Federated hybrid domain generalized target detection method, device, electronic device and medium

Abstract

The present invention relates to a method for detecting a generalized target in a federated mixed domain, the method comprising: inputting a plurality of targets to be detected into each client respectively; obtaining detection results of the plurality of targets to be detected respectively; wherein each client comprises a pre-trained detection model; the pre-trained detection model is obtained by training the detection model with a mixed domain data set; the mixed domain data set is obtained by mixing data sets from a plurality of different domains. Through the above technical scheme, the detection model in the client is trained with a mixed domain data set, thereby obtaining a pre-trained detection model with good generalization and discrimination, and then the pre-trained detection model is used to detect the target to be detected, and the detection model in the cloud server and each client is used to realize the detection of a generalized target in a federated mixed domain, which has a good detection effect for mixed domain data or non-mixed domain data, and is more in line with the actual scene.

Description

Technical Field

[0001] The present invention relates to the field of federated generalized target detection, and in particular to a federated mixed-domain generalized target detection method, device, electronic device and medium. Background Art

[0002] Federated learning is a decentralized approach where each local client sends its parameters to a cloud server while saving data locally. Federated learning has gradually attracted attention due to its privacy protection advantages. Domain generalization object detection aims to transfer detection models with superior performance trained on a source domain to an unseen target domain. Recent studies have shown that extracting domain-invariant representations to mitigate the effects of domain shift given multiple source domains is a popular approach. Many methods have been proposed to improve the performance of federated learning models.

[0003] Existing methods propose to use adversarial attacks to accelerate the convergence of federated learning and alleviate forgetting in local federated training. However, this method is based on average visual features and is prone to infringing data privacy. Although these methods are effective, they usually require each client model to use a dataset from the same domain, which cannot be directly applied to mixed domain scenarios. Although most existing methods have been proven to be effective, these methods require access to multiple datasets from different fields, which cannot be directly applied to mixed domain dataset learning scenarios. Due to data privacy restrictions, some generalized object detection methods cannot be applied to federated learning. In addition, many existing federated domain generalization methods focus on improving the generalization ability of distributed single-domain datasets and cannot be directly applied to mixed datasets. Summary of the invention

[0004] In order to solve the above problems existing in the prior art, the present invention provides a federated hybrid domain generalized target detection method, device, electronic device and storage medium.

[0005] According to a first aspect of an embodiment of the present invention, a federated hybrid domain generalized target detection method is provided, the method comprising:

[0006] Input multiple targets to be detected into each client respectively;

[0007] Obtaining detection results of each of the multiple targets to be detected;

[0008] Each of the clients includes a pre-trained detection model; the pre-trained detection model is obtained by training the detection model with a mixed domain data set; the mixed domain data set is obtained by mixing data sets from multiple different domains.

[0009] Optionally, the training process of the pre-trained detection model includes:

[0010] For each of the clients, training the detection model using the mixed domain dataset;

[0011] When the detection model training is completed, the pre-trained detection model is obtained;

[0012] When the detection model training is not completed, the parameters of the detection model are updated according to the cloud server, and the detection model is trained again until the detection model training is completed.

[0013] Optionally, for each client, training the detection model using the mixed domain dataset includes:

[0014] Inputting the mixed domain data set into the backbone network in the detection model to obtain initial features corresponding to the mixed domain data set;

[0015] Inputting a preset prototype set into the backbone network to obtain a target prototype set;

[0016] Obtaining a descriptor according to the target prototype set and the initial features;

[0017] Obtaining fused features using the initial features and the descriptors;

[0018] The fused features are respectively input into the region proposal network and the ROI feature extractor of the detection model to obtain the classification corresponding to the fused features.

