Image recognition model self-adaptive adjustment, recognition method, device and system

Abstract

The disclosure provides an image recognition model adaptive adjustment, recognition method, device and system, the method responds to the target area of the image to be identified, using a pre-trained model, predicting the initial classification features of the image to be identified; the initial classification features include a probability matrix of each classification result; the classification result represents the category of the target object included in the image to be identified; by analyzing the correlation between the initial classification features, the probability of each classification result is corrected to obtain the target classification feature, which improves the accuracy of the target classification feature to a certain extent, and then the loss value of the model can be calculated based on the target classification feature and the preset parameter, and the model parameters of the model are adjusted based on the loss value, so as to optimize the model, improve the accuracy of the model in different use environments, and enhance the robustness of the model.

Description

Technical Field

[0001] This disclosure relates to the field of image processing technology, specifically to an image recognition model adaptive adjustment, recognition method, apparatus, and system. Background Technology

[0002] 3D perception refers to acquiring data about a target area through sensors and other devices, and then using pre-trained models to reconstruct the 3D physical world based on that data. For example, in the field of autonomous driving, images of the target area can be acquired using cameras deployed on the vehicle. A pre-trained image recognition model can then be used to identify the images, obtaining information such as the depth, pose, and object category of the target objects included in the images. Based on this information, the physical environment of the target area corresponding to the vehicle can be constructed to enable functions such as path planning.

[0003] During the use of the model, the differences in the environment in which the vehicle is located will lead to differences in the acquired images. As a result, the model has low accuracy in image recognition and cannot accurately construct the physical environment of the target area corresponding to the vehicle, thus failing to achieve accurate path planning and other functions. Summary of the Invention

[0004] This disclosure proposes an adaptive adjustment method, model, and computer device for image recognition models.

[0005] The first aspect of this disclosure proposes an adaptive adjustment method for an image recognition model. In response to acquiring an image to be recognized of a target region, a pre-trained model is used to predict the initial classification features of the image to be recognized. The initial classification features include probability matrices corresponding to each classification result. The classification result represents the category of the target object included in the image to be recognized.

[0006] Based on the correlation between the initial classification features, the probabilities of each classification result are corrected to obtain the target classification features;

[0007] The loss value of the model is calculated based on the target classification features and preset parameters;

[0008] The model parameters are adjusted based on the loss value to optimize the model.

[0009] In this embodiment of the disclosure, the step of correcting the probabilities of each classification result based on the correlation between the initial classification features to obtain the target classification features includes:

[0010] The probability distribution of the initial classification features is corrected to obtain correction parameters;

[0011] The initial classification features are adjusted based on the correction parameters to obtain the target classification features.

[0012] In this embodiment of the disclosure, the step of correcting the probability distribution of the initial classification features to obtain correction parameters includes:

[0013] The probability matrix is ​​linearly coupled to obtain the outer product parameter; wherein the outer product parameter is used to enhance the influence of the positive sample label.

[0014] The diagonal matrix of the probability matrix is ​​subjected to probability regularization to obtain a balance parameter; the balance parameter is used to eliminate the influence of negative sample labels.

[0015] The correction parameter is calculated based on the outer product parameter and the balance parameter.

[0016] In this embodiment of the disclosure, the outer product parameter is calculated in the following manner:

[0017] A = (1 - logp)·pp T

[0018] Where A is the outer product parameter; p is the probability matrix.

[0019] In this embodiment of the disclosure, the balance parameter is calculated in the following manner:

[0020] B = logp·diag(p)

[0021] Where B is the balance parameter; diag(p) ​​is the diagonal matrix of the probability matrix.

[0022] In this embodiment of the disclosure, calculating the loss value of the model based on the target classification features and preset parameters includes:

[0023] The target classification features are used as the classification features of the model, and the loss value is determined using the model's preset loss function.

[0024] A second aspect of this disclosure provides an image recognition method, the method comprising:

[0025] Acquire the image of the target region to be identified;

[0026] The image to be identified is identified according to a pre-trained model, and the pose information and category information of the target object included in the content of the image to be identified are obtained. The pre-trained model is optimized according to the image recognition model adaptive adjustment method as described in the first aspect or any optional embodiment of the first aspect.

[0027] A third aspect of this disclosure provides an adaptive adjustment device for an image recognition model, the device comprising:

[0028] An initial prediction module is used to predict the initial classification features of the image to be identified using a pre-trained model in response to the acquisition of the image to be identified of the target region; the initial classification features include the probability matrix corresponding to each classification result; the classification result represents the category of the target object included in the image to be identified;

[0029] The correction module is used to correct the probability of each classification result based on the correlation between the initial classification features to obtain the target classification features;

[0030] The loss calculation module is used to calculate the loss value of the model based on the target classification features and preset parameters;

[0031] An optimization module is used to adjust the model parameters of the model based on the loss value in order to optimize the model.

