Method and device for updating target detection model

Abstract

The present application discloses a method and device for updating a target detection model. The method obtains a data set for updating the target detection model, and uses a preset confidence upper bound algorithm to update the target detection model according to the data set to obtain model parameters and binary codes of the importance corresponding to the model parameters; and pairs the model parameters and the importance binary codes to obtain key-value pairs; finally, the key-value pairs are sent to an intelligent robot, so that the target detection model deployed on the intelligent robot is updated according to the key-value pairs, thereby facilitating the robot to select the model parameters to be updated, reducing the risk of the model forgetting the original target features due to a large number of changes in model parameters, and improving the model update effect. At the same time, the model is pre-updated on a computer device to avoid errors when updating the model directly on the robot and causing unpredictable losses, thereby further improving the model update effect.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method and device for updating a target detection model. Background Art

[0002] With the rise of artificial intelligence, intelligent robots are increasingly used in public places such as shopping malls, airports, and train stations. By deploying trained object detection models on intelligent robots, they can detect and recognize objects.

[0003] Currently, object detection models used in intelligent robots are unable to recognize objects of unknown categories. There are two main approaches to training for unknown objects. One is to integrate the new object into the training set, train a new model, and deploy it on the robot. However, this involves complex model updates and iterations and is prone to errors. The other is to retrain the object detection model using data from the new object. However, due to the high correlation between model parameters and the identified object, updating a large number of model parameters can cause the model to forget previously learned features, resulting in a decrease in recognition accuracy for the original object. Therefore, current object detection models suffer from poor update performance. Summary of the Invention

[0004] The present application provides a method and device for updating a target detection model to solve the technical problem that the current target detection model has poor updating effect.

[0005] In order to solve the above technical problems, in a first aspect, an embodiment of the present application provides a method for updating a target detection model, which is applied to a computer device. The updating method includes:

[0006] Obtain a dataset for updating the object detection model;

[0007] Using the preset confidence upper bound algorithm, the target detection model is updated according to the data set to obtain the new model parameters and the corresponding binary encoding of the importance of the new model parameters;

[0008] Pair the new model parameters and their binary encodings to obtain key-value pairs;

[0009] The key-value pairs are sent to the intelligent robot, and the key-value pairs are used to update the target detection model deployed on the intelligent robot according to the importance binary encoding and the new model parameters.

[0010] This embodiment obtains a data set for updating the target detection model, and uses a preset confidence upper bound algorithm to update the target detection model according to the data set to obtain model parameters and the corresponding importance binary codes of the model parameters, thereby obtaining each model parameter after the target detection model is updated and the corresponding importance degree of each model parameter; and pairs the model parameters and the importance binary codes to obtain key-value pairs, so as to correspond the model parameters and the importance binary codes one by one, thereby facilitating the robot to select the model parameters to be updated, reducing the risk of the model forgetting the original target features due to a large number of changes in model parameters, and improving the model update effect; finally, the key-value pairs are sent to the intelligent robot, so that the target detection model deployed on the intelligent robot is updated according to the key-value pairs, thereby realizing the pre-update of the model on the computer device, so as to avoid errors when updating the model directly on the robot and causing unpredictable losses, and further improving the model update effect.

[0011] In one embodiment, a preset confidence upper bound algorithm is used to update the target detection model based on the data set to obtain new model parameters and binary codes of importance corresponding to the new model parameters, including:

[0012] Using the preset confidence upper bound algorithm, the target detection model is iteratively trained according to the data set until the initial target detection model reaches the preset convergence condition, the iteration is stopped, and the new model parameters are obtained;

[0013] According to the new model parameters and the original model parameters of the target detection model, the importance binary coding corresponding to the new model parameters is determined.

[0014] This embodiment compares the new model parameters with the original model parameters to determine the importance of the new model parameters, thereby facilitating the subsequent selective updating of the model by the intelligent robot and effectively avoiding the catastrophic problem of forgetting the model.

[0015] In one embodiment, determining the importance binary codes corresponding to the model parameters based on the new model parameters and the original model parameters of the target detection model includes:

[0016] Determine the variance between the new model parameters and the original model parameters;

[0017] Based on the variance, the importance of the new model parameters is binary coded.