[0019] Optionally, when the detection model training is not completed, updating the parameters of the detection model according to the cloud server, and training the detection model again until the detection model training is completed, including:

[0020] According to the classification corresponding to the fused feature and the real classification of the initial feature, the loss value is obtained by using the loss function;

[0021] After each training of the detection model, uploading the detection model obtained in this training and the target prototype set obtained in this training to the cloud server;

[0022] Connect the target prototype sets obtained by each client in this training to obtain a prototype set;

[0023] Obtaining aggregated prototypes according to the adjacency matrix and the prototype set;

[0024] When the loss value is greater than or equal to the preset value, the aggregated prototype is downloaded on each client, the model parameters of the detection model are updated according to the aggregated prototype, and the detection model is continuously trained until the loss value is less than the preset threshold.

[0025] Optionally, the descriptor is obtained according to the following formula:

[0026]

[0027] Among them, V i represents the descriptor of the i-th prototype in the target prototype set, m represents the width of the initial feature F, h represents the height of the initial feature F, e represents the Napier constant, S j,i is the value of the i-th prototype in the j-th dimension in the soft assignment. D is the number of prototypes in the target prototype set, p i Represents the i-th prototype in the target prototype set.

[0028] Optionally, the adjacency matrix is ​​obtained according to the following formula:

[0029]

[0030] in, represents the prototype set, represents the transpose of the prototype set, I is the identity matrix, A' is the intermediate matrix, and A is the adjacency matrix.

[0031] Optionally, the aggregated prototype is obtained according to the following formula:

[0032]

[0033] in, is the prototype after aggregation, and Q represents the maximum pooling operation in the channel dimension.

[0034] According to a second aspect of an embodiment of the present invention, a federated hybrid domain generalized target detection device is provided, the device comprising:

[0035] An input module, used to input multiple targets to be detected into each client respectively;

[0036] A detection module, used to obtain detection results of each of the multiple targets to be detected;

[0037] Each of the clients includes a pre-trained detection model; the pre-trained detection model is obtained by training the detection model with a mixed domain data set; the mixed domain data set is obtained by mixing data sets from multiple different domains.

[0038] According to a third aspect of an embodiment of the present disclosure, there is provided a federated hybrid domain generalized target detection apparatus, comprising: a processor; a memory for storing instructions executable by the processor;

[0039] The processor is configured to: execute the executable instructions to implement the steps of the federated mixed domain generalized target detection method described in any implementation of the first aspect above.

[0040] According to a fourth aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored. When the program instructions are executed by a processor, the steps of the federated mixed domain generalized target detection method described in the first aspect of the present invention are implemented.

[0041] The technical solution provided by the embodiments of the present invention may have the following beneficial effects:

[0042] In the above technical solution, multiple targets to be detected are input into each client respectively; the detection results of the multiple targets to be detected are obtained; wherein each client includes a pre-trained detection model; the pre-trained detection model is obtained by training the detection model through a mixed domain data set; the mixed domain data set is obtained by mixing data sets from multiple different domains. Through the above technical solution, the detection model in the client is trained through a mixed domain data set, so that the pre-trained detection model has good generalization and discrimination, and then the pre-trained detection model is used to detect the target to be detected, which has a good detection effect for mixed domain data or non-mixed domain data, and is more in line with the actual scene.

[0043] Other features and advantages of the present invention will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the present invention but do not constitute a limitation of the present invention. In the accompanying drawings:

[0045] Figure 1 The present invention is a flowchart of a method for generalized target detection in a federated mixed domain according to an exemplary embodiment.

[0046] Figure 2 It is a schematic diagram showing a detection model training process according to an exemplary embodiment.

[0047] Figure 3 It is a block diagram of a federated hybrid domain generalized target detection device according to an exemplary embodiment.

[0048] Figure 4 The present invention is a block diagram of an electronic device for a federated mixed-domain generalized target detection method according to an exemplary embodiment.

[0049] Figure 5It is a block diagram of another electronic device for a federated mixed-domain generalized target detection method according to an exemplary embodiment. DETAILED DESCRIPTION

[0050] The specific implementation of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the present invention, and is not used to limit the present invention.

[0051] Figure 1 is a flowchart of a federated hybrid domain generalized target detection method according to an exemplary embodiment. Figure 1 As shown, the method includes the following steps.

[0052] In S101, a plurality of targets to be detected are input into each client respectively.

