threshold and the preset fusion times, curve fitting is performed to obtain the mapping relationship.

6. The method according to claim 5, characterized in that The curve fitting is performed based on the comparison threshold and the preset fusion number to obtain the mapping relationship, including: Obtaining training sample feature data and test sample feature data; Based on the training sample feature data and the comparison threshold used in the current round of training, the fused feature data of each sample object in the sample feature database is updated, and when the number of updates of the fused feature data of each sample object reaches the preset number of fusions in the current round of training, the latest fused feature data of each sample object is obtained; Determining a comparison threshold for the next round of training based on the test sample feature data, the latest fused feature data of each sample object, and the target error recognition rate; When the number of training rounds reaches the preset number of training rounds, curve fitting is performed based on the comparison threshold used in each round of training and the preset number of fusions in each round of training to obtain the mapping relationship.

7. The method according to claim 6, characterized in that The determining of a comparison threshold for the next round of training based on the test sample feature data, the latest fused feature data of each sample object, and the target error recognition rate includes: Comparing the test sample feature data with the latest fused feature data of each sample object to obtain a corresponding test comparison score; The error rate is calculated based on the test comparison score and the candidate comparison threshold. From all candidate comparison thresholds that can make the calculated error rate meet the target error rate, the lowest candidate comparison threshold is selected as the comparison threshold for the next round of training.

8. The method according to any one of claims 1 to 7, characterized in that The step of fusing the feature data of the object to be identified with the fused feature data of the matching object to obtain updated fused feature data of the matching object includes: Determining a fusion weight of the feature data of the object to be identified according to the fusion score; According to the fusion weight of the feature data of the object to be identified, the feature data of the object to be identified and the fused feature data of the matching object are weighted averaged to obtain updated fused feature data of the matching object.

9. The method according to any one of claims 1 to 7, characterized in that Also includes: In response to the maximum comparison score being greater than the target threshold, determining the recognition result of the object to be recognized as the matching object; In response to the maximum comparison score being less than or equal to the target threshold, determining that the recognition result of the object to be recognized is recognition failure.

10. A feature processing device, characterized in that: The device comprises: an extraction unit, configured to obtain an image to be identified, wherein the image to be identified includes an object to be identified, and extract the object to be identified and feature data of the object to be identified from the image to be identified; a first comparison unit, configured to compare the feature data of the object to be identified with the fused feature data of each candidate object in a feature database, respectively, to obtain a comparison score corresponding to each candidate object, wherein the feature database contains pre-entered original feature data and the fused feature data of each candidate object; a second comparison unit, configured to obtain a maximum comparison score from all the comparison scores, and in response to the maximum comparison score being greater than a target threshold, compare the feature data of the object to be identified with the original feature data of the matching object in the feature database to obtain a fusion score, wherein the matching object is the candidate object corresponding to the maximum comparison score; a fusion unit, configured to, in response to a situation where the fusion score is greater than the target threshold, fuse the feature data of the object to be identified with the fused feature data of the matching object to obtain updated fused feature data of the matching object; The target threshold is related to the number of feature fusions of the matching object.

11. An electronic device, characterized in that: The method comprises at least one processor and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the at least one processor implements the method according to any one of claims 1 to 9 by executing the instructions stored in the memory.

12. A computer-readable storage medium, characterized in that The computer-readable storage medium stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by a processor to implement the method according to any one of claims 1 to 9.

Images