A multi-target rapid identification method and system

Abstract

The present invention discloses a multi-target rapid recognition method and system, which belongs to the field of computer vision, wherein the method includes: setting a target recognition set; configuring an adaptive threshold anchor frame based on the spatial distribution of the anchor frame space; when performing recognition of the target recognition set, performing feature frame selection through the adaptive threshold anchor frame, and reconstructing the anchor frame according to the feature frame selection result; establishing a feature mapping between the anchor frame size and the target feature, performing test recognition, inputting the test recognition result into a tolerance configuration network, and generating a tolerance configuration optimization result; performing mapping tolerance adjustment of the feature mapping through the tolerance configuration optimization result; completing rapid target recognition of the target recognition set according to the mapping tolerance adjustment result and reconstructing the anchor frame. The present application solves the problem that it is difficult to cope with complex spatial distribution and large computing resource overhead in multi-target recognition in the prior art, and achieves the technical effect of improving target recognition accuracy and efficiency, realizing real-time recognition and adjusting tolerance to balance accuracy and speed.

Description

Technical Field

[0001] The present invention relates to the field of computer vision, and in particular to a multi-target rapid recognition method and system. Background Art

[0002] Object recognition is widely used in many fields such as monitoring, autonomous driving, and intelligent robots. However, when dealing with multi-target scenes, traditional object recognition methods use fixed threshold selection, which cannot adapt to the different characteristics and complex spatial distribution of multiple targets, resulting in missed detection and false detection. At the same time, due to the large computing resource overhead, it is difficult to meet the real-time requirements, limiting the application scope of multi-target recognition. Summary of the invention

[0003] The present application provides a method and system for rapid multi-target identification, aiming to solve the technical problems in the prior art that multi-target identification is difficult to cope with complex spatial distribution and has high computing resource overhead, resulting in low accuracy and efficiency of multi-target identification.

[0004] In view of the above problems, the present application provides a multi-target rapid identification method and system.

[0005] In a first aspect disclosed in the present application, a method for rapid identification of multiple targets is provided, the method comprising: setting a target identification set, wherein the target identification set is a target data set to be identified, performing initial data interaction based on the target identification set, performing feature extraction according to the initial data interaction result, and establishing an anchor frame space for feature representation; configuring an adaptive threshold anchor frame based on the spatial distribution of the anchor frame space, wherein the spatial distribution includes an anchor frame space extreme value distribution and an anchor frame space concentrated distribution; when identifying the target identification set, performing feature selection through the adaptive threshold anchor frame, generating selection feedback according to the feature selection results of each adaptive threshold anchor frame, and reconstructing the anchor frame according to the selection feedback result; establishing a feature mapping between the anchor frame size and the target feature through the spatial distribution, and performing test identification through the feature mapping and the reconstructed anchor frame; inputting the test identification result into a tolerant configuration network to generate a tolerant configuration optimization result; performing mapping tolerance adjustment of the feature mapping through the tolerant configuration optimization result; and completing rapid target identification of the target identification set according to the mapping tolerance adjustment result and the reconstructed anchor frame.

[0006] Another aspect disclosed in the present application provides a multi-target rapid recognition system, the system comprising: a recognition set setting module, used to set a target recognition set, wherein the target recognition set is a target data set to be recognized, initial data interaction is performed based on the target recognition set, and feature extraction is performed according to the initial data interaction result to establish an anchor frame space for feature representation; a threshold anchor frame setting module, used to configure an adaptive threshold anchor frame based on the spatial distribution of the anchor frame space, wherein the spatial distribution includes an extreme value distribution of the anchor frame space and a concentrated distribution of the anchor frame space; an anchor frame reconstruction module, used to reconstruct the target recognition set through the adaptive threshold anchor frame when recognizing the target recognition set; Execute feature selection, generate selection feedback according to the feature selection results of each adaptive threshold anchor frame, and reconstruct the anchor frame according to the selection feedback result; a test recognition module is used to establish a feature mapping between the anchor frame size and the target feature through the spatial distribution, and perform test recognition through the feature mapping and the reconstructed anchor frame; an optimization result generation module is used to input the test recognition result into the tolerance configuration network to generate a tolerance configuration optimization result; a mapping tolerance adjustment module is used to perform mapping tolerance adjustment of the feature mapping through the tolerance configuration optimization result; a fast target recognition module is used to complete the fast target recognition of the target recognition set according to the mapping tolerance adjustment result and the reconstructed anchor frame.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0008] The target recognition set is set to clearly define the target data set to be processed; initial data interaction is performed on the target recognition set and feature extraction is performed to establish an anchor box space for feature representation; based on the spatial extreme value distribution of the anchor box space and the concentrated distribution of the anchor box space, an adaptive threshold anchor box is configured to adjust the threshold according to the spatial distribution characteristics of different anchor boxes to achieve more accurate feature box selection; on the target recognition set, feature box selection is performed using an adaptive threshold anchor box, and box selection feedback is generated based on the feature box selection results of each adaptive threshold anchor box to guide anchor box adjustment and reconstruct the anchor box; test recognition is performed based on the feature mapping of the anchor box size and the target feature, as well as the reconstructed anchor box. To achieve rapid recognition of the target; input the test recognition result into the tolerance configuration network, generate the tolerance configuration optimization result, and realize the tolerance configuration of the recognition result according to different needs to balance the accuracy and speed; perform mapping tolerance adjustment of the feature mapping through the tolerance configuration optimization result, and realize the adjustment of the feature mapping to further improve the recognition accuracy; according to the mapping tolerance adjustment result and the reconstructed anchor frame, complete the technical solution of rapid target recognition of the target recognition set, solve the problem that it is difficult to cope with complex spatial distribution and large computing resource overhead in multi-target recognition in the prior art, and achieve the technical effect of improving the target recognition accuracy and efficiency, realizing real-time recognition and adjusting the tolerance to balance the accuracy and speed.