[0053] In S102, the detection results of the multiple targets to be detected are obtained;

[0054] Each client includes a pre-trained detection model; the pre-trained detection model is obtained by training the detection model with a mixed domain dataset; the mixed domain dataset is obtained by mixing datasets from multiple different domains.

[0055] It can be understood that in the federated hybrid domain generalized target detection, the detection model in each client is aggregated into a federated detection model after being uploaded to the cloud server to achieve the federated hybrid domain generalized target detection. In the federated hybrid domain generalized target detection, (x, y) can be used to represent the image and label space, where x represents the image and y represents the label. The source domain is represented as D L ={D1,D2,...,D k}, where k is the number of source domains. The source domain D of each client L K It is a mixture of multiple datasets, that is, the mixed domain dataset can be K is the number of clients.

[0056] Optionally, the training process of the pre-trained detection model includes:

[0057] For each client, the detection model is trained using the mixed domain dataset;

[0058] When the detection model training is completed, a pre-trained detection model is obtained;

[0059] When the detection model training is not completed, the parameters of the detection model are updated according to the cloud server, and the detection model is trained again until the detection model training is completed.

[0060] Optionally, for each client, a detection model is trained using a mixed domain dataset, including:

[0061] Input the mixed domain dataset into the backbone network in the detection model to obtain the initial features corresponding to the mixed domain dataset;

[0062] Input the preset prototype set into the backbone network to obtain the target prototype set;

[0063] Obtain a descriptor based on the target prototype set and initial features;

[0064] Use the initial features and descriptors to obtain fused features;

[0065] The fused features are respectively input into the region proposal network and ROI feature extractor of the detection model to obtain the classification corresponding to the fused features.

[0066] Understandably, Figure 2 is a schematic diagram showing a detection model training process according to an exemplary embodiment. Figure 2 As shown, the detection model in each client of the present invention adopts Faster RCNN, in which the backbone network is ResNet101. The mixed domain dataset is input into the backbone network in the detection model to obtain the initial features corresponding to the mixed domain dataset. m represents the width of the initial feature F, h represents the height of the initial feature F, and d represents the number of channels of the initial feature F.

[0067] In one implementation, the descriptor is obtained according to the following formula:

[0068]

[0069] Among them, V i represents the descriptor of the i-th prototype in the target prototype set, m represents the width of the initial feature F, h represents the height of the initial feature F, e represents the Napier constant, S j,i is the value of the i-th prototype in the j-th dimension in the soft assignment. D is the number of prototypes in the target prototype set, p i represents the i-th prototype in the target prototype set. In addition, F j -p i Expressed as residual operation, soft allocation S = W g *F+b g , W g and b g are the convolution parameters.

[0070] It is worth mentioning that Figure 2 As shown, the fusion features obtained by using the initial features and descriptors can be referred to: V i =F+V i , the obtained fusion feature Vi The region proposal network (RPN) and ROI feature extractor that will be fed into the detection model. RPN generates anchor boxes by sliding a small window on the feature map of the input image. RPN uses anchor boxes as a basis, each of which has a fixed scale and aspect ratio. For each anchor box, RPN predicts two values: the probability of whether it contains an object and the regression parameters that adjust the boundaries of the candidate box. By applying these predictions to all anchor boxes, RPN can generate a series of candidate object boxes. RPN classifies and regresses the anchor boxes, filters out some candidate boxes based on the predicted probability and boundary adjustment, and selects the candidate boxes with high probability as candidate object boxes. The ROI feature extractor uses the candidate object boxes generated by RPN for object detection. The input of the ROI feature extractor is the candidate object boxes generated by RPN and the feature maps corresponding to these candidate object boxes. First, the ROI feature extractor crops and scales each candidate object box to match the fixed input size. Then, these cropped feature maps are fed into a fully connected layer network to extract feature representations associated with each candidate box. Finally, for each candidate object box, the ROI feature extractor predicts the object category and bounding box adjustment through a classification head and a regression head. The classification head is used to predict the class probability of the object, and the regression head is used to adjust the bounding box coordinates of the candidate box to more accurately frame the object.

