Camouflage target detection method and device, equipment and storage medium

By extracting and enhancing the feature of the camouflage target image, the problem of poor detection of camouflage targets in the prior art is solved, more accurate camouflage target recognition is achieved, and the accuracy of the recognition of camouflage target objects is improved.

CN119091176APending Publication Date: 2024-12-06TCL TECHNOLOGY GROUP CORPORATION
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202310657819.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2023-06-05
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

The existing deep object detection model cannot accurately detect the object of the camouflage target in the camouflage target scenario, resulting in a large amount of feature information of the camouflage target, the difference between the background and the camouflage camouflage texture is reduced, and the detection effect is poor.

Method used

By obtaining the camouflage target image to be detected, the position and edge features are obtained through feature extraction, and horizontal and vertical features are enhanced, including node vectorization, information transmission, pyramid pooling, convolutional operations and cross-domain information aggregation. Generate enhanced positional and edge features, and finally classify mapping to achieve camouflage target recognition.

Benefits of technology

The recognition accuracy of camouflage target objects is improved, the detection ability of camouflage target images is enhanced, and the target user location and edge features can be detected more accurately.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119091176A_ABST
    Figure CN119091176A_ABST
Patent Text Reader

Abstract

The invention provides a camouflage target detection method and device, equipment and a storage medium. The method comprises the steps that a to-be-detected camouflage target image is acquired; performing feature extraction on the camouflage target image to obtain a position feature and an edge feature; performing feature enhancement on the position feature and the edge feature to obtain a processed position feature and a processed edge feature; and performing classification mapping according to the processed position features and the processed edge features to obtain a camouflage target recognition result of the camouflage target image. According to the technical scheme, edge information enhancement is carried out on the camouflage target image, so that the recognition accuracy of the camouflage target object is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method, device, equipment and storage medium for detecting a camouflaged target. Background Art

[0002] At present, with the rapid development of VR field and target detection technology, various enterprises have launched VR applications based on target detection to provide different content scenes, so that users can get rid of venue restrictions and experience different entertainment scenes anytime and anywhere by using VR devices. However, in scenes such as VR gun battles that include camouflaged targets, since users often wear camouflage-colored clothing and props when playing games, which have strong concealment, the existing deep target detection model does not fully explore the correlation between features, and the feature extraction process will cause a large amount of feature information of the camouflaged target to be lost, and the difference between the background and the camouflage texture is further reduced, resulting in the detection effect on such camouflaged targets is not outstanding, and the target user position and edge features cannot be accurately detected. Summary of the invention

[0003] The embodiments of the present application provide a disguised target detection method, apparatus, device and storage medium, which aim to solve the technical problem in the prior art that a disguised target object cannot be accurately detected in a disguised target recognition scenario.

[0004] On the one hand, an embodiment of the present application provides a disguised target detection method, the disguised target detection method comprising the following steps:

[0005] Acquire a camouflaged target image to be detected;

[0006] Extracting features from the camouflaged target image to obtain position features and edge features;

[0007] Performing feature enhancement on the position feature and the edge feature to obtain a processed position feature and a processed edge feature;

[0008] Classification mapping is performed according to the processed position features and the processed edge features to obtain a disguised target recognition result of the disguised target image.

[0009] In a possible implementation of the present application, the step of performing feature enhancement on the position feature and the edge feature to obtain the processed position feature and the processed edge feature includes:

[0010] Performing lateral feature enhancement on the position feature and the edge feature to obtain enhanced position feature and enhanced edge feature;

[0011] The enhanced position feature and the enhanced edge feature are subjected to longitudinal feature enhancement to obtain a processed position feature and a processed edge feature.

[0012] In a possible implementation of the present application, the step of performing lateral feature enhancement on the position feature and the edge feature to obtain an enhanced position feature and an enhanced edge feature includes:

[0013] Performing node vectorization on the position feature and the edge feature to obtain a position feature domain and an edge feature domain;

[0014] Transmitting node information of the location feature domain and the edge feature domain to obtain location domain edge information, edge domain edge information and cross-domain edge information;

[0015] Node information is updated according to the location domain side information, the edge domain side information and the cross-domain side information to obtain enhanced location features and enhanced edge features.

[0016] In a possible implementation of the present application, performing node vectorization on the position feature and the edge feature to obtain a position feature domain and an edge feature domain includes:

[0017] Performing pyramid pooling processing on the position features and the edge features to obtain pooled position features and pooled edge features;

[0018] Performing interpolation processing on the pooled position features and the pooled edge features according to a preset interpolation layer to obtain a target position node and a target edge node;

[0019] Each of the target position nodes is aggregated to obtain a position feature domain of the disguised target image, and each of the target edge nodes is aggregated to obtain an edge feature domain of the disguised target image.

[0020] In a possible implementation of the present application, the node information is transmitted to the location feature domain and the edge feature domain to obtain location domain side information, edge domain side information and cross-domain side information, including:

[0021] Performing a convolution operation on a first position node and a second position node in the position feature domain to obtain position domain edge information;

[0022] Performing a convolution operation on the first edge node and the second edge node in the edge feature domain to obtain edge domain edge information;

[0023] A dynamic convolution kernel associated with the first position node and / or the first edge node is generated based on a preset learner, and a convolution operation is performed according to the dynamic convolution kernel, the first position node and the first edge node to obtain cross-domain edge information.

[0024] In a possible implementation of the present application, updating the node information according to the location domain side information, the edge domain side information, and the cross-domain side information to obtain enhanced location features and enhanced edge features includes:

[0025] Aggregating the same-domain information of the location domain side information and the corresponding target location node in the location feature domain according to a preset activation function to obtain a location aggregation feature;

[0026] Aggregating the same-domain information of the edge domain side information and the corresponding target edge nodes in the edge feature domain according to a preset activation function to obtain edge aggregation features;

[0027] Obtain a preset adapter to generate a cross-domain convolution kernel that associates the cross-domain side information with a position feature domain or an edge feature domain, and perform cross-domain information aggregation according to the cross-domain convolution kernel, the cross-domain side information, and a target position node or a target edge node corresponding to the cross-domain side information to obtain a cross-domain aggregation feature;

[0028] An enhanced position feature and an enhanced edge feature are generated according to the position aggregation feature, the edge aggregation feature and the cross-domain aggregation feature.

[0029] In a possible implementation of the present application, generating enhanced location features and enhanced edge features according to the location aggregation features, the edge aggregation features, and the cross-domain aggregation features includes:

[0030] Obtaining a historical location node corresponding to the cross-domain aggregation feature, inputting the cross-domain aggregation feature, the edge aggregation feature, and the historical location node into a preset application gated loop unit for state update, and obtaining an iterative enhanced location feature;

[0031] The historical edge node corresponding to the cross-domain aggregation feature is obtained, and the cross-domain aggregation feature, the location aggregation feature and the historical edge node are input into a preset application gated loop unit for state update to obtain an iterated enhanced edge feature.