[0009] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 A flowchart of a multi-target rapid identification method is provided for an embodiment of the present application;

[0011] Figure 2 A schematic diagram of a flow chart for generating a tolerant configuration optimization result in a multi-target rapid identification method is provided for an embodiment of the present application;

[0012] Figure 3 A structural schematic diagram of a multi-target rapid identification system is provided for an embodiment of the present application.

[0013] Explanation of the reference numerals: recognition set setting module 11, threshold anchor box setting module 12, anchor box reconstruction module 13, test recognition module 14, optimization result generation module 15, mapping tolerance adjustment module 16, fast target recognition module 17. DETAILED DESCRIPTION

[0014] The overall idea of ​​the technical solution provided by this application is as follows:

[0015] The embodiments of the present application provide a method and system for rapid identification of multiple targets. First, initial data interaction and feature extraction are performed on the target identification set to establish an anchor frame space for feature representation. Next, based on the spatial distribution of the anchor frame space, an adaptive threshold anchor frame is configured to achieve more accurate feature frame selection. Subsequently, in target identification, an adaptive threshold anchor frame is used for feature frame selection, and frame selection feedback is generated to guide the reconstruction of the anchor frame. Through feature mapping and tolerance configuration networks, test identification and tolerance configuration are achieved, thereby effectively balancing the accuracy and speed of identification. Finally, through mapping tolerance adjustment, the accuracy of feature mapping is improved, and rapid identification of multiple targets is completed.

[0016] After introducing the basic principles of the present application, various non-limiting implementation methods of the present application will be specifically described below in conjunction with the drawings in the specification.

[0017] Embodiment 1

[0018] like Figure 1 As shown, the embodiment of the present application provides a method for rapid identification of multiple targets, the method comprising:

[0019] Setting a target recognition set, wherein the target recognition set is a target data set to be recognized, performing initial data interaction based on the target recognition set, performing feature extraction according to the initial data interaction result, and establishing an anchor frame space for feature representation;

[0020] In an embodiment of the present application, first, a target recognition set is set, which is a group of target data to be recognized, such as images, video frames or other visual data, which come from different acquisition sources, such as surveillance cameras, sensors or image databases.

[0021] After setting the target recognition set, initial data interaction is performed based on the target recognition set, that is, the target data in the target recognition set is preprocessed for subsequent feature extraction and anchor frame establishment. Among them, data interaction includes image enhancement, denoising, size standardization, etc., to obtain the initial data interaction results to ensure that the data in the target recognition set obtains the best performance in subsequent processing. Then, feature extraction is performed on the initial data interaction results to extract useful features. For example, visual characteristics such as color, texture, shape, etc. of the data in the target recognition set are extracted to distinguish different targets.

[0022] Next, an anchor box space for feature representation is established based on the result of feature extraction. The anchor box space is a rectangular area used for target detection, and its position and size can be adjusted to accommodate different targets.