[0071] Optionally, when the detection model training is not completed, the parameters of the detection model are updated according to the cloud server, and the detection model is trained again until the detection model training is completed, including:

[0072] According to the classification corresponding to the fused features and the true classification of the initial features, the loss value is obtained using the loss function;

[0073] After each training of the detection model, the detection model obtained in this training and the target prototype set obtained in this training are uploaded to the cloud server;

[0074] Connect the target prototype sets obtained by each client in this training to obtain the prototype set;

[0075] Obtain the aggregated prototype according to the adjacency matrix and the prototype set;

[0076] When the loss value is greater than or equal to the preset value, the aggregated prototype is downloaded on each client, the model parameters of the detection model are updated according to the aggregated prototype, and the detection model is continuously trained until the loss value is less than the preset threshold.

[0077] It is understandable that if Figure 2As shown in the figure, after each training of the detection model, the detection model obtained by the client in this training and the target prototype set obtained in this training are uploaded to the cloud server, and the target prototype set obtained by each client in this training is connected according to the prototype graph network to obtain the prototype set, and then the aggregated prototype is obtained according to the adjacency matrix and the prototype set. When the loss value is greater than or equal to the preset value, the aggregated prototype is downloaded from each client, the model parameters of the detection model are updated according to the aggregated prototype, and the detection model is continuously trained until the loss value is less than the preset threshold.

[0078] It can be understood that the adjacency matrix is ​​obtained according to the following formula:

[0079]

[0080] in, represents a collection of prototypes, represents the transpose of the prototype set, I is the identity matrix, A' is the intermediate matrix, and A is the adjacency matrix.

[0081] The aggregated prototype is obtained according to the following formula:

[0082]

[0083] in, is the prototype after aggregation, and Q represents the maximum pooling operation in the channel dimension.

[0084] It is worth mentioning that the loss function used in the present invention will optimize the detection model in each client, and the loss function can be referred to as follows:

[0085]

[0086] in, and They represent classification loss - cross entropy loss and regression loss - SmoothL1 loss respectively. is the loss of the region proposal network, which includes classification loss and regression loss.

[0087] In the above technical solutions, the proposed federated hybrid domain generalized target detection method is more in line with the actual scenario. Prototypes are used to capture rich object-related features, and prototype sets are used to aggregate features, thereby extracting more generalized fusion features, and the aggregated prototypes are used to update the parameters of the detection model, making the features learned by the client more generalized and discriminative, and the aggregated prototypes are used to update the parameters of the detection model.

[0088] Figure 3 is a block diagram of a federated hybrid domain generalized target detection device according to an exemplary embodiment. Figure 3, the federated hybrid domain generalized target detection device 300 may include.

[0089] An input module 301 is used to input multiple targets to be detected into each client respectively;

[0090] A detection module 302 is used to obtain detection results of each of the multiple targets to be detected;

[0091] Each client includes a pre-trained detection model; the pre-trained detection model is obtained by training the detection model with a mixed domain dataset; the mixed domain dataset is obtained by mixing datasets from multiple different domains.

[0092] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0093] Figure 4 is a block diagram of an electronic device for a federated mixed domain generalized target detection method according to an exemplary embodiment. Figure 4 As shown, the electronic device 400 may include: a processor 401 and a memory 402. The electronic device 400 may also include one or more of a multimedia component 403, an input / output (I / O) interface 404, and a communication component 405.