[0032] In a possible implementation of the present application, the step of performing longitudinal feature enhancement on the enhanced position feature and the enhanced edge feature to obtain a processed position feature and a processed edge feature includes:

[0033] Acquire historical position information corresponding to the enhanced position feature and historical edge information corresponding to the enhanced edge feature;

[0034] Calculate node similarity based on the historical position information and the transposed position information corresponding to the historical position information to obtain a first graph adjacency matrix;

[0035] Calculating node similarity between the historical edge information and the transposed edge information corresponding to the historical edge information to obtain a second graph adjacency matrix;

[0036] Performing graph reasoning according to the historical position information and the first graph adjacency matrix to obtain reasoned position information, and performing graph reasoning according to the historical edge information and the second graph adjacency matrix to obtain reasoned edge information;

[0037] The inferred position information and the enhanced position feature are feature-merged to obtain a processed position feature, and the inferred edge information and the enhanced edge feature are feature-merged to obtain a processed edge feature.

[0038] In a possible implementation of the present application, the camouflaged target image is subjected to feature extraction to obtain position features and edge features, including:

[0039] Inputting the disguised target image into a preset disguised target detection model, wherein the disguised target detection model includes a channel fusion module and a target attention module;

[0040] Acquiring image features in the disguised target image based on the channel fusion module, and weighting the image features according to the channel feature vectors corresponding to the image features to obtain weighted image features;

[0041] Based on the target attention module, multi-dimensional information detection is performed on the weighted image features to obtain position features and edge features of the disguised target image.

[0042] On the other hand, the present application provides a disguised target detection device, the disguised target detection device comprising:

[0043] An image acquisition module is configured to acquire an image of a camouflaged target to be detected;

[0044] A feature extraction module is configured to extract features of the camouflaged target image to obtain position features and edge features;

[0045] A feature enhancement module is configured to perform feature enhancement on the position feature and the edge feature to obtain a processed position feature and a processed edge feature, wherein the feature enhancement operation includes horizontal feature enhancement and vertical feature enhancement;

[0046] The feature classification module is configured to perform classification mapping according to the processed position features and the processed edge features to obtain a disguised output result and an edge output result of the disguised target image.

[0047] On the other hand, the present application also provides a disguised target detection device, the disguised target detection device comprising:

[0048] one or more processors;

[0049] Memory; and

[0050] One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the processor to implement the steps of the camouflaged target detection method.

[0051] On the other hand, the present application also provides a computer-readable storage medium on which a computer program is stored, and the computer program is loaded by a processor to execute the steps in the camouflaged target detection method.

[0052] In the present application, a camouflaged target image to be detected is obtained; the camouflaged target image is subjected to feature extraction to obtain position features and edge features; the position features and edge features are subjected to feature enhancement to obtain processed position features and processed edge features, wherein the feature enhancement operation includes horizontal feature enhancement and vertical feature enhancement; classification mapping is performed based on the processed position features and processed edge features to obtain a camouflaged target recognition result of the camouflaged target image. By enhancing the edge information of the camouflaged target image, the recognition accuracy of the camouflaged target object is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0054] Figure 1 A schematic diagram of a scene of a camouflaged target detection method according to an embodiment of the present application;

[0055] Figure 2 A schematic diagram of a flow chart of an embodiment of a camouflaged target detection method in an embodiment of the present application;

[0056] Figure 3 A schematic diagram of a model of a feature extraction network in a disguised target detection model provided in an embodiment of the present application;

[0057] Figure 4 A flowchart of an embodiment of node information transmission in a disguised target detection method provided in an embodiment of the present application;

[0058] Figure 5 A flow chart of an embodiment of performing lateral feature enhancement on position features and edge features in a camouflaged target detection method provided in an embodiment of the present application;

[0059] Figure 6 A schematic diagram of a flow chart of an embodiment of performing longitudinal feature enhancement in a camouflaged target detection method provided in an embodiment of the present application;

[0060] Figure 7 A schematic diagram of the structure of an embodiment of a camouflaged target detection device provided in an embodiment of the present application;

[0061] Figure 8 A schematic diagram of the structure of an embodiment of a camouflaged target detection device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0062] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. 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 those skilled in the art without creative work are within the scope of protection of the present invention.

[0063] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, which are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first" and "second" are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0064] In this application, the word "exemplary" is used to mean "serving as an example, illustration, or illustration." Any embodiment described in this application as "exemplary" is not necessarily to be construed as being preferred or advantageous over other embodiments. The following description is given to enable any person skilled in the art to implement and use the invention. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can recognize that the invention can be implemented without using these specific details. In other instances, well-known structures and processes are not elaborated in detail to avoid obscuring the description of the invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed in this application.

[0065] At present, with the rapid development of VR field and target detection technology, various enterprises have launched VR applications based on target detection to provide different content scenes, so that users can get rid of venue restrictions and experience different entertainment scenes anytime and anywhere by using VR devices. However, in scenes such as VR gun battles that include camouflaged targets, since users often wear camouflage-colored clothing and props when playing games, which have strong concealment, the existing deep target detection model does not fully explore the correlation between features, and the feature extraction process will cause a large amount of feature information of the camouflaged target to be lost, and the difference between the background and the camouflage texture is further reduced, resulting in the detection effect on such camouflaged targets is not outstanding, and the target user position and edge features cannot be accurately detected.

[0066] Based on this, the present application proposes a disguised target detection method, device, equipment and computer-readable storage medium to solve the technical problem in the prior art that the disguised target object cannot be accurately detected in a disguised target recognition scenario.

[0067] The disguised target detection method in the embodiment of the present invention is applied to a disguised target detection device, which is arranged in a disguised target detection device. The disguised target detection device is provided with one or more processors, memories, and one or more applications, wherein the one or more applications are stored in the memories and configured to be executed by the processor to implement the disguised target detection method; wherein the disguised target detection device can be an intelligent terminal, such as a mobile phone, a tablet computer, an all-in-one VR machine, and a smart computer; optionally, the disguised target detection device can also be a server, or a service cluster composed of multiple servers.

[0068] like Figure 1 As shown, Figure 1This is a scene diagram of the disguised target detection method in an embodiment of the present application. The disguised target detection scene in the embodiment of the present invention includes a disguised target detection device 100 (a disguised target detection device is integrated in the disguised target detection device 100), and a computer-readable storage medium corresponding to the disguised target detection method is running in the disguised target detection device 100 to execute the steps of the disguised target detection method.

[0069] Understandably, Figure 1 The disguised target detection device in the disguised target detection method scenario shown, or the device included in the disguised target detection device, does not constitute a limitation on the embodiments of the present invention, that is, the number of devices and types of devices of the disguised target detection device included in the scenario of the disguised target detection method, or the number and types of devices included in each device do not affect the overall implementation of the technical solution in the embodiments of the present invention, and can all be regarded as equivalent replacements or derivatives of the technical solution claimed to be protected in the embodiments of the present invention.

[0070] The disguised target detection device 100 in the embodiment of the present invention is mainly used to: obtain a disguised target image to be detected; perform feature extraction on the disguised target image to obtain position features and edge features; perform feature enhancement on the position features and the edge features to obtain processed position features and processed edge features, wherein the feature enhancement operation includes horizontal feature enhancement and vertical feature enhancement; perform classification mapping according to the processed position features and the processed edge features to obtain a disguised target recognition result of the disguised target image.