[0018] In one embodiment, the new model parameters and the binary encoding of the importance are paired to obtain key-value pairs, including:

[0019] Arrange the new model parameters to obtain an arrangement order, and determine the arrangement position of each new model parameter in the arrangement order;

[0020] Determine the importance binary coding as the target new model parameter of the preset coding;

[0021] The target new model parameter and the arrangement position of the target new model parameter are combined to obtain a key-value pair.

[0022] In one embodiment, obtaining a dataset for updating an object detection model includes:

[0023] Receive a target image sent by the intelligent robot, where the target image is an image detected by the target detection model with a confidence level lower than a preset threshold;

[0024] Combine multiple target images into a dataset.

[0025] This embodiment uses low-confidence images collected by the intelligent robot in actual scenarios as training samples to enable the target detection model to be applicable to more application scenarios and enhance the model performance of the target detection model.

[0026] In a second aspect, an embodiment of the present application provides a method for updating a target detection model, which is applied to an intelligent robot. The updating method includes:

[0027] receiving a key-value pair sent by a computer device, the key-value pair including a new model parameter and a binary code of the importance of the new model parameter, the key-value pair being obtained by the computer device by updating the target detection model based on the data set using a preset confidence upper bound algorithm, and pairing the updated new model parameter with the binary code of the importance;

[0028] The object detection model deployed on the intelligent robot is updated based on the importance binary coding and new model parameters.

[0029] In a third aspect, an embodiment of the present application provides a device for updating a target detection model, comprising:

[0030] An acquisition module is used to obtain a dataset for updating the target detection model;

[0031] The first updating module is used to update the target detection model according to the data set using a preset confidence upper bound algorithm to obtain new model parameters and binary codes of importance corresponding to the new model parameters;

[0032] The pairing module is used to pair the new model parameters with the binary encoding of the importance to obtain key-value pairs;

[0033] The sending module is used to send key-value pairs to the intelligent robot. The key-value pairs are used to update the target detection model deployed on the intelligent robot according to the importance binary coding and new model parameters.

[0034] In a fourth aspect, an embodiment of the present application provides a device for updating a target detection model, comprising:

[0035] a receiving module, configured to receive a key-value pair sent by a computer device, the key-value pair including a new model parameter and a binary code of the importance of the new model parameter. The key-value pair is obtained by pairing the updated new model parameter and the binary code of the importance by the computer device using a preset confidence upper bound algorithm to update the target detection model according to the data set;

[0036] The second updating module is used to update the target detection model deployed on the intelligent robot according to the importance binary coding and the new model parameters.

[0037] In a fifth aspect, an embodiment of the present application provides a computer device comprising a processor and a memory, wherein the memory is used to store a computer program, and when the computer program is executed by the processor, the method for updating the target detection model as in the first aspect is implemented.

[0038] In a sixth aspect, an embodiment of the present application provides an intelligent robot comprising a processor and a memory, wherein the memory is used to store a computer program, and when the computer program is executed by the processor, the method for updating the target detection model as in the second aspect is implemented.

[0039] It should be noted that the beneficial effects of the second to sixth aspects mentioned above can be referred to the relevant description of the first aspect mentioned above, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A flowchart of a method for updating an object detection model provided in one embodiment of the present application;

[0041] Figure 2 A schematic diagram of model parameter update provided in an embodiment of the present application;

[0042] Figure 3 A schematic diagram of detection update of the target detection model provided in an embodiment of the present application;

[0043] Figure 4 A schematic flow chart of a method for updating an object detection model provided in another embodiment of the present application;

[0044] Figure 5 A schematic diagram of the structure of an updating device for an object detection model provided in one embodiment of the present application;

[0045] Figure 6 A schematic structural diagram of an object detection model updating device provided in another embodiment of the present application;

[0046] Figure 7 A schematic diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0048] Currently, object detection models used in intelligent robots are unable to recognize objects of unknown categories. There are two main approaches to training for unknown objects. One is to integrate the new object into the training set, train a new model, and deploy it on the robot. However, this involves complex model updates and iterations and is prone to errors. The other is to retrain the object detection model using data from the new object. However, due to the high correlation between model parameters and the identified object, updating a large number of model parameters can cause the model to forget previously learned features, resulting in a decrease in recognition accuracy for the original object. Therefore, current object detection models suffer from poor update performance.