[0094] Among them, the processor 401 is used to control the overall operation of the electronic device 400 to complete all or part of the steps in the above-mentioned federated hybrid domain generalized target detection method. The memory 402 is used to store various types of data to support the operation of the electronic device 400, and these data may include, for example, instructions for any application or method used to operate on the electronic device 400, and application-related data, such as contact data, sent and received messages, pictures, audio, video, etc. The memory 402 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read-Only Memory, referred to as EPROM), programmable read-only memory (Programmable Read-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic memory, flash memory, disk or optical disk. The multimedia component 403 may include a screen and an audio component. The screen may be, for example, a touch screen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signal may be further stored in the memory 402 or sent through the communication component 405. The audio component also includes at least one speaker for outputting audio signals. The I / O interface 404 provides an interface between the processor 401 and other interface modules, and the other interface modules may be keyboards, mice, buttons, etc. These buttons may be virtual buttons or physical buttons. The communication component 405 is used for wired or wireless communication between the electronic device 400 and other devices. Wireless communication, such as Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, 4G, NB-IOT, eMTC, or other 5G, etc., or a combination of one or more of them, is not limited here. Therefore, the corresponding communication component 405 may include: Wi-Fi module, Bluetooth module, NFC module, etc.

[0095] In an exemplary embodiment, the electronic device 400 can be implemented by one or more application specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing devices (DSPD), programmable logic devices (PLD), field programmable gate arrays (FPGA), controllers, microcontrollers, microprocessors or other electronic components to execute the above-mentioned federal hybrid domain generalized target detection method.

[0096] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, and when the program instructions are executed by a processor, the steps of the above-mentioned federated hybrid domain generalized target detection method are implemented. For example, the computer-readable storage medium can be the above-mentioned memory 402 including program instructions, and the above-mentioned program instructions can be executed by the processor 401 of the electronic device 400 to complete the above-mentioned federated hybrid domain generalized target detection method.

[0097] Figure 5 is a block diagram of another electronic device for a federated mixed domain generalized target detection method according to an exemplary embodiment. For example, the electronic device 500 may be provided as a server. Figure 5 The electronic device 500 includes a processor 522, which may be one or more, and a memory 532 for storing a computer program executable by the processor 522. The computer program stored in the memory 532 may include one or more modules, each corresponding to a set of instructions. In addition, the processor 522 may be configured to execute the computer program to perform the above-mentioned federated mixed domain generalized target detection method.

[0098] In addition, the electronic device 500 may further include a power supply component 526 and a communication component 550, wherein the power supply component 526 may be configured to perform power management of the electronic device 500, and the communication component 550 may be configured to implement communication, for example, wired or wireless communication, of the electronic device 500. In addition, the electronic device 500 may further include an input / output (I / O) interface 558. The electronic device 500 may operate based on an operating system stored in the memory 532.

[0099] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, and when the program instructions are executed by a processor, the steps of the above-mentioned federated hybrid domain generalized target detection method are implemented. For example, the non-transitory computer-readable storage medium can be the above-mentioned memory 532 including program instructions, and the above-mentioned program instructions can be executed by the processor 522 of the electronic device 500 to complete the above-mentioned federated hybrid domain generalized target detection method.

[0100] In another exemplary embodiment, a computer program product is also provided, which includes a computer program that can be executed by a programmable device, and the computer program has a code portion for executing the above-mentioned federated mixed-domain generalized target detection method when executed by the programmable device.

[0101] The preferred embodiments of the present invention are described in detail above in conjunction with the accompanying drawings. However, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, a variety of simple modifications can be made to the technical solution of the present invention, and these simple modifications all belong to the protection scope of the present invention.

[0102] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.

[0103] In addition, various embodiments of the present invention may be arbitrarily combined, and as long as they do not violate the concept of the present invention, they should also be regarded as the contents disclosed by the present invention.