[0032] A fourth aspect of this disclosure provides an image recognition apparatus, the apparatus comprising:

[0033] The image acquisition module is used to acquire the image to be identified of the target area;

[0034] The recognition module is used to recognize the image to be recognized according to a pre-trained model, and to obtain the pose information and category information of the target object included in the content of the image to be recognized, wherein the pre-trained model is optimized according to the image recognition model adaptive adjustment method as described in the first aspect or any optional embodiment of the first aspect.

[0035] A fifth aspect of this disclosure describes an image recognition system, comprising: a terminal device and a processor; the terminal device includes an image acquisition component and a transmission component.

[0036] The image acquisition component is used to acquire the image to be identified in the target area;

[0037] The transmission component is used to transmit the image to be identified to the processor;

[0038] The processor is configured to invoke the method described in the second aspect to identify the pose information and category information of the target object included in the content of the image to be identified, or to invoke the method described in the first aspect or any optional implementation thereof to optimize the model.

[0039] A sixth aspect of this disclosure provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being executed by the processor to cause the electronic device to perform the methods described in the first aspect, any alternative implementation of the first aspect, or the second aspect.

[0040] According to a seventh aspect of the present application, a computer-readable storage medium is provided having a computer program stored thereon, the program being executed by a processor to implement the method as described in the first aspect, any optional implementation of the first aspect, or the second aspect.

[0041] The technical solutions provided in this disclosure have at least the following technical effects or advantages:

[0042] In response to acquiring an image of the target region to be identified, a pre-trained model is used to predict the initial classification features of the image to be identified. The initial classification features include a probability matrix of each classification result. The classification result represents the category of the target object included in the image to be identified. By analyzing the correlation between the initial classification features, the probabilities of each classification result are corrected to obtain the target classification features, which improves the accuracy of the target classification features to a certain extent. Then, the loss value of the model can be calculated based on the target classification features and preset parameters, and the model parameters of the model can be adjusted based on the loss value to optimize the model, improve the accuracy of the model in different usage environments, and enhance the robustness of the model.

[0043] Additional aspects and advantages of this disclosure will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this disclosure. Attached Figure Description

[0044] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this disclosure. Furthermore, the same reference numerals denote the same parts throughout the drawings.

[0045] In the attached diagram:

[0046] Figure 1 This illustration shows a schematic diagram of the structure of an image recognition model adaptive adjustment system provided in an embodiment of the present disclosure;

[0047] Figure 2 A flowchart of an image recognition model adaptive adjustment method provided in an embodiment of this disclosure is shown;

[0048] Figure 3 A flowchart of an image recognition method provided in an embodiment of this disclosure is shown;

[0049] Figure 4 A schematic diagram of an image recognition model adaptive adjustment device provided in an embodiment of the present disclosure is shown;

[0050] Figure 5A schematic diagram of an image recognition device provided in one embodiment of the present disclosure is shown;

[0051] Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of the present disclosure is shown;

[0052] Figure 7 A schematic diagram of a storage medium provided according to an embodiment of the present disclosure is shown. Detailed Implementation

[0053] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0054] It should be noted that, unless otherwise stated, the technical or scientific terms used in this disclosure shall have the ordinary meaning as understood by one of ordinary skill in the art to which this disclosure pertains.

[0055] This disclosure proposes an image recognition system, such as... Figure 1 As shown, the image recognition system of this embodiment includes a terminal device 101 and a processor 102, wherein the terminal device 101 includes an image acquisition component 1011 and a transmission component 1012.

[0056] Image acquisition component 1011 is used to acquire the image to be identified in the target area;

[0057] The transmission unit 1012 is used to transmit the image to be recognized to the processor;

[0058] The processor 102 is used by the image recognition method to identify the pose information and category information of the target object included in the content of the image to be recognized, or to call the image recognition model adaptive adjustment method to optimize the model.