[0049] To this end, an embodiment of the present application provides a method and device for updating a target detection model, by obtaining a data set for updating the target detection model, and using a preset confidence upper bound algorithm to update the target detection model according to the data set, obtaining model parameters and corresponding importance binary codes of the model parameters, thereby obtaining each model parameter after the target detection model is updated and the corresponding importance degree of each model parameter; and pairing the model parameters and the importance binary codes to obtain key-value pairs, so as to correspond the model parameters and the importance binary codes one by one, thereby facilitating the robot to select the model parameters to be updated, reducing the risk of the model forgetting the original target features due to a large number of changes in model parameters, and improving the model update effect; finally, sending the key-value pairs to the intelligent robot, so that the target detection model deployed on the intelligent robot is updated according to the key-value pairs, thereby realizing the pre-update of the model on the computer device, so as to avoid errors when updating the model directly on the robot and causing unpredictable losses, and further improving the model update effect.

[0050] Please refer to Figure 1 , Figure 1 The present invention provides a flowchart of a target detection model update method. The present invention can be applied to computer devices, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, physical servers, cloud servers and other computing devices. Figure 1 As shown, the target detection model updating method includes steps S101 to S104, which are described in detail as follows:

[0051] Step S101: Acquire a data set for updating the target detection model.

[0052] In this step, the dataset consists of low-confidence images captured by the intelligent robot in real-world scenarios. The intelligent robot includes a camera module, a communication module, a data storage module, a central control module, and a deep learning module. The camera module receives commands from the central control module and sets parameters such as the camera's height and tilt angle. It acquires image data, which the deep learning module then detects and stores in the data storage module. It also provides feedback on the robot's current operating status to the central control module. The communication module serves as the communication interface between the robot system and the outside world, enabling bidirectional communication with the central control module. It can both receive control commands from the outside world and provide feedback on the robot's current operating status. The data storage module can be a memory device, receiving commands from the central control module, providing feedback on the robot's current operating status to the central control module, and storing large amounts of data collected by the camera module. The central control module can be a processor, serving as the decision-making center of the robot system. It can obtain operating status information from the remaining modules and send commands to them, thereby coordinating their operations. The deep learning module encapsulates a target detection model as a logical module for the robot to distinguish data. It can identify targets such as people or objects in the images captured by the camera, and can receive model parameter update instructions from the central control module to update the model parameters that need to be updated, thereby realizing online updates.

[0053] It is understandable that the module composition and connection structure of the above-mentioned intelligent robot are used as examples rather than limitations, and other embodiments may include more or fewer modules, such as only a memory and a processor.

[0054] For example, a user sends a command to the robot to initiate a target detection task. Upon receiving the command, the robot's communication module transmits it to the central control module. The central control module then obtains the status of each relevant module, initializes it, and initiates real-time monitoring of its status, such as data storage capacity and data transmission rate. The deep learning module then begins its work, and the target detection task begins: The current target detection performance is assessed. If the confidence level is below a preset threshold, the current frame is captured and stored in the data storage module. Once a certain number of frames are captured, the images are sent to the computer. If the confidence level is above the preset threshold, the current target detection task continues.

[0055] Step S102 : Using a preset confidence upper bound algorithm, the target detection model is updated according to the data set to obtain new model parameters and binary codes of importance corresponding to the new model parameters.

[0056] In this step, the original model parameters of the target detection model are the model parameters obtained from the previous training, rather than the random parameters of the model trained from scratch. The original model parameters can also be assigned binary importance codes, for example, the binary importance codes of the original model parameters are all 0. Based on the degree of change in the updated parameters, the binary importance codes of the original model parameters are also updated, with high importance being 1 and low importance being 0. It is understood that model parameters with high importance are important to the target detection model, and therefore these model parameters are not updated when the target detection model is updated.

[0057] Step S103 : Pair the new model parameters with the importance binary codes to obtain key-value pairs.