[0071] The disguised target detection device 100 in the embodiment of the present invention can be an independent disguised target detection device, such as a smart terminal such as a mobile phone, a tablet computer, a VR all-in-one machine, a server, and a smart computer, or it can be a disguised target detection network or a disguised target detection cluster composed of multiple disguised target detection devices.

[0072] The embodiments of the present application provide a camouflaged target detection method, apparatus, device, and computer-readable storage medium, which are described in detail below.

[0073] It can be understood by those skilled in the art that Figure 1 The application environment shown in is only one of the application scenarios related to the present application solution and does not constitute a limitation on the application scenario of the present application solution. Other application environments may also include Figure 1 More or fewer disguised target detection devices, or disguised target detection network connection relationships, such as Figure 1Only one disguised target detection device is shown in the figure. It can be understood that the scenario of the disguised target detection method can also include one or more disguised target detection devices, which is not specifically limited here. The disguised target detection device 100 can also include a memory for storing a disguised target image and other data.

[0074] It should be noted that Figure 1 The scene diagram of the disguised target detection method shown is only an example. The scene of the disguised target detection method described in the embodiment of the present invention is to more clearly illustrate the technical solution of the embodiment of the present invention, and does not constitute a limitation on the technical solution provided by the embodiment of the present invention.

[0075] Based on the scenario of the foregoing disguised target detection method, various embodiments of the disguised target detection method disclosed in the present invention are proposed.

[0076] like Figure 2 As shown, Figure 2 This is a flow chart of an embodiment of a disguised target detection method in an embodiment of the present application. The disguised target detection method includes the following steps 201 to 204:

[0077] 201. Obtain a camouflaged target image to be detected;

[0078] The disguised target detection method in this embodiment is applied to a disguised target detection device, and the type and number of the disguised target detection device are not specifically limited, that is, the disguised target detection device can be one or more intelligent terminals. In a specific embodiment, the disguised target detection device is a smart computer.

[0079] Specifically, the camouflaged target detection device is configured to pre-set a camouflaged target detection model, and extract features of the camouflaged target image to be detected through a multi-task backbone network in the camouflaged target detection model to obtain position features and edge features of the camouflaged target image, and enhance the position features and edge features through an edge feature enhancement module, and classify and map the enhanced position features and enhanced edge features to obtain a camouflaged target recognition result.

[0080] Specifically, during the operation of the disguised target detection device, the disguised target detection request is responded to. The disguised target detection request is an operation request to drive the disguised target detection device to perform disguised target recognition on the disguised target image. The triggering request of the disguised target detection request is not specifically limited here, that is, the disguised target detection request can be triggered when the user starts an application such as VR gun battle with a disguised target image.

[0081] Specifically, after receiving the disguised target detection request, the disguised target detection device obtains a disguised target image to be detected, wherein the disguised target image is an image in which the disguised object to be detected has a high similarity with the image background in the image, resulting in blurred edges between the object and the background. For example, in a specific embodiment, the disguised target image is an image of a game character wearing camouflage clothes in a VR game application scenario.

[0082] After acquiring the disguised target image to be detected, the disguised target detection device inputs the disguised target image into a preset disguised target detection model to perform disguised target detection and outputs a disguised target recognition result.

[0083] 202. Extracting features from the camouflaged target image to obtain position features and edge features;

[0084] Specifically, the disguised target detection device pre-sets a disguised target detection model for performing disguised target recognition on a disguised target image, and the disguised target detection model includes a feature extraction network, an edge feature enhancement module, and a classification output module. In a specific embodiment, the feature extraction network is a multi-task backbone network including an improved target residual module. For example, the feature extraction network can be based on Θ MTFE A multi-branch FCN network based on ResNet with parameters.

[0085] Specifically, after acquiring the disguised target image, the disguised target detection device inputs the disguised target image into a feature extraction network in the disguised target detection model for feature extraction to obtain the disguised features and edge features of the disguised target image.

[0086] like Figure 3 As shown, Figure 3 A model schematic diagram of a feature extraction network in a camouflaged target detection model provided in an embodiment of the present application. Specifically, the camouflaged target detection device improves the residual module in the feature extraction network through a channel fusion module and a target attention module, and adds a channel fusion module and a target attention module to the residual module to generate an improved target residual module. In a specific embodiment, the channel fusion module is a SE (Squeeze and Excitation) structure, which can effectively fuse information between different channels of the input feature map and enhance the sensitivity of the algorithm to channel information. The target attention module is a Triplet Attention structure.

[0087] Specifically, the disguised target detection device obtains the image of the disguised target image through the channel fusion module, and weights the image features according to the channel feature vectors corresponding to the image features to obtain weighted image features, and performs multi-dimensional information fusion on the weighted image features through the three-branch structure in the target attention module, thereby capturing the feature information of the disguised target position in the disguised target image and obtaining the position features and edge features of the disguised target image.

[0088] Optionally, in a specific embodiment, after obtaining the weighted image features, the target attention module calculates the feature connections between the weighted image features in the high dimension and the wide dimension, the high dimension and the channel dimension, and the wide dimension and the channel dimension through the high dimension, the wide dimension, and the channel dimension to obtain the position features and edge features of the disguised target image. Specifically, the feature representation of the position features and edge features of the disguised target image is as follows:

[0089] F C = MTFE (I; θ MTFE ), E = MTFE (I; θ MTFE )

[0090] in, is a position feature with h*w spatial resolution and c channels; It is a position feature with h*w spatial resolution and c channels; I is the camouflaged target image.

[0091] 203. Perform feature enhancement on the position feature and the edge feature to obtain a processed position feature and a processed edge feature;

[0092] Specifically, after the disguised target detection device extracts features of the disguised target image through the target residual module in the feature extraction network, it also enhances the position features and the edge features to obtain processed position features and processed edge features.

[0093] Specifically, since the camouflaged object to be detected in the camouflaged target image has a high similarity with the image background in the image, the edge of the object and the background are blurred. Therefore, after obtaining the position features and edge features of the camouflaged target image, the camouflaged target detection device also performs horizontal feature enhancement and vertical feature enhancement on the position features and edge features, thereby making full use of the complementary information of the time dimension and the space dimension in the position features and edge features, realizing information enhancement of the position features and edge features, and obtaining processed position features and processed edge features. Among them, horizontal feature enhancement is a feature extraction method that uses the complementary relationship between the position features and edge features in the camouflaged target image to extract features; vertical feature enhancement is a feature extraction method that uses the information of the previous moment to supplement the information of the current moment.

[0094] Specifically, the camouflaged target detection device inputs the position features and edge features into the edge information enhancement module for feature enhancement, and drives the 1*1 convolution layer in the edge information enhancement module to perform feature dimensionality reduction on the position features and edge features to obtain the reduced dimensionality position features and edge features.

[0095] The disguised target detection device obtains the position features after dimensionality reduction and the edge features after dimensionality reduction, and also performs node vectorization on the position features and edge features to obtain the position feature domain and edge feature domain of the disguised target image. The position feature domain is a node feature domain including each target position node of the disguised target image. The disguise feature domain is a node feature domain including each target edge node of the disguised target image.