[0023] configuring an adaptive threshold anchor frame based on a spatial distribution of the anchor frame space, wherein the spatial distribution includes an extreme value distribution of the anchor frame space and a concentrated distribution of the anchor frame space;

[0024] In the embodiment of the present application, the spatial distribution of the anchor box space refers to the arrangement and distribution of the anchor boxes in the image, including information such as the position, density and scale of the anchor boxes, and the extreme value distribution of the anchor box space and the concentrated distribution of the anchor box space are determined from the information of the anchor boxes. Among them, the extreme value distribution of the anchor box space means that the position of the anchor box in the anchor box space corresponds to the extreme or boundary position where the target may appear; the concentrated distribution of the anchor box space means that the position of the anchor box in the anchor box space corresponds to the area where the target is densely distributed, which is highly correlated with the target features.

[0025] First, a detailed analysis is performed on the anchor boxes, including their position, size, density, and relationship with target features in the image, so as to understand the distribution of different anchor boxes in the image. Then, based on the analysis of the anchor box space, the extreme value distribution of the anchor box space and the concentrated distribution of the anchor box space are determined to configure the adaptive threshold anchor box. For the extreme value distribution of the anchor box space, a relatively low adaptive threshold anchor box is selected to improve the sensitivity of detection; for the concentrated distribution of the anchor box space, a relatively high adaptive threshold anchor box is selected to improve the specificity of detection.

[0026] During the object detection process, the configured adaptive threshold anchor frame is applied according to the position and size of the anchor frame, that is, different thresholds are used at different anchor frame positions to adaptively determine whether there are target features in the anchor frame, thereby improving the accuracy and robustness of object detection.

[0027] When performing recognition of the target recognition set, feature selection is performed through the adaptive threshold anchor frame, and selection feedback is generated according to the feature selection results of each adaptive threshold anchor frame, and the anchor frame is reconstructed according to the selection feedback results;

[0028] In an embodiment of the present application, when performing target recognition, the configured adaptive threshold anchor box is used to perform feature selection in the target recognition set, to achieve adaptive selection of the target data to be identified in the target recognition set, and to obtain the feature selection results of each adaptive threshold anchor box, indicating whether the target feature is detected at each anchor box position.

[0029] For the feature selection results of each adaptive threshold anchor box, the coordinate position and size information are recorded, and the similarity between the features in the anchor box and the target features is compared to obtain the feature similarity of each anchor box, which indicates the matching degree of the target features in the anchor box as the selection feedback result. Then, based on the selection feedback results, determine which anchor boxes need to be adjusted in position, for example, fine-tune the coordinate position of the anchor box to more accurately match the target features; based on the feature similarity score, determine which anchor boxes need to be resized to adapt to the size of the target features. By adjusting the position and size, the anchor box can be reconstructed to match the target features more accurately, which helps to improve the recognition accuracy of the target.

[0030] Establishing a feature map between the anchor box size and the target feature through the spatial distribution, and performing test recognition through the feature map and the reconstructed anchor box;

[0031] In the embodiment of the present application, first, the relationship between the target feature and the anchor frames of different sizes is investigated, and a feature map is established. The feature map is an association table that establishes a connection between the anchor frames of different sizes and the target features, so as to associate the anchor frames of different sizes with the matching target features to better adapt to the size change of the target. Then, the size of each anchor frame in the reconstructed anchor frame is identified, and the target feature is identified according to the feature map based on the anchor frame size, thereby achieving test recognition and obtaining a test recognition result.

[0032] Input the test identification results into the tolerant configuration network to generate the tolerant configuration optimization results;

[0033] Further, such as Figure 2 As shown, this step includes:

[0034] After the test recognition result is input into the tolerant configuration network, the data of the test recognition result is analyzed by the analysis subunit to determine the anchor box deviation data set of the selected anchor box and the real anchor box;

[0035] The anchor frame deviation data set is transmitted to the analysis subunit, a tolerance calibration interval and a tolerance limit interval are set by the analysis subunit, and an interval call frequency optimization is performed on the tolerance calibration interval and the tolerance limit interval;

[0036] The optimization result, the tolerance calibration interval and the tolerance limit interval are taken as the tolerance configuration optimization result.

[0037] In a preferred embodiment, after obtaining the test recognition result, the test recognition result is input into the tolerant configuration network. The tolerant configuration network uses a parsing subunit to carefully analyze the test recognition result, associate the selected anchor box with the corresponding real anchor box, and calculate the deviation data between each selected anchor box and its corresponding real anchor box, indicating the deviation degree of the precise position of the target in the image, and obtaining the anchor box deviation data set.