[0058] In this step, the key-value pair can be {model position: model parameter at the corresponding position}. For example, if the original model parameters are {1:1, 2:2, 3:3}, and the key-value pair that needs to be updated based on server model training is {2:5}, then the updated model parameter key-value pair will be {1:1, 2:5, 3:3}. The key-value pair is used to update the model parameters at the corresponding position. Using key-value pairs clearly indicates the parameter position and the new model parameters that need to be updated, ensuring the accuracy of the model update.

[0059] Step S104: sending the key-value pair to the intelligent robot, wherein the key-value pair is used to update the target detection model deployed on the intelligent robot according to the importance binary code and the new model parameters.

[0060] In this step, if Figure 3 The following diagram shows the detection update of an object detection model. The object detection model recognizes squares, rectangles, or other shapes. Assume that the new object detection model needs to be able to represent them with bounding boxes of different colors. This means that the new object detection model and the target detection model differ by one color representation task. When training the new task using the model parameters of the original task as initial values, only the intermediate layer information related to color and garbage bags changes, leaving the information unrelated to recognizing square, rectangular, and other shaped paper boxes. Therefore, continuous learning can be achieved by updating only a few parameters.

[0061] In one embodiment, Figure 1 Based on the embodiment shown, the above step S101 includes:

[0062] Receiving a target image sent by the intelligent robot, where the target image is an image detected by the target detection model with a confidence level lower than a preset threshold;

[0063] A plurality of the target images are combined into the dataset.

[0064] In this embodiment, the target image can be an image in the current training set with poor object recognition performance, a new image that is not in the training set but is very similar to the training set, or a new image that is not in the training set but may need to be labeled later. For example, if the current training set contains an image of a white cardboard box, but in the real scene, there is a white book on the ground that is not in the training set, the object detection model will generally recognize the white book as a white cardboard box, resulting in a low confidence score.

[0065] It's important to note that target images need to be annotated. The dataset that data engineers use isn't the massive amount of data from the past, but rather data with poor recognition performance. This data might be new objects being recognized as old ones, or old objects being poorly recognized. Once a certain number of target images are reached, a dataset is formed for model update training.

[0066] This embodiment uses low-confidence images collected by the intelligent robot in actual scenarios as training samples to enable the target detection model to be applicable to more application scenarios and enhance the model performance of the target detection model.

[0067] In one embodiment, Figure 1 Based on the embodiment shown, the above step S102 includes:

[0068] Iteratively training the target detection model according to the data set using a preset confidence upper bound algorithm until the initial target detection model reaches a preset convergence condition, stopping the iteration, and obtaining new model parameters;

[0069] Determine the importance binary codes corresponding to the new model parameters according to the new model parameters and the original model parameters of the target detection model.

[0070] In this embodiment, the model update training is performed according to the confidence upper bound UCB algorithm. The initial parameters of the model are the model parameters trained last time, and the initial model importance binary codes are all 0. When training the model, the network usually has many layers, and these layers have a large number of parameters that need to be trained. After the first large-scale training task, the network parameters have been determined. Then, after adding a new task (new target data set), the network parameters will change to varying degrees. The parameters with a lower degree of change should be the parameters that are highly correlated with the previous task, while the parameters with a higher degree of change are the parameters related to the new task. The definition of importance is the degree of change of the parameters. This embodiment compares the new model parameters with the original model parameters to determine the importance of the new model parameters, thereby facilitating the subsequent intelligent robot to selectively update the model and effectively avoiding the catastrophic problem of forgetting the model.

[0071] Optionally, determining the importance binary codes corresponding to the model parameters according to the new model parameters and the original model parameters of the target detection model includes:

[0072] determining the variance between the new model parameters and the original model parameters;

[0073] According to the variance, a binary code of the importance of the new model parameter is determined.

[0074] In this alternative approach, each layer of the model has variance, and the UCB algorithm is related to variance. A larger variance indicates a higher learning rate for the parameters in that layer, indicating greater information uncertainty in that layer. Therefore, the importance of new model parameters with variances greater than the preset variance is binary-coded to 0, while the importance of new model parameters with variances less than the preset variance is binary-coded to 1.