[0096] Specifically, the camouflaged target detection device performs pyramid pooling processing on the reduced dimension position features and the reduced dimension edge features to obtain pooled position features and pooled edge features, and interpolates the pooled position features and the pooled edge features through a preset interpolation layer to obtain target position nodes and target edge nodes. Specifically, the expressions of the target position nodes and the target edge nodes are as follows:

[0097]

[0098]

[0099] Among them, R(·) is the interpolation operation and P(·) is the pyramid pooling operation.

[0100] Specifically, after acquiring the target position node, the disguised target detection device aggregates each target position node to generate a position feature domain of the position target image; and aggregates each target edge node to generate an edge feature domain of the disguised target image.

[0101] Specifically, the camouflaged target detection device also transfers node information between the position feature domain and the edge feature domain to obtain position domain side information, edge domain side information and cross-domain side information. Among them, the position domain side information is a multi-scale relationship between target position nodes in the position feature domain. The edge domain side information is a multi-scale relationship between target edge nodes in the edge feature domain. The cross-domain side information is complementary information between nodes between the position feature domain and the edge feature domain.

[0102] Specifically, after obtaining the position domain side information, edge domain side information and cross-domain side information, the disguised target detection device updates the node information according to the position domain side information, edge domain side information and cross-domain side information to obtain enhanced position features and enhanced edge features.

[0103] Specifically, after performing horizontal feature enhancement to obtain enhanced camouflage information and enhanced edge features, the camouflaged target detection device also makes full use of the spatiotemporal information of the camouflaged target image to perform further contextual learning of the features, that is, performing vertical feature enhancement on the enhanced camouflage information and enhanced edge features to obtain processed position features and processed edge features.

[0104] 204. Perform classification mapping according to the processed position features and the processed edge features to obtain a disguised target recognition result of the disguised target image.

[0105] Optionally, before responding to the disguised target detection request, the disguised target detection device also pre-trains the disguised target detection model based on the disguised training image. That is, the disguised target detection device extracts the training position features and training edge features of the disguised training image through the feature extraction network of the disguised target detection model, and performs feature enhancement on the training position features and training edge features to obtain processed training position features and processed training edge features, and inputs the processed training position features and processed training edge features into the classification output module for classification mapping to obtain training position output features and training edge output features, and calculates the model loss based on the training position output features and training edge output features, and trains the disguised target detection model through the model loss to obtain the trained disguised target detection model.

[0106] Specifically, after obtaining the training position output features and the training edge output features, the camouflaged target detection device uses bilinear interpolation to upsample the output image to the original size to calculate the loss, and uses cross entropy loss for the training position output features and the training edge output features. Specifically, the calculation formula is as follows:

[0107]

[0108] Among them, L is the model loss, G C and G Erefers to the true value label, γ is the weight, is the training position loss, is the training edge loss, C is the training position feature, and E is the training edge feature.

[0109] Specifically, after obtaining the trained disguised target detection model, the disguised target detection device inputs the processed position features and the processed edge features into the classification output module in the disguised target detection model for classification mapping to obtain the disguised target recognition result of the disguised target image. The classification output module is a 1*1 convolution layer. The disguised target recognition result includes a target position result and a target edge result.

[0110] In this embodiment, the camouflaged target detection device obtains a camouflaged target image to be detected; extracts features from the camouflaged target image to obtain position features and edge features; enhances the position features and edge features to obtain processed position features and processed edge features; and performs classification mapping based on the processed position features and processed edge features to obtain a camouflaged target recognition result of the camouflaged target image. By enhancing the edge information of the camouflaged target image, the recognition accuracy of the camouflaged target object is improved.

[0111] like Figure 4 As shown, Figure 4 A flowchart of an embodiment of node information transmission in a disguised target detection method provided in an embodiment of the present application, specifically, including steps 301 to 303:

[0112] 301. Perform a convolution operation on a first position node and a second position node in the position feature domain to obtain position domain edge information;

[0113] Based on the above embodiments, in this embodiment, during the process of performing lateral feature enhancement on position features and edge features, the camouflaged target detection device node-izes the position features and edge features to obtain position feature and edge feature domains, and then performs node information transmission on the position feature domain and the edge feature domain to obtain position domain side information, edge domain side information, and cross-domain side information.

[0114] Specifically, the camouflaged target detection device performs same-domain edge vectorization on the position feature domain to obtain position domain edge information, and connects each target position node in the position feature domain through the position domain edge information.

[0115] Specifically, the disguised target detection device obtains the first position node and the second position node in the position feature domain, and performs a convolution operation on the first position node and the second position node through a preset relationship function to extract the multi-scale relationship between the first position node and the second position node to obtain the position domain edge information. Specifically, the extraction method is as follows:

[0116]

[0117] in, is the position domain side information, g(·,·) is the preset relationship function, Conv(·) represents the convolution operation, is the first position node, The second location node.

[0118] 302. Perform a convolution operation on the first edge node and the second edge node in the edge feature domain to obtain edge domain edge information;

[0119] Specifically, the disguised target detection device performs same-domain edge vectorization on the edge feature domain to obtain edge domain edge information, and connects each target edge node in the edge feature domain through the edge domain edge information.

[0120] Specifically, the disguised target detection device obtains the first edge node and the second edge node in the edge feature domain, and performs a convolution operation on the first edge node and the second edge node through a preset relationship function to extract the multi-scale relationship between the first edge node and the second edge node to obtain edge domain edge information. Specifically, the extraction method is as follows:

[0121]

[0122] in, is the edge information of the edge domain, g(·,·) is the preset relationship function, Conv(·) represents the convolution operation, is the first edge node, The second edge node.

[0123] 303. Generate a dynamic convolution kernel associated with the first position node and / or the first edge node based on a preset learner, and perform a convolution operation on the dynamic convolution kernel, the first position node, and the first edge node to obtain cross-domain edge information.

[0124] Specifically, the disguised target detection device further obtains complementary information in different domain nodes, that is, generates a dynamic convolution kernel associated with the first position node and / or the first edge node based on a preset learner, and performs a convolution operation according to the dynamic convolution kernel, the first position node and the first edge node to obtain cross-domain edge information. The preset learner is a one-time learner that can predict the dynamic convolution kernel from a single sample.

[0125] Specifically, the disguised target detection device obtains the dynamic convolution kernel of the first position node through a preset learner, and performs a convolution operation on the first edge node through the dynamic convolution kernel of the first position node to obtain a position cross-domain expression of the first position node. Specifically, the calculation method is as follows:

[0126]

[0127] Among them, * is the convolution operation,, i is the dynamic convolution kernel, It is a preset learner.

[0128] Specifically, after obtaining the position cross-domain expression, the disguised target detection device performs a convolution operation on the position cross-domain expression and the first edge node to obtain the cross-domain edge information from the first position node in the position feature domain to the first edge node in the edge feature domain. Specifically, the calculation method is as follows:

[0129]

[0130] Similarly, the edge cross-domain expression of the first edge node can be obtained:

[0131]

[0132] The cross-domain edge information expression from the first edge node in the edge feature domain to the first position node in the position feature domain is as follows:

[0133]

[0134] Specifically, after obtaining the position domain side information, edge domain side information and cross-domain side information, the camouflaged target detection device performs transverse feature enhancement on the position features and edge features according to the position domain side information, edge domain side information and cross-domain side information to obtain enhanced position features and enhanced edge features.