[0059] In some embodiments, the image recognition system proposed in this disclosure can be applied to or deployed in a three-dimensional perception system, for example, it can be applied to fields such as autonomous driving of vehicles and path planning of robots. This disclosure uses autonomous driving as an example. The terminal device 101 can be a vehicle, the image acquisition component 1011 can be a camera device in the vehicle for acquiring images of the vehicle's surroundings, and the transmission component 1012 can be a data bus for communication with the processor 102, or a wireless communication device, such as a Bluetooth device, a wireless transmission device, or a 5G communication device. The transmission component 1012 is used to transmit the images acquired by the image acquisition component 1011 to the processor 102, so that the processor 102 can identify the vehicle's surrounding environment and reconstruct the physical environment around the vehicle to achieve the corresponding functions of autonomous driving. This disclosure does not limit the specific forms of the terminal device, image acquisition component, and transmission component; those skilled in the art can determine them according to the actual situation.

[0060] The processor 102 may be an integrated circuit chip deployed in a vehicle, used to execute a program corresponding to an image recognition method or a program for an image recognition model adaptive adjustment method. In some embodiments, the processor 101 may also be a cloud server, a cloud service cluster, an external computing device, etc. It is used to recognize images sent by a transmission device and to adaptively adjust the model of the recognized image to optimize the model and improve the accuracy of model recognition.

[0061] This disclosure proposes an adaptive adjustment method for an image recognition model, such as... Figure 2 As shown, it may include the following steps:

[0062] In step S21, in response to acquiring the image to be identified in the target region, the initial classification features of the image to be identified are predicted using a pre-trained model.

[0063] The initial classification features include the probability matrix corresponding to each classification result; the classification result represents the category of the target object included in the image to be identified.

[0064] In the field of autonomous driving, due to changes in the vehicle's environment, the acquired images for identification can be blurry due to environmental interference and equipment differences. Furthermore, in practical use, the classification labels corresponding to the images are often unknown, leading to inaccurate identification of objects within the images and consequently, inaccurate reconstruction of the vehicle's surrounding physical environment. Specifically, in image recognition within autonomous driving systems, the classification labels can be the categories of target objects included in the image content, such as pedestrians, vehicles, and road obstacles. Vehicles can be further categorized into trucks, cars, construction vehicles, and freight trucks.

[0065] In this embodiment, the pre-trained model is a model that has been trained based on accurate samples. This pre-trained model can be a MonoFlex, MonoGround, PVRCNN, or DSVT model, etc. In practical use, the pre-trained model can be used to obtain the initial classification features of the image to be identified. These initial classification features represent the category of the target object corresponding to the content of the image to be identified. Based on the preceding content, these can be the probabilities corresponding to pedestrians, vehicles, and roadblocks, respectively. For example, the initial classification features can be represented as [20%, 35%, 45%], where 20% is the probability of a pedestrian, 35% is the probability of a vehicle, and 45% is the probability of a roadblock.

[0066] In the actual recognition process, the image to be recognized can also be preprocessed. For example, the size of the image to be recognized can be adjusted to fit the input of the model. Then, the pixel values ​​of the image to be recognized can be standardized, such as linearly mapping the pixel values ​​from the integer range of 0-255 to the floating-point range of [-1, 1] or [0, 1] to eliminate the influence of differences in illumination, contrast and other factors on the model prediction.

[0067] In unsupervised recognition, the model determines initial classification features based solely on the image to be recognized, and the lack of accurate labels (classification features) may lead to errors. Therefore, in this embodiment, the initial classification features can be further corrected to optimize the model during use, making it more suitable for the current application environment and improving the model's recognition accuracy.

[0068] In step S22, the probabilities of each classification result are corrected based on the correlation between the initial classification features to obtain the target classification features.

[0069] For example, the correlation between initial classification features can be calculated using historical data from the model, by statistically analyzing the co-occurrence frequency of different classification results in real-world scenarios. For instance, the co-occurrence probability of the "pedestrian" and "vehicle" categories is relatively high, while the co-occurrence probability of the "roadblock" and "vehicle" categories approaches zero. Statistical methods are then used to quantify the correlation between the probability vectors of each classification result, such as the Pearson correlation coefficient or the Jaccard similarity coefficient. An N×N correlation matrix is ​​constructed, where the matrix elements R0... i,j This represents the correlation between the results of class i and class j, with a value ranging from [-1, 1]. A correlation threshold is set based on the noise level of the application scenario. For example, in complex background scenarios, the threshold is lowered to retain weak correlations; in simple scenarios, the threshold is increased to filter noise.

[0070] Furthermore, regarding the correlation coefficient R... i,jFor classification features with a value <0, if both classification probabilities are high in the initial probability, then the probability values ​​of both are reduced proportionally to suppress contradictory results; for the correlation coefficient R... i,j For class pairs with a correlation coefficient R > 0 and a co-occurrence frequency exceeding a threshold, if the probability of one class is significant, the probability of the other class is increased proportionally to the strength of the correlation. i,j For categories whose probability is approximately 0