[0075] In one embodiment, Figure 1 Based on the embodiment shown, the above step S103 includes:

[0076] Arranging the new model parameters to obtain an arrangement order, and determining an arrangement position of each new model parameter in the arrangement order;

[0077] Determine the importance binary code as a target new model parameter of a preset code;

[0078] The target new model parameter and the arrangement position of the target new model parameter are combined to obtain the key-value pair.

[0079] In this embodiment, if Figure 2 The model parameter update diagram shown in Figure 1 is a diagram. The object detection model has three original model parameters: [1.1, 1.2, 1.3], with their importance binary encoded as [1, 0, 0]. After model training is complete, the model parameters become [1, 1.3, 1.8]. The original model parameter 1.1 has an importance of 1, so it is not updated and remains fixed. The difference between the original model parameter 1.2 and the new model parameter 1.3 is small, indicating that both the new and old tasks consider this model parameter important. Therefore, the new model parameters and parameter importance binary encoding become [1.1, 1.3, 1.8] and [1, 1, 0]. As can be seen, only the model parameters in the second and third positions are updated, which can be integrated into the key-value pair {2:1.3, 3:1.8}.

[0080] Please refer to Figure 4 , Figure 4The present invention provides a flowchart of a target detection model update method. The present invention can be applied to an intelligent robot, which includes a processor and a memory. The memory is used to store a computer program. When the computer program is executed by the processor, the following is achieved: Figure 4 The update method shown, the intelligent robot includes but is not limited to inspection robots, welcoming robots and cleaning robots and other equipment.

[0081] like Figure 4 As shown, the target detection model updating method includes steps S401 to S402:

[0082] Step S401: receiving a key-value pair sent by a computer device, wherein the key-value pair includes a new model parameter and a binary code of importance corresponding to the new model parameter, wherein the key-value pair is obtained by the computer device using a preset confidence upper bound algorithm to update the object detection model according to the data set, and pairing the updated new model parameter with the binary code of importance;

[0083] Step S402: updating the target detection model deployed on the intelligent robot according to the importance binary code and the new model parameters.

[0084] It is understandable that Figure 4 The embodiment shown is Figure 1 The embodiments shown have the same technical features, so the explanation of steps S401 and S402 can be found in the above Figure 1 The relevant description of the illustrated embodiment will not be repeated here.

[0085] In order to execute the target detection model update method corresponding to the above method embodiment, to achieve the corresponding functions and technical effects. Figure 5 , Figure 5 The following is a block diagram of a target detection model update device provided by an embodiment of the present application. For ease of explanation, only the parts related to this embodiment are shown. The target detection model update device provided by an embodiment of the present application includes:

[0086] An acquisition module 501 is used to acquire a data set for updating a target detection model;

[0087] A first updating module 502 is configured to update the object detection model according to the data set using a preset confidence upper bound algorithm to obtain new model parameters and binary codes of importance corresponding to the new model parameters;

[0088] A pairing module 503 is used to pair the new model parameters with the importance binary codes to obtain key-value pairs;

[0089] The sending module 504 is used to send the key-value pair to the intelligent robot, and the key-value pair is used to update the target detection model deployed on the intelligent robot according to the importance binary code and the new model parameters.

[0090] In one embodiment, the first update module 502 includes:

[0091] A training unit, applying a preset confidence upper bound algorithm to iteratively train the target detection model according to the data set until the initial target detection model reaches a preset convergence condition, stopping the iteration, and obtaining new model parameters;

[0092] The first determining unit is used to determine the importance binary code corresponding to the new model parameters according to the new model parameters and the original model parameters of the target detection model.

[0093] In one embodiment, the first determining unit includes:

[0094] A first determining subunit, configured to determine the variance between the new model parameters and the original model parameters;

[0095] The second determining subunit is configured to determine the importance binary code of the new model parameters according to the variance.

[0096] In one embodiment, the pairing module 503 includes:

[0097] a second determining unit, configured to arrange the new model parameters to obtain an arrangement order, and determine an arrangement position of each new model parameter in the arrangement order;

[0098] A third determining unit is used to determine the target new model parameter whose importance binary code is a preset code;

[0099] The first combining unit is configured to combine the target new model parameter with the arrangement position of the target new model parameter to obtain the key-value pair.