[0038] Then, the anchor deviation data set is transmitted to the analysis subunit. Based on the target recognition performance and requirements, the analysis subunit determines the tolerance calibration interval, which is used to represent the normal deviation interval of the anchor deviation data; at the same time, the analysis subunit sets the tolerance limit interval, which represents the maximum acceptable deviation degree for the deviation data. Subsequently, the analysis subunit optimizes the calling frequency of the tolerance calibration interval and the tolerance limit interval, that is, selects the appropriate tolerance calibration interval and tolerance limit interval in different scenarios to achieve better performance.

[0039] Subsequently, the optimization results, tolerance calibration intervals, and tolerance limit intervals are used as tolerance configuration optimization results to achieve the best tolerance configuration for more accurate target detection and recognition, which helps to improve the performance and robustness of the multi-target rapid recognition method.

[0040] Performing mapping tolerance adjustment of the feature mapping according to the tolerance configuration optimization result;

[0041] In an embodiment of the present application, after obtaining the tolerance configuration optimization result, based on the tolerance configuration optimization result, the mapping tolerance adjustment of the feature mapping is performed, including adjusting the rules, parameters or configurations of the feature mapping to more flexibly map the size and position of the anchor box to the feature space of the target feature, thereby obtaining the mapping tolerance adjustment result. By adjusting the feature mapping, the tolerance of the mapping is improved to make it more adaptable to targets of different sizes and features, which helps to increase the robustness of target detection to adapt to various scenarios and target changes.

[0042] The rapid target recognition of the target recognition set is completed according to the mapping tolerance adjustment result and the reconstructed anchor frame.

[0043] Furthermore, this step includes:

[0044] Extracting target key features from the target recognition set to generate a target key feature extraction set;

[0045] Performing irregular sub-anchor frame coverage planning on the target key feature set to establish an identification sub-anchor frame, wherein the identification sub-anchor frame includes relative position coordinates with the corresponding anchor frame;

[0046] When performing the reconstructed anchor frame recognition by the target recognition set, auxiliary matching recognition is performed by using the corresponding recognition sub-anchor frame of the reconstructed anchor frame;

[0047] Rapid target recognition is completed based on the auxiliary matching recognition results.

[0048] In a preferred embodiment, first, the key features of the target, including shape, color, texture, etc., are extracted from the target recognition set to help better identify the target, and the extracted key features are organized into a target key feature extraction set. Then, the target key feature set is analyzed to determine which features are associated, that is, which features appear in the same target at the same time. Based on the feature association, irregular sub-anchor frames are planned to ensure that they cover the key features related to the target, and a set of identification sub-anchor frames is established, where each sub-anchor frame includes relative position coordinate information with the corresponding anchor frame, and defines the offset of the sub-anchor frame relative to the original anchor frame to better match the features of the target.

[0049] Subsequently, the target recognition set is used to perform reconstructed anchor frame recognition, each reconstructed anchor frame is matched with the corresponding recognition sub-anchor frame, feature matching is performed, the features of the target are compared with the features of the recognition sub-anchor frame, and it is determined whether the target matches the specific sub-anchor frame, and the auxiliary matching recognition result is generated. Finally, the rapid target recognition is completed through the results of auxiliary matching recognition, including determining the precise location of the target, the bounding box, and the final recognition result.

[0050] Furthermore, the embodiment of the present application also includes:

[0051] Establishing a deep feature layer, wherein the deep feature layer is constructed based on the identification sub-anchor frame;

[0052] When performing anchor frame recognition corresponding to the reconstructed anchor frame, calling the depth feature layer based on the current reconstructed anchor frame, and configuring the corresponding recognition sub-anchor frame;

[0053] Generate a first auxiliary recognition value by identifying the coverage value of the sub-anchor box;

[0054] Generate a second auxiliary recognition value by identifying the feature matching value of the sub-anchor frame;

[0055] generating a matching result of an auxiliary feature layer based on the first auxiliary identification value and the second auxiliary identification value;

[0056] The matching result is used as the auxiliary matching recognition result to complete rapid target recognition.

[0057] In a preferred embodiment, the deep feature layer stores the feature description of each identified sub-anchor frame, including information about the shape, size and features of the sub-anchor frame, as well as the relationship and coverage information between the identified sub-anchor frames. When performing anchor frame recognition corresponding to the reconstructed anchor frame, in order to assist in anchor frame recognition, the deep feature layer is called to obtain detailed information about the sub-anchor frame, and the identified sub-anchor frames corresponding to the current anchor frame are configured to ensure that they match the position and features of the anchor frame.