[0135] In this embodiment, the disguised target detection device obtains the position domain side information by performing convolution operation according to the first position node and the second position node in the position feature domain; performs convolution operation according to the first edge node and the second edge node in the edge feature domain to obtain the edge domain side information; generates the first position node and / or the first edge node associated dynamic convolution kernel based on the preset learner, and performs convolution operation according to the dynamic convolution kernel, the first position node and the first edge node to obtain cross-domain side information. The complementary information in the position feature and edge feature of the disguised target image is obtained, and the position feature and edge feature are enhanced horizontally through the complementary information, so as to improve the accuracy of the disguised target detection of the disguised target image.

[0136] like Figure 5 As shown, Figure 5 A flow chart of an embodiment of performing lateral feature enhancement on position features and edge features in a camouflaged target detection method provided in an embodiment of the present application, specifically, comprising steps 401 to 404:

[0137] 401. Aggregate the same-domain information of the location domain side information and the corresponding target location nodes in the location feature domain according to a preset activation function to obtain a location aggregation feature;

[0138] Based on the above embodiments, in this embodiment, after obtaining the position domain side information, edge domain side information and cross-domain side information, the camouflaged target detection device performs lateral feature enhancement according to the position domain side information, edge domain side information and cross-domain side information to obtain enhanced position features and enhanced edge features.

[0139] Specifically, after obtaining the position domain side information, the camouflaged target detection device integrates the internal information of the position feature domain according to the position domain side information, thereby aggregating the feature information in the position feature domain.

[0140] Specifically, the disguised target detection device aggregates the same-domain information of the position domain edge information and the corresponding target position nodes in the position feature domain according to the preset activation function to obtain the position aggregation feature. The calculation method is as follows:

[0141]

[0142] in, is the position aggregation feature, S(·) is the preset activation function, is the location domain side information, The target location node.

[0143] 402. Aggregate the same-domain information of the edge domain side information and the corresponding target edge nodes in the edge feature domain according to a preset activation function to obtain edge aggregation features;

[0144] Specifically, after obtaining the edge domain side information, the camouflaged target detection device integrates the internal information of the edge feature domain according to the edge domain side information, thereby aggregating the feature information in the edge feature domain.

[0145] Specifically, the disguised target detection device aggregates the same-domain information of the edge domain side information and the corresponding target edge nodes in the edge feature domain according to the preset activation function to obtain the edge aggregation feature. The calculation method is as follows:

[0146]

[0147] in, is the edge aggregation feature, S(·) is the preset activation function, is the edge domain information, is the target edge node.

[0148] 403. Obtain a preset adapter to generate a cross-domain convolution kernel that associates the cross-domain side information with a position feature domain or an edge feature domain, and perform cross-domain information aggregation according to the cross-domain convolution kernel, the cross-domain side information, and a target position node or a target edge node corresponding to the cross-domain side information to obtain a cross-domain aggregation feature;

[0149] Specifically, the camouflaged target detection device also performs cross-domain information aggregation on the position feature domain and the edge feature domain according to the cross-domain side information to obtain a cross-domain aggregated feature.

[0150] Specifically, the camouflaged target detection device is pre-set with a preset adapter, in which a convolutional neural network module is provided, and a cross-domain convolution kernel associating cross-domain side information with a position feature domain or an edge feature domain is generated through the convolutional neural network module in the preset adapter. The cross-domain convolution kernel contains cross-domain guidance information, and cross-domain information is aggregated through the cross-domain convolution kernel, the cross-domain side information and the target position node or the target edge node corresponding to the cross-domain side information to obtain a cross-domain aggregated feature.

[0151] Optionally, when the target domain is an edge feature domain, that is, the calculation process of the information transfer process from the target position node in the position feature domain to the edge feature domain is as follows:

[0152]

[0153] in, Represents the information transfer function between nodes in two different domains. The target domain is the marginal feature domain. The node vector of i is a cross-domain convolution kernel.

[0154] Optionally, when the target domain is a location feature domain, that is, the calculation process of the information transfer process from the target edge node in the edge feature domain to the location feature domain is as follows:

[0155]

[0156] 404. Generate enhanced position features and enhanced edge features according to the position aggregation features, the edge aggregation features, and the cross-domain aggregation features.

[0157] Specifically, after acquiring the position aggregation features, edge aggregation features, and cross-domain aggregation features, the camouflaged target detection device generates enhanced position features and enhanced edge features according to the position aggregation features, edge aggregation features, and cross-domain aggregation features.

[0158] Specifically, the disguised target detection device pre-sets an application gated loop unit as an update function, and after obtaining the cross-domain aggregation features and the edge aggregation features, obtains the historical position node corresponding to the cross-domain aggregation features to obtain the intermediate position node of the target position node, wherein the historical position node is the position node of the target position node at the previous time point.

[0159] The disguised target detection device inputs the cross-domain aggregation features, edge aggregation features and historical location nodes into the preset application gated loop unit for state update to obtain the iterative enhanced location features. Specifically, the calculation method is as follows:

[0160]

[0161] in, To enhance the location features, U GRU () is the application of gated recurrent unit, is the historical location node, is the cross-domain aggregation feature. is the edge aggregation feature.

[0162] Specifically, the disguised target detection device also obtains the historical edge node corresponding to the cross-domain aggregation feature, inputs the cross-domain aggregation feature, the position aggregation feature and the historical edge node into the preset application gated loop unit for state update, and obtains the iterative enhanced edge feature. Specifically, the calculation method is as follows:

[0163]

[0164] in, To enhance edge features, U GRU () is the application of gated recurrent unit, is the historical edge node, is the cross-domain aggregation feature. Aggregate features for locations.

[0165] Specifically, after performing horizontal feature enhancement to obtain enhanced camouflage information and enhanced edge features, the camouflaged target detection device also makes full use of the spatiotemporal information of the camouflaged target image to perform further contextual learning on the features, that is, performing vertical feature enhancement on the enhanced camouflage information and enhanced edge features to obtain processed position features and processed edge features, and inputting the processed position features and processed edge features into the classification output module in the camouflaged target detection model for classification mapping to obtain the camouflaged target recognition result of the camouflaged target image.

[0166] In this embodiment, the disguised target detection device aggregates the same-domain information of the position domain side information and the corresponding target position node in the position feature domain according to a preset activation function to obtain a position aggregation feature;

[0167] According to the preset activation function, the edge domain side information and the corresponding target edge nodes in the edge feature domain are aggregated in the same domain to obtain edge aggregation features; a preset adapter is obtained to generate a cross-domain convolution kernel that associates the cross-domain side information with the position feature domain or the edge feature domain, and cross-domain information aggregation is performed according to the cross-domain convolution kernel, the cross-domain side information, and the target position node or target edge node corresponding to the cross-domain side information to obtain cross-domain aggregation features; enhanced position features and enhanced edge features are generated according to the position aggregation features, the edge aggregation features, and the cross-domain aggregation features. Horizontal feature enhancement of feature nodes in different feature domains is achieved to improve the subsequent recognition accuracy of camouflage target objects.