[0100] In one embodiment, the acquisition module 501 includes:

[0101] a receiving unit, configured to receive a target image sent by the intelligent robot, wherein the target image is an image detected by the target detection model with a confidence level lower than a preset threshold;

[0102] The second combining unit is configured to combine a plurality of the target images into the data set.

[0103] In order to execute the target detection model update method corresponding to the above method embodiment, to achieve the corresponding functions and technical effects. Figure 6 , Figure 6 The following is a block diagram of a target detection model update device provided by an embodiment of the present application. For ease of explanation, only the parts related to this embodiment are shown. The target detection model update device provided by an embodiment of the present application includes:

[0104] A receiving module 601 is configured to receive a key-value pair sent by a computer device, wherein the key-value pair includes a new model parameter and a binary code of importance corresponding to the new model parameter. The key-value pair is obtained by pairing the updated new model parameter with the binary code of importance by the computer device using a preset confidence upper bound algorithm to update the object detection model based on the data set.

[0105] The second updating module 602 is used to update the target detection model deployed on the intelligent robot according to the importance binary code and the new model parameters.

[0106] The above-mentioned target detection model updating device can implement the target detection model updating method of the above-mentioned method embodiment. The optional options in the above-mentioned method embodiment are also applicable to this embodiment and will not be described in detail here. The remaining contents of the embodiment of this application can refer to the contents of the above-mentioned method embodiment and will not be repeated in this embodiment.

[0107] Figure 7 This is a schematic diagram of the structure of a computer device provided in one embodiment of the present application. Figure 7 As shown, the computer device 7 of this embodiment includes: at least one processor 70 ( Figure 7 Only one is shown in the figure) a processor, a memory 71, and a computer program 72 stored in the memory 71 and executable on the at least one processor 70, wherein the processor 70 implements the steps of any of the above method embodiments when executing the computer program 72.

[0108] The computer device 7 may be a computing device such as a smart phone, a tablet computer, a desktop computer, a cloud server, etc. The computer device may include but is not limited to a processor 70 and a memory 71. It will be understood by those skilled in the art that Figure 7 This is merely an example of the computer device 7 and does not constitute a limitation on the computer device 7 . The computer device 7 may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, it may also include input and output devices, network access devices, etc.

[0109] The processor 70 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor may be a microprocessor or any conventional processor.

[0110] In some embodiments, the memory 71 may be an internal storage unit of the computer device 7, such as a hard disk or memory of the computer device 7. In other embodiments, the memory 71 may also be an external storage device of the computer device 7, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the computer device 7. Furthermore, the memory 71 may include both an internal storage unit of the computer device 7 and an external storage device. The memory 71 is used to store an operating system, application programs, a boot loader, data, and other programs, such as the program code of the computer program. The memory 71 may also be used to temporarily store data that has been output or is about to be output.

[0111] In addition, an embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in any of the above method embodiments are implemented.

[0112] An embodiment of the present application provides a computer program product. When the computer program product is run on a terminal device, the terminal device implements the steps in the above-mentioned method embodiments when executing the computer program product.

[0113] In several embodiments provided in the present application, it is understood that each box in the flow chart or block diagram can represent a part of a module, program segment or code, and the part of the module, program segment or code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, which depends on the functions involved.

[0114] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a terminal device to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, and other media that can store program codes.

[0115] The specific embodiments described above further illustrate the objectives, technical solutions, and beneficial effects of this application. It should be understood that the above descriptions are merely specific embodiments of this application and are not intended to limit the scope of protection of this application. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of this application by those skilled in the art should be included within the scope of protection of this application.

Claims

1. A method for updating a target detection model, characterized in that: Applied to computer equipment, the updating method includes: Obtaining a data set for updating the target detection model; the data set is low-confidence images collected by the intelligent robot in actual scenes; Using a preset confidence upper bound algorithm, the target detection model is updated according to the data set to obtain new model parameters and binary codes of importance corresponding to the new model parameters, including: using a preset confidence upper bound algorithm, iteratively training the target detection model according to the data set until the target detection model reaches a preset convergence condition, stopping the iteration, and obtaining new model parameters; determining the binary codes of importance corresponding to the new model parameters according to the new model parameters and original model parameters of the target detection model; Pairing the new model parameters with the importance binary codes to obtain key-value pairs; The key-value pair is sent to the intelligent robot, and the key-value pair is used to update the target detection model deployed on the intelligent robot according to the importance binary code and the new model parameters.