[0058] Then, the coverage degree of each identified sub-anchor frame and the anchor frame is calculated, that is, the overlapping area between them, and the coverage value of the identified sub-anchor frame is obtained as the first auxiliary recognition value to measure the matching degree between the identified sub-anchor frame and the anchor frame. At the same time, the feature description of the sub-anchor frame and the feature information of the anchor frame are analyzed, and the similarity between the features is calculated to obtain the feature matching value of the identified sub-anchor frame as the second auxiliary recognition value, which is used to measure the feature matching degree of each sub-anchor frame and the anchor frame.

[0059] Next, according to the calculation results of the first auxiliary recognition value and the second auxiliary recognition value, the matching result of each sub-anchor frame is generated, which reflects the matching situation between the sub-anchor frame and the anchor frame, including the coverage degree and the feature matching degree, as the auxiliary matching recognition result to achieve rapid target recognition.

[0060] Furthermore, the embodiment of the present application also includes:

[0061] Perform anchor frame edge positioning evaluation on the current reconstructed anchor frame to determine the positioning fuzzy result;

[0062] Setting N rounds of positioning edges based on the positioning fuzzy results, and identifying the distribution of sub-anchor frames by the relative position coordinates after the positioning of the N rounds of positioning edges is completed;

[0063] Perform coverage recognition based on the distributed recognition sub-anchor frames, and take the distribution result with the largest coverage value in coverage recognition as the terminal result;

[0064] The first auxiliary identification value and the second auxiliary identification value are generated by the terminal result.

[0065] In a preferred embodiment, first, the edge positioning of the current reconstructed anchor frame is evaluated to determine whether the edge of the anchor frame accurately frames the target to determine whether there is a positioning ambiguity problem, thereby obtaining the positioning ambiguity result of each anchor frame. Secondly, based on the obtained positioning ambiguity result, N rounds of positioning edges are set to further refine the positioning of the target, where the number of N positioning rounds is set according to the complexity and needs of the problem. After each round of positioning, the relative position coordinates are used to reposition the identification sub-anchor frame to avoid the target not being accurately identified due to the noise positioning of the large frame.

[0066] Then, according to the distributed recognition sub-anchor frames, the coverage value of each recognition sub-anchor frame and the target is calculated to determine the degree of matching with the target, and the distribution result with the largest coverage value is selected as the terminal result. Subsequently, the first auxiliary recognition value is generated using the coverage value of the recognition sub-anchor frame in the terminal result, and the second auxiliary recognition value is generated using the feature matching value of the recognition sub-anchor frame in the terminal result, thereby reducing positioning errors and improving the performance of the multi-target fast recognition method in complex situations, providing more accurate target information.

[0067] Furthermore, the embodiment of the present application also includes:

[0068] Establishing an update unit, wherein the update unit is an update unit for adaptive threshold anchor boxes and feature maps;

[0069] After completing the identification of any target in the target identification set, transmitting the identification result to the updating unit;

[0070] Performing an update on the adaptive threshold anchor box and the feature map based on the recognition result by the update unit;

[0071] The subsequent target recognition set recognition is realized based on the updated adaptive threshold anchor box and feature map.

[0072] In a preferred embodiment, first, an update unit is established, which is specifically used to manage and execute the update operation of the adaptive threshold anchor frame and feature map. Once any target is successfully identified in the target recognition set, these recognition results are transmitted to the established update unit, and the adaptive threshold anchor frame and feature map are updated according to the latest recognition results through the update unit, that is, the adaptive threshold anchor frame is adjusted and optimized according to the actual situation of target recognition to better adapt to different targets and scenes, and the updated adaptive threshold anchor frame and feature map are obtained. Then, based on the updated adaptive threshold anchor frame and feature map, the recognition of the subsequent target recognition set is realized, ensuring that the constantly changing targets and scenes can be handled more accurately and reliably, and the performance and applicability of the multi-target recognition method are improved.

[0073] Furthermore, the embodiment of the present application also includes:

[0074] Marking the recognition result as trustworthy and configuring a pre-stored threshold of the update unit;

[0075] If the credible identifier cannot satisfy the pre-stored threshold, the corresponding recognition result is added to the fuzzy space;

[0076] If the credible identifier can satisfy the pre-stored threshold, the updating unit updates the adaptive threshold anchor box and the feature map based on the recognition result.

[0077] In a preferred embodiment, after obtaining the recognition result of any target in the target recognition set, the target recognition result is first analyzed based on the target feature to obtain a credible identification of the recognition result, which is a numerical representation. At the same time, a pre-stored threshold is configured for the update unit based on specific needs, and the pre-stored threshold is a numerical representation for determining whether the recognition result is credible.