Claims

1. A federated mixed domain generalized target detection method, characterized in that: The method comprises: Input multiple targets to be detected into each client respectively; Obtaining detection results of each of the multiple targets to be detected; Each of the clients includes a pre-trained federated detection model; the pre-trained federated detection model is obtained by training the federated detection model with a mixed domain dataset; the mixed domain dataset is obtained by mixing datasets from multiple different domains; The training process of the pre-trained federated detection model includes: For each of the clients, training the federated detection model using the mixed domain dataset; When the training of the federated detection model is completed, obtaining the pre-trained federated detection model; When the training of the federated detection model is not completed, updating the parameters of the federated detection model according to the cloud server, and training the federated detection model again until the training of the federated detection model is completed; The step of training the federated detection model using the mixed domain dataset for each client includes: Inputting the mixed domain data set into the backbone network in the federated detection model to obtain initial features corresponding to the mixed domain data set; Inputting a preset prototype set into the backbone network to obtain a target prototype set; Obtaining a descriptor according to the target prototype set and the initial features; Obtaining fused features using the initial features and the descriptors; Inputting the fused features into the region proposal network and the ROI feature extractor of the federated detection model respectively to obtain the classification corresponding to the fused features; The descriptor is obtained according to the following formula: Among them, V i represents the descriptor of the i-th prototype in the target prototype set, m represents the width of the initial feature F, h represents the height of the initial feature F, e represents the Napier constant, S j,i is the value of the i-th prototype in the j-th dimension in the soft assignment, D is the number of prototypes in the target prototype set, and p i Represents the i-th prototype in the target prototype set.

2. The federated hybrid domain generalized target detection method according to claim 1, characterized in that: When the training of the federated detection model is not completed, updating the parameters of the federated detection model according to the cloud server, and training the federated detection model again until the training of the federated detection model is completed, including: According to the classification corresponding to the fused feature and the real classification of the initial feature, a loss value is obtained by using a loss function; After each training of the federated detection model, uploading the federated detection model obtained in this training and the target prototype set obtained in this training to the cloud server; Connect the target prototype sets obtained by each client in this training to obtain a prototype set; Obtaining aggregated prototypes according to the adjacency matrix and the prototype set; When the loss value is greater than or equal to a preset value, the aggregated prototype is downloaded on each client, the model parameters of the federated detection model are updated according to the aggregated prototype, and the federated detection model is continuously trained until the loss value is less than a preset threshold.

3. The federated hybrid domain generalized target detection method according to claim 2, characterized in that: The adjacency matrix is ​​obtained according to the following formula: in, represents the prototype set, represents the transpose of the prototype set, I is the identity matrix, A' is the intermediate matrix, and A is the adjacency matrix.

4. The federated hybrid domain generalized target detection method according to claim 3, characterized in that: The polymerized prototype is obtained according to the following formula: in, is the prototype after aggregation, and Q represents the maximum pooling operation in the channel dimension.

5. A federated hybrid domain generalized target detection device, characterized in that: The device comprises: An input module, used to input multiple targets to be detected into each client respectively; A detection module, used to obtain detection results of each of the multiple targets to be detected; Each of the clients includes a pre-trained federated detection model; the pre-trained federated detection model is obtained by training the federated detection model with a mixed domain dataset; the mixed domain dataset is obtained by mixing datasets from multiple different domains; The training process of the pre-trained federated detection model includes: For each of the clients, training the federated detection model using the mixed domain dataset; When the training of the federated detection model is completed, obtaining the pre-trained federated detection model; When the training of the federated detection model is not completed, updating the parameters of the federated detection model according to the cloud server, and training the federated detection model again until the training of the federated detection model is completed; The step of training the federated detection model using the mixed domain dataset for each client includes: Inputting the mixed domain data set into the backbone network in the federated detection model to obtain initial features corresponding to the mixed domain data set; Inputting a preset prototype set into the backbone network to obtain a target prototype set; Obtaining a descriptor according to the target prototype set and the initial features; Obtaining fused features using the initial features and the descriptors; Inputting the fused features into the region proposal network and the ROI feature extractor of the federated detection model respectively to obtain the classification corresponding to the fused features; The descriptor is obtained according to the following formula: Among them, V i represents the descriptor of the i-th prototype in the target prototype set, m represents the width of the initial feature F, h represents the height of the initial feature F, e represents the Napier constant, S j,i is the value of the i-th prototype in the j-th dimension in the soft assignment, D is the number of prototypes in the target prototype set, and p i Represents the i-th prototype in the target prototype set.

6. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to: execute the executable instructions to implement the steps of the federated mixed domain generalized target detection method according to any one of claims 1 to 4.

7. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the program instructions are executed by a processor, the steps of the federated mixed domain generalized target detection method described in any one of claims 1 to 4 are implemented.

Images