2. The method for updating the target detection model according to claim 1, wherein: The step of determining the importance binary codes corresponding to the new model parameters according to the new model parameters and the original model parameters of the target detection model includes: determining the variance between the new model parameters and the original model parameters; According to the variance, a binary code of the importance of the new model parameter is determined.

3. The method for updating the target detection model according to claim 1, wherein: The pairing of the new model parameters and the importance binary codes to obtain key-value pairs includes: Arranging the new model parameters to obtain an arrangement order, and determining an arrangement position of each new model parameter in the arrangement order; Determine the importance binary code as a target new model parameter of a preset code; The target new model parameter and the arrangement position of the target new model parameter are combined to obtain the key-value pair.

4. The method for updating the target detection model according to claim 1, wherein: The obtaining of a data set for updating the target detection model includes: Receiving a target image sent by the intelligent robot, where the target image is an image detected by the target detection model with a confidence level lower than a preset threshold; A plurality of the target images are combined into the dataset.

5. A method for updating a target detection model, characterized in that: Applied to an intelligent robot, the updating method includes: Receiving a key-value pair sent by a computer device, the key-value pair comprising new model parameters and binary codes of importance corresponding to the new model parameters, including: using a preset confidence upper bound algorithm to iteratively train the target detection model according to a data set until the target detection model reaches a preset convergence condition, stopping iteration, and obtaining new model parameters; determining the binary codes of importance corresponding to the new model parameters according to the new model parameters and original model parameters of the target detection model; the key-value pair is obtained by pairing the updated new model parameters with the binary codes of importance by the computer device using a preset confidence upper bound algorithm to update the target detection model according to the data set; the data set is low-confidence images collected by an intelligent robot in an actual scene; The target detection model deployed on the intelligent robot is updated according to the importance binary code and the new model parameters.

6. A device for updating a target detection model, characterized in that: include: An acquisition module is used to obtain a dataset for updating the target detection model; A first updating module is configured to update the target detection model using a preset confidence upper bound algorithm based on the data set to obtain new model parameters and binary codes of importance corresponding to the new model parameters, including: iteratively training the target detection model using the preset confidence upper bound algorithm based on the data set until the target detection model reaches a preset convergence condition, stopping iteration, and obtaining new model parameters; and determining binary codes of importance corresponding to the new model parameters based on the new model parameters and original model parameters of the target detection model; the data set is low-confidence images collected by the intelligent robot in actual scenarios; a pairing module, configured to pair the new model parameters with the importance binary codes to obtain key-value pairs; A sending module is used to send the key-value pair to the intelligent robot, and the key-value pair is used to update the target detection model deployed on the intelligent robot according to the importance binary code and the new model parameters.

7. A device for updating a target detection model, characterized in that: include: A receiving module is configured to receive a key-value pair sent by a computer device, the key-value pair comprising new model parameters and binary codes of importance corresponding to the new model parameters, including: using a preset confidence upper bound algorithm to iteratively train the target detection model according to a data set until the target detection model reaches a preset convergence condition, stopping iteration, and obtaining new model parameters; determining the binary codes of importance corresponding to the new model parameters according to the new model parameters and the original model parameters of the target detection model; the key-value pair is obtained by pairing the updated new model parameters with the binary codes of importance by the computer device using a preset confidence upper bound algorithm to update the target detection model according to the data set; the data set is low-confidence images collected by an intelligent robot in an actual scenario; The second updating module is used to update the target detection model deployed on the intelligent robot according to the importance binary code and the new model parameters.

8. A computer device, characterized in that: The method comprises a processor and a memory, wherein the memory is used to store a computer program, and when the computer program is executed by the processor, the method for updating the target detection model according to any one of claims 1 to 4 is implemented.

9. An intelligent robot, characterized in that: The method comprises a processor and a memory, wherein the memory is used to store a computer program, and when the computer program is executed by the processor, the method for updating the target detection model as claimed in claim 5 is implemented.

Images