[0168] like Figure 6 As shown, Figure 6 A flowchart of an embodiment of performing longitudinal feature enhancement in the camouflaged target detection method provided in an embodiment of the present application, specifically, including steps 501 to 505:

[0169] 501. Acquire historical position information corresponding to the enhanced position feature and historical edge information corresponding to the enhanced edge feature;

[0170] Based on the above embodiments, in this embodiment, in order to fully utilize the spatiotemporal information of the disguised target image, the disguised target detection device also performs context learning on the enhanced position features and the enhanced edge features to obtain processed position features and processed edge features.

[0171] Specifically, the disguised target detection device obtains the historical position information of the enhanced position feature in the previous preset time period, and obtains the historical edge information of the enhanced edge feature in the previous preset time period.

[0172] 502. Calculate node similarity according to the historical position information and the transposed position information corresponding to the historical position information to obtain a first graph adjacency matrix;

[0173] Specifically, after obtaining the historical position information, the disguised target detection device calculates the node similarity according to the historical position information and the transposed position information corresponding to the historical position information, and generates the first graph adjacency matrix. Specifically, the calculation method is as follows:

[0174]

[0175] in, is the first graph adjacency matrix, To transpose the position information, For historical location information.

[0176] 503. Calculate the node similarity between the historical edge information and the transposed edge information corresponding to the historical edge information to obtain a second graph adjacency matrix;

[0177] Specifically, after obtaining the historical edge information, the disguised target detection device calculates the node similarity according to the historical edge information and the transposed edge information corresponding to the historical edge information, and generates the second graph adjacency matrix. Specifically, the calculation method is as follows:

[0178]

[0179] in, is the adjacency matrix of the second graph, To transpose the edge information, For historical marginal information.

[0180] 504. Perform graph reasoning according to the historical position information and the first graph adjacency matrix to obtain reasoned position information, and perform graph reasoning according to the historical edge information and the second graph adjacency matrix to obtain reasoned edge information;

[0181] Specifically, after obtaining the first graph adjacency matrix and the second graph adjacency matrix, the disguised target detection device performs graph reasoning according to the historical position information and the first graph adjacency matrix to obtain the reasoned position information, and performs graph reasoning according to the historical edge information and the second graph adjacency matrix to obtain the reasoned edge information. Specifically, the calculation method is as follows:

[0182]

[0183] in, To infer location information, is the inference edge information, g(·) is the preset nonlinear activation function, and W C and W E are the learnable parameters of the graph convolutional layer.

[0184] 505. Merge the inferred position information and the enhanced position feature to obtain a processed position feature, and merge the inferred edge information and the enhanced edge feature to obtain a processed edge feature.

[0185] Specifically, after obtaining the inferred position information, the camouflaged target detection device merges the inferred position information and the enhanced position feature, and obtains the processed position feature after reprojection.

[0186] Specifically, after obtaining the inferred edge information, the camouflaged target detection device merges the inferred edge information with the enhanced edge features, and obtains the processed edge features after reprojection.

[0187] After acquiring the processed position features and the processed edge features, the disguised target detection device inputs the processed position features and the processed edge features into a classification output module in the disguised target detection model for classification mapping to obtain a disguised target recognition result of the disguised target image.

[0188] In this embodiment, the camouflaged target detection device obtains the historical position information corresponding to the enhanced position feature and the historical edge information corresponding to the enhanced edge feature; calculates the node similarity according to the historical position information and the transposed position information corresponding to the historical position information to obtain a first graph adjacency matrix; calculates the node similarity between the historical edge information and the transposed edge information corresponding to the historical edge information to obtain a second graph adjacency matrix; performs graph reasoning according to the historical position information and the first graph adjacency matrix to obtain the inferred position information, performs graph reasoning according to the historical edge information and the second graph adjacency matrix to obtain the inferred edge information; performs feature merging of the inferred position information and the enhanced position feature to obtain the processed position feature, and performs feature merging of the inferred edge information and the enhanced edge feature to obtain the processed edge feature. The image features in the camouflaged target image are enhanced longitudinally by using the spatiotemporal information in the camouflaged target image, so as to improve the subsequent recognition accuracy of the camouflaged target object.

[0189] In order to better implement the disguised target detection method in the embodiment of the present application, based on the disguised target detection method, the embodiment of the present application also provides a disguised target detection device, such as Figure 7 As shown, Figure 7 The schematic diagram of the structure of the disguised target detection device provided in the embodiment of the present application is as follows. Specifically, the disguised target detection device 600 includes:

[0190] The image acquisition module 601 is configured to acquire a disguised target image to be detected;

[0191] A feature extraction module 602 is configured to extract features of the camouflaged target image to obtain position features and edge features;

[0192] A feature enhancement module 603 is configured to perform feature enhancement on the position feature and the edge feature to obtain a processed position feature and a processed edge feature;

[0193] The feature classification module 604 is configured to perform classification mapping according to the processed position features and the processed edge features to obtain a disguised output result and an edge output result of the disguised target image.

[0194] In a possible implementation of this embodiment, the camouflaged target detection device performs feature enhancement on the position feature and the edge feature to obtain a processed position feature and a processed edge feature, including:

[0195] Performing lateral feature enhancement on the position feature and the edge feature to obtain enhanced position feature and enhanced edge feature;

[0196] The enhanced position feature and the enhanced edge feature are subjected to longitudinal feature enhancement to obtain a processed position feature and a processed edge feature.

[0197] In a possible implementation of this embodiment, the camouflaged target detection device performs lateral feature enhancement on the position feature and the edge feature to obtain enhanced position feature and enhanced edge feature, including:

[0198] Performing node vectorization on the position feature and the edge feature to obtain a position feature domain and an edge feature domain;

[0199] Transmitting node information of the location feature domain and the edge feature domain to obtain location domain edge information, edge domain edge information and cross-domain edge information;

[0200] Node information is updated according to the location domain side information, the edge domain side information and the cross-domain side information to obtain enhanced location features and enhanced edge features.

[0201] In a possible implementation of this embodiment, the disguised target detection device performs node vectorization on the position feature and the edge feature to obtain a position feature domain and an edge feature domain, including:

[0202] Performing pyramid pooling processing on the position features and the edge features to obtain pooled position features and pooled edge features;

[0203] Performing interpolation processing on the pooled position features and the pooled edge features according to a preset interpolation layer to obtain a target position node and a target edge node;

[0204] Each of the target position nodes is aggregated to obtain a position feature domain of the disguised target image, and each of the target edge nodes is aggregated to obtain an edge feature domain of the disguised target image.

[0205] In a possible implementation of this embodiment, the disguised target detection device transfers node information on the position feature domain and the edge feature domain to obtain position domain side information, edge domain side information and cross-domain side information, including:

[0206] Performing a convolution operation on a first position node and a second position node in the position feature domain to obtain position domain edge information;

[0207] Performing a convolution operation on the first edge node and the second edge node in the edge feature domain to obtain edge domain edge information;

[0208] A dynamic convolution kernel associated with the first position node and / or the first edge node is generated based on a preset learner, and a convolution operation is performed according to the dynamic convolution kernel, the first position node and the first edge node to obtain cross-domain edge information.