[0078] Then, the credible identification is compared with the pre-stored threshold. When the credible identification is less than the pre-stored threshold, it is a credible identification that cannot meet the pre-stored threshold. At this time, the untrustworthy recognition result is added to the fuzzy space. When the credible identification is greater than or equal to the pre-stored threshold, it is a credible identification that can meet the pre-stored threshold. At this time, for the credible recognition results, they are used to update the adaptive threshold anchor box and feature map to further optimize the performance of multi-target rapid recognition.

[0079] In summary, the multi-target rapid identification method provided by the embodiment of the present application has the following technical effects:

[0080] A target recognition set is set, wherein the target recognition set is a target data set to be recognized, initial data interaction is performed based on the target recognition set, and feature extraction is performed based on the initial data interaction results, and an anchor box space for feature representation is established to provide a clear target set for the subsequent recognition process and facilitate the subsequent feature selection. An adaptive threshold anchor box is configured based on the spatial distribution of the anchor box space, wherein the spatial distribution includes the extreme value distribution of the anchor box space and the concentrated distribution of the anchor box space to achieve more accurate feature selection. When the target recognition set is recognized, feature selection is performed through the adaptive threshold anchor box, and selection feedback is generated according to the feature selection results of each adaptive threshold anchor box, and the anchor box is reconstructed according to the selection feedback results to recognize the target and provide selection results for the subsequent. A feature mapping of the anchor box size and the target feature is established through the spatial distribution, and test recognition is performed through the feature mapping and the reconstructed anchor box. By mapping the target feature to the anchor box space, the target is quickly recognized. The test recognition result is input into the tolerance configuration network to generate the tolerance configuration optimization result, so that the recognition result is tolerantly configured according to different requirements to balance accuracy and speed. The mapping tolerance adjustment of the feature mapping is performed through the tolerance configuration optimization result to improve the recognition accuracy. According to the mapping tolerance adjustment result, the anchor frame is reconstructed to complete the rapid target recognition of the target recognition set, complete the rapid recognition of the entire multi-target, obtain the final target recognition result, and improve the target recognition accuracy and efficiency.

[0081] Embodiment 2

[0082] Based on the same inventive concept as the multi-target rapid identification method in the aforementioned embodiment, Figure 3 As shown, the embodiment of the present application provides a multi-target rapid identification system, the system comprising:

[0083] The recognition set setting module 11 is used to set a target recognition set, wherein the target recognition set is a target data set to be recognized, perform initial data interaction based on the target recognition set, perform feature extraction based on the initial data interaction result, and establish an anchor frame space for feature representation;

[0084] A threshold anchor box setting module 12, configured to configure an adaptive threshold anchor box based on the spatial distribution of the anchor box space, wherein the spatial distribution includes an extreme value distribution of the anchor box space and a concentrated distribution of the anchor box space;

[0085] An anchor box reconstruction module 13 is used to perform feature box selection through the adaptive threshold anchor box when identifying the target recognition set, generate box selection feedback according to the feature box selection results of each adaptive threshold anchor box, and reconstruct the anchor box according to the box selection feedback results;

[0086] A test recognition module 14 is used to establish a feature map between the anchor box size and the target feature through the spatial distribution, and perform test recognition through the feature map and the reconstructed anchor box;

[0087] The optimization result generation module 15 is used to input the test identification result into the tolerant configuration network to generate a tolerant configuration optimization result;

[0088] A mapping tolerance adjustment module 16, configured to perform mapping tolerance adjustment of the feature mapping according to the tolerance configuration optimization result;

[0089] The fast target recognition module 17 is used to complete the fast target recognition of the target recognition set according to the mapping tolerance adjustment result and the reconstructed anchor frame.

[0090] Furthermore, the optimization result generation module 15 includes the following execution steps:

[0091] After the test recognition result is input into the tolerant configuration network, the data of the test recognition result is analyzed by the analysis subunit to determine the anchor box deviation data set of the selected anchor box and the real anchor box;

[0092] The anchor frame deviation data set is transmitted to the analysis subunit, a tolerance calibration interval and a tolerance limit interval are set by the analysis subunit, and an interval call frequency optimization is performed on the tolerance calibration interval and the tolerance limit interval;

[0093] The optimization result, the tolerance calibration interval and the tolerance limit interval are taken as the tolerance configuration optimization result.

[0094] Furthermore, the rapid target recognition module 17 includes the following execution steps:

[0095] Extracting target key features from the target recognition set to generate a target key feature extraction set;

[0096] Performing irregular sub-anchor frame coverage planning on the target key feature set to establish an identification sub-anchor frame, wherein the identification sub-anchor frame includes relative position coordinates with the corresponding anchor frame;

[0097] When performing the reconstructed anchor frame recognition by the target recognition set, auxiliary matching recognition is performed by using the corresponding recognition sub-anchor frame of the reconstructed anchor frame;

[0098] Rapid target recognition is completed based on the auxiliary matching recognition results.