[0209] In a possible implementation of this embodiment, the disguised target detection device updates the node information according to the position domain side information, the edge domain side information, and the cross-domain side information to obtain enhanced position features and enhanced edge features, including:

[0210] Aggregating the same-domain information of the location domain side information and the corresponding target location node in the location feature domain according to a preset activation function to obtain a location aggregation feature;

[0211] Aggregating the same-domain information of the edge domain side information and the corresponding target edge nodes in the edge feature domain according to a preset activation function to obtain edge aggregation features;

[0212] Obtain a preset adapter to generate a cross-domain convolution kernel that associates the cross-domain side information with a position feature domain or an edge feature domain, and perform cross-domain information aggregation according to the cross-domain convolution kernel, the cross-domain side information, and a target position node or a target edge node corresponding to the cross-domain side information to obtain a cross-domain aggregation feature;

[0213] An enhanced position feature and an enhanced edge feature are generated according to the position aggregation feature, the edge aggregation feature and the cross-domain aggregation feature.

[0214] In a possible implementation of this embodiment, the disguised target detection device generates enhanced position features and enhanced edge features according to the position aggregation features, the edge aggregation features, and the cross-domain aggregation features, including:

[0215] Obtaining a historical location node corresponding to the cross-domain aggregation feature, inputting the cross-domain aggregation feature, the edge aggregation feature, and the historical location node into a preset application gated loop unit for state update, and obtaining an iterative enhanced location feature;

[0216] The historical edge node corresponding to the cross-domain aggregation feature is obtained, and the cross-domain aggregation feature, the location aggregation feature and the historical edge node are input into a preset application gated loop unit for state update to obtain an iterated enhanced edge feature.

[0217] In a possible implementation of this embodiment, the camouflaged target detection device performs longitudinal feature enhancement on the enhanced position feature and the enhanced edge feature to obtain a processed position feature and a processed edge feature, including:

[0218] Acquire historical position information corresponding to the enhanced position feature and historical edge information corresponding to the enhanced edge feature;

[0219] Calculate node similarity based on the historical position information and the transposed position information corresponding to the historical position information to obtain a first graph adjacency matrix;

[0220] Calculating node similarity between the historical edge information and the transposed edge information corresponding to the historical edge information to obtain a second graph adjacency matrix;

[0221] Performing graph reasoning according to the historical position information and the first graph adjacency matrix to obtain reasoned position information, and performing graph reasoning according to the historical edge information and the second graph adjacency matrix to obtain reasoned edge information;

[0222] The inferred position information and the enhanced position feature are feature-merged to obtain a processed position feature, and the inferred edge information and the enhanced edge feature are feature-merged to obtain a processed edge feature.

[0223] In a possible implementation of this embodiment, the disguised target detection device extracts features from the disguised target image to obtain position features and edge features, including:

[0224] Inputting the disguised target image into a preset disguised target detection model, wherein the disguised target detection model includes a channel fusion module and a target attention module;

[0225] Acquiring image features in the disguised target image based on the channel fusion module, and weighting the image features according to the channel feature vectors corresponding to the image features to obtain weighted image features;

[0226] Based on the target attention module, multi-dimensional information detection is performed on the weighted image features to obtain position features and edge features of the disguised target image.

[0227] In this embodiment, the camouflaged target detection device obtains a camouflaged target image to be detected; extracts features from the camouflaged target image to obtain position features and edge features; enhances the position features and edge features to obtain processed position features and processed edge features; and performs classification mapping based on the processed position features and processed edge features to obtain a camouflaged target recognition result of the camouflaged target image. By enhancing the edge information of the camouflaged target image, the recognition accuracy of the camouflaged target object is improved.

[0228] The embodiment of the present invention also provides a disguised target detection device, such as Figure 8 As shown, Figure 8 This is a schematic diagram of the structure of an embodiment of a camouflaged target detection device provided in an embodiment of the present application.

[0229] The disguised target detection device integrates any one of the disguised target detection devices provided by the embodiments of the present invention, and the disguised target detection device includes:

[0230] one or more processors;

[0231] Memory; and

[0232] One or more applications, wherein the one or more applications are stored in the memory and are configured to execute, by the processor, the steps of the disguised target detection method described in any of the above-mentioned disguised target detection method embodiments.

[0233] Specifically, the camouflaged target detection device may include one or more processing core processors 701, one or more computer-readable storage media memories 702, a power supply 703, an input unit 704 and other components. Those skilled in the art will understand that Figure 8 The structure of the disguised target detection device shown in the figure does not constitute a limitation on the disguised target detection device, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0234] The processor 701 is the control center of the disguised target detection device. It uses various interfaces and lines to connect the various parts of the entire disguised target detection device. By running or executing the software programs and / or modules stored in the memory 702, and calling the data stored in the memory 702, it performs various functions of the disguised target detection device and processes data, thereby monitoring the disguised target detection device as a whole. Optionally, the processor 701 may include one or more processing cores; preferably, the processor 701 may integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user interface, and application programs, etc., and the modem processor mainly processes wireless communications. It is understandable that the above-mentioned modem processor may not be integrated into the processor 701.

[0235] The memory 702 can be used to store software programs and modules. The processor 701 executes various functional applications and data processing by running the software programs and modules stored in the memory 702. The memory 702 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area may store data created according to the use of the camouflaged target detection device, etc. In addition, the memory 702 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage devices. Accordingly, the memory 702 may also include a memory controller to provide the processor 701 with access to the memory 702.

[0236] The camouflaged target detection device also includes a power supply 703 for supplying power to various components. Preferably, the power supply 703 can be logically connected to the processor 701 through a power management system, so as to manage charging, discharging, and power consumption through the power management system. The power supply 703 can also include any components such as one or more DC or AC power supplies, recharging systems, power failure detection circuits, power converters or inverters, and power status indicators.

[0237] The disguised target detection device may further include an input unit 704, which may be used to receive input digital or character information and generate keyboard, mouse, joystick, optical or trackball signal input related to user settings and function control.

[0238] Although not shown, the disguised target detection device may also include a display unit, etc., which will not be described in detail here. Specifically in this embodiment, the processor 701 in the disguised target detection device will load the executable files corresponding to the processes of one or more application programs into the memory 702 according to the following instructions, and the processor 701 will run the application programs stored in the memory 702, thereby realizing various functions, as follows:

[0239] Acquire a camouflaged target image to be detected;

[0240] Extracting features from the camouflaged target image to obtain position features and edge features;

[0241] Performing feature enhancement on the position feature and the edge feature to obtain a processed position feature and a processed edge feature;

[0242] Classification mapping is performed according to the processed position features and the processed edge features to obtain a disguised target recognition result of the disguised target image.

[0243] To this end, an embodiment of the present invention provides a computer-readable storage medium, which may include: a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, etc. A computer program is stored thereon, and the computer program is loaded by a processor to execute the steps in any of the camouflaged target detection methods provided in the embodiments of the present invention. For example, the computer program may be loaded by a processor to execute the following steps:

[0244] Acquire a camouflaged target image to be detected;

[0245] Extracting features from the camouflaged target image to obtain position features and edge features;

[0246] Performing feature enhancement on the position feature and the edge feature to obtain a processed position feature and a processed edge feature;

[0247] Classification mapping is performed according to the processed position features and the processed edge features to obtain a disguised target recognition result of the disguised target image.

[0248] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the detailed description of other embodiments above, and will not be repeated here.

[0249] In specific implementation, the above units or structures can be implemented as independent entities, or can be arbitrarily combined to be implemented as the same or several entities. The specific implementation of the above units or structures can refer to the previous method embodiments, which will not be repeated here.

[0250] The specific implementation of the above operations can be found in the previous embodiments, which will not be described in detail here.

[0251] The above is a detailed introduction to a camouflaged target detection method provided in an embodiment of the present application. Specific embodiments are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for technical personnel in this field, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A camouflaged target detection method, characterized in that: The camouflaged target detection method comprises: Acquire a camouflaged target image to be detected; Extracting features of the camouflaged target image to obtain position features and edge features; Performing feature enhancement on the position feature and the edge feature to obtain processed position feature and processed edge feature, wherein the feature enhancement operation includes horizontal feature enhancement and vertical feature enhancement; Classification mapping is performed according to the processed position features and the processed edge features to obtain a disguised target recognition result of the disguised target image.

2. The camouflaged target detection method according to claim 1, characterized in that: The step of performing feature enhancement on the position feature and the edge feature to obtain processed position feature and processed edge feature includes: Performing lateral feature enhancement on the position feature and the edge feature to obtain enhanced position feature and enhanced edge feature; The enhanced position feature and the enhanced edge feature are subjected to longitudinal feature enhancement to obtain a processed position feature and a processed edge feature.

3. The camouflaged target detection method according to claim 2, characterized in that: The step of performing lateral feature enhancement on the position feature and the edge feature to obtain enhanced position feature and enhanced edge feature includes: Performing node vectorization on the position feature and the edge feature to obtain a position feature domain and an edge feature domain; Transmitting node information of the location feature domain and the edge feature domain to obtain location domain edge information, edge domain edge information and cross-domain edge information; Node information is updated according to the location domain side information, the edge domain side information and the cross-domain side information to obtain enhanced location features and enhanced edge features.

4. The camouflaged target detection method according to claim 3, characterized in that: The performing node vectorization on the position feature and the edge feature to obtain a position feature domain and an edge feature domain includes: Performing pyramid pooling processing on the position features and the edge features to obtain pooled position features and pooled edge features; Performing interpolation processing on the pooled position features and the pooled edge features according to a preset interpolation layer to obtain a target position node and a target edge node; Each of the target position nodes is aggregated to obtain a position feature domain of the disguised target image, and each of the target edge nodes is aggregated to obtain an edge feature domain of the disguised target image.

5. The camouflaged target detection method according to claim 3, characterized in that: The node information of the location feature domain and the edge feature domain is transmitted to obtain location domain side information, edge domain side information and cross-domain side information, including: Performing a convolution operation on a first position node and a second position node in the position feature domain to obtain position domain edge information; Performing a convolution operation on the first edge node and the second edge node in the edge feature domain to obtain edge domain edge information; A dynamic convolution kernel associated with the first position node and / or the first edge node is generated based on a preset learner, and a convolution operation is performed according to the dynamic convolution kernel, the first position node and the first edge node to obtain cross-domain edge information.

6. The camouflaged target detection method according to claim 3, characterized in that: The updating of node information according to the location domain side information, the edge domain side information and the cross-domain side information to obtain enhanced location features and enhanced edge features includes: Aggregating the same-domain information of the location domain side information and the corresponding target location node in the location feature domain according to a preset activation function to obtain a location aggregation feature; Aggregating the same-domain information of the edge domain side information and the corresponding target edge nodes in the edge feature domain according to a preset activation function to obtain edge aggregation features; Obtain a preset adapter to generate a cross-domain convolution kernel that associates the cross-domain side information with a position feature domain or an edge feature domain, and perform cross-domain information aggregation according to the cross-domain convolution kernel, the cross-domain side information, and a target position node or a target edge node corresponding to the cross-domain side information to obtain a cross-domain aggregation feature; An enhanced position feature and an enhanced edge feature are generated according to the position aggregation feature, the edge aggregation feature and the cross-domain aggregation feature.

7. The camouflaged target detection method according to claim 6, characterized in that: The generating enhanced position features and enhanced edge features according to the position aggregation features, the edge aggregation features and the cross-domain aggregation features includes: Obtaining a historical location node corresponding to the cross-domain aggregation feature, inputting the cross-domain aggregation feature, the edge aggregation feature, and the historical location node into a preset application gated loop unit for state update, and obtaining an iterative enhanced location feature; The historical edge node corresponding to the cross-domain aggregation feature is obtained, and the cross-domain aggregation feature, the location aggregation feature and the historical edge node are input into a preset application gated loop unit for state update to obtain an iterated enhanced edge feature.

8. The camouflaged target detection method according to claim 2, characterized in that: The performing longitudinal feature enhancement on the enhanced position feature and the enhanced edge feature to obtain processed position feature and processed edge feature includes: Acquire historical position information corresponding to the enhanced position feature and historical edge information corresponding to the enhanced edge feature; Calculate node similarity based on the historical position information and the transposed position information corresponding to the historical position information to obtain a first graph adjacency matrix; Calculating node similarity between the historical edge information and the transposed edge information corresponding to the historical edge information to obtain a second graph adjacency matrix; Performing graph reasoning according to the historical position information and the first graph adjacency matrix to obtain reasoned position information, and performing graph reasoning according to the historical edge information and the second graph adjacency matrix to obtain reasoned edge information; The inferred position information and the enhanced position feature are feature-merged to obtain a processed position feature, and the inferred edge information and the enhanced edge feature are feature-merged to obtain a processed edge feature.

9. The camouflaged target detection method according to any one of claims 1 to 8, characterized in that: The step of extracting features from the camouflaged target image to obtain position features and edge features includes: Inputting the disguised target image into a preset disguised target detection model, wherein the disguised target detection model includes a channel fusion module and a target attention module; Acquiring image features in the disguised target image based on the channel fusion module, and weighting the image features according to the channel feature vectors corresponding to the image features to obtain weighted image features; Based on the target attention module, multi-dimensional information detection is performed on the weighted image features to obtain position features and edge features of the disguised target image.

10. A camouflaged target detection device, characterized in that: The camouflaged target detection device comprises: An image acquisition module is configured to acquire an image of a camouflaged target to be detected; A feature extraction module is configured to extract features from the camouflaged target image to obtain position features and edge features; A feature enhancement module is configured to perform feature enhancement on the position feature and the edge feature to obtain a processed position feature and a processed edge feature; The feature classification module is configured to perform classification mapping according to the processed position features and the processed edge features to obtain a disguised output result and an edge output result of the disguised target image.

11. A disguised target detection device, characterized in that: The camouflaged target detection device comprises: one or more processors; Memory; and One or more applications, wherein the one or more applications are stored in the memory and are configured to be executed by the processor to implement the steps of the camouflaged target detection method according to any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and the computer program is loaded by a processor to execute the steps of the camouflaged target detection method according to any one of claims 1 to 9.