[0099] Furthermore, the rapid target recognition module 17 further includes the following execution steps:

[0100] Establishing a deep feature layer, wherein the deep feature layer is constructed based on the identification sub-anchor frame;

[0101] When performing anchor frame recognition corresponding to the reconstructed anchor frame, calling the depth feature layer based on the current reconstructed anchor frame, and configuring the corresponding recognition sub-anchor frame;

[0102] Generate a first auxiliary recognition value by identifying the coverage value of the sub-anchor box;

[0103] Generate a second auxiliary recognition value by identifying the feature matching value of the sub-anchor frame;

[0104] generating a matching result of an auxiliary feature layer based on the first auxiliary identification value and the second auxiliary identification value;

[0105] The matching result is used as the auxiliary matching recognition result to complete rapid target recognition.

[0106] Furthermore, the rapid target recognition module 17 further includes the following execution steps:

[0107] Perform anchor frame edge positioning evaluation on the current reconstructed anchor frame to determine the positioning fuzzy result;

[0108] Setting N rounds of positioning edges based on the positioning fuzzy results, and identifying the distribution of sub-anchor frames by the relative position coordinates after the positioning of the N rounds of positioning edges is completed;

[0109] Perform coverage recognition based on the distributed recognition sub-anchor frames, and take the distribution result with the largest coverage value in coverage recognition as the terminal result;

[0110] The first auxiliary identification value and the second auxiliary identification value are generated by the terminal result.

[0111] Furthermore, the embodiment of the present application also includes an anchor frame mapping update module, which includes the following execution steps:

[0112] Establishing an update unit, wherein the update unit is an update unit for adaptive threshold anchor boxes and feature maps;

[0113] After completing the identification of any target in the target identification set, transmitting the identification result to the updating unit;

[0114] Performing an update on the adaptive threshold anchor box and the feature map based on the recognition result by the update unit;

[0115] The subsequent target recognition set recognition is realized based on the updated adaptive threshold anchor box and feature map.

[0116] Furthermore, the anchor box mapping update module further includes the following execution steps:

[0117] Marking the recognition result as trustworthy and configuring a pre-stored threshold of the update unit;

[0118] If the credible identifier cannot satisfy the pre-stored threshold, the corresponding recognition result is added to the fuzzy space;

[0119] If the credible identifier can satisfy the pre-stored threshold, the updating unit updates the adaptive threshold anchor box and the feature map based on the recognition result.

[0120] Any step of the method described above can be stored as a computer instruction or program in an unlimited computer memory, and can be called and recognized by an unlimited computer processor to implement any method in the embodiments of the present application, without any unnecessary restrictions.

[0121] Furthermore, the first or second mentioned above may not only represent an order relationship, but may also represent a specific concept, and / or refer to multiple elements that can be selected individually or in full. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A multi-target rapid identification method, characterized in that: The method comprises: A target recognition set is set, wherein the target recognition set is a target data set to be recognized, initial data interaction is performed based on the target recognition set, and feature extraction is performed according to the initial data interaction result to establish an anchor frame space for feature representation, wherein the initial data interaction is to preprocess the target data in the target recognition set, and the data interaction includes image enhancement, denoising, and size standardization to obtain the initial data interaction result; configuring an adaptive threshold anchor frame based on a spatial distribution of the anchor frame space, wherein the spatial distribution includes an extreme value distribution of the anchor frame space and a concentrated distribution of the anchor frame space; When performing recognition of the target recognition set, feature selection is performed through the adaptive threshold anchor frame, and selection feedback is generated according to the feature selection results of each adaptive threshold anchor frame, and the anchor frame is reconstructed according to the selection feedback results; Establishing a feature map between the anchor box size and the target feature through the spatial distribution, and performing test recognition through the feature map and the reconstructed anchor box; Input the test identification results into the tolerant configuration network to generate the tolerant configuration optimization results; Performing mapping tolerance adjustment of the feature mapping according to the tolerance configuration optimization result; The rapid target recognition of the target recognition set is completed according to the mapping tolerance adjustment result and the reconstructed anchor frame.

2. The method according to claim 1, characterized in that The method further comprises: After the test recognition result is input into the tolerant configuration network, the data of the test recognition result is analyzed by the analysis subunit to determine the anchor box deviation data set of the selected anchor box and the real anchor box; The anchor frame deviation data set is transmitted to the analysis subunit, a tolerance calibration interval and a tolerance limit interval are set by the analysis subunit, and an interval call frequency optimization is performed on the tolerance calibration interval and the tolerance limit interval; The optimization result, the tolerance calibration interval and the tolerance limit interval are taken as the tolerance configuration optimization result.

3. The method according to claim 1, characterized in that The method further comprises: Extracting target key features from the target recognition set to generate a target key feature extraction set; Performing irregular sub-anchor frame coverage planning on the target key feature set to establish an identification sub-anchor frame, wherein the identification sub-anchor frame includes relative position coordinates with the corresponding anchor frame; When performing the reconstructed anchor frame recognition by the target recognition set, auxiliary matching recognition is performed by using the corresponding recognition sub-anchor frame of the reconstructed anchor frame; Rapid target recognition is completed based on the auxiliary matching recognition results.

4. The method according to claim 3, characterized in that The method further comprises: Establishing a deep feature layer, wherein the deep feature layer is constructed based on the identification sub-anchor frame; When performing anchor frame recognition corresponding to the reconstructed anchor frame, calling the depth feature layer based on the current reconstructed anchor frame, and configuring the corresponding recognition sub-anchor frame; Generate a first auxiliary recognition value by identifying the coverage value of the sub-anchor box; Generate a second auxiliary recognition value by identifying the feature matching value of the sub-anchor frame; generating a matching result of an auxiliary feature layer based on the first auxiliary identification value and the second auxiliary identification value; The matching result is used as the auxiliary matching recognition result to complete rapid target recognition.

5. The method according to claim 4, characterized in that The method further comprises: Perform anchor frame edge positioning evaluation on the current reconstructed anchor frame to determine the positioning fuzzy result; Setting N rounds of positioning edges based on the positioning fuzzy results, and identifying the distribution of sub-anchor frames by the relative position coordinates after the positioning of the N rounds of positioning edges is completed; Perform coverage recognition based on the distributed recognition sub-anchor frames, and take the distribution result with the largest coverage value in coverage recognition as the terminal result; The first auxiliary identification value and the second auxiliary identification value are generated by the terminal result.

6. The method according to claim 1, characterized in that The method further comprises: Establishing an update unit, wherein the update unit is an update unit for adaptive threshold anchor boxes and feature maps; After completing the identification of any target in the target identification set, transmitting the identification result to the updating unit; Performing an update on the adaptive threshold anchor box and the feature map based on the recognition result by the update unit; The subsequent target recognition set recognition is realized based on the updated adaptive threshold anchor box and feature map.

7. The method according to claim 6, characterized in that The method further comprises: Marking the recognition result as trustworthy and configuring a pre-stored threshold of the update unit; If the credible identifier cannot satisfy the pre-stored threshold, the corresponding recognition result is added to the fuzzy space; If the credible identifier can satisfy the pre-stored threshold, the updating unit updates the adaptive threshold anchor box and the feature map based on the recognition result.

8. A multi-target rapid identification system, characterized in that: The system comprises: An identification set setting module, which is used to set a target identification set, wherein the target identification set is a target data set to be identified, performs initial data interaction based on the target identification set, and performs feature extraction according to the initial data interaction result to establish an anchor frame space for feature representation, wherein the initial data interaction is to preprocess the target data in the target identification set, and the data interaction includes image enhancement, denoising, and size standardization to obtain the initial data interaction result; A threshold anchor box setting module, the threshold anchor box setting module is used to configure an adaptive threshold anchor box based on the spatial distribution of the anchor box space, wherein the spatial distribution includes an extreme value distribution of the anchor box space and a concentrated distribution of the anchor box space; An anchor box reconstruction module, wherein the anchor box reconstruction module is used to perform feature box selection through the adaptive threshold anchor box when performing recognition of the target recognition set, generate box selection feedback according to the feature box selection results of each adaptive threshold anchor box, and reconstruct the anchor box according to the box selection feedback results; A test recognition module, the test recognition module is used to establish a feature map between the anchor box size and the target feature through the spatial distribution, and perform test recognition through the feature map and the reconstructed anchor box; An optimization result generation module, wherein the optimization result generation module is used to input the test recognition result into the tolerant configuration network to generate a tolerant configuration optimization result; A mapping tolerance adjustment module, the mapping tolerance adjustment module is used to perform mapping tolerance adjustment of the feature mapping according to the tolerance configuration optimization result; A fast target recognition module is used to complete the fast target recognition of the target recognition set according to the mapping tolerance adjustment result and the reconstructed anchor frame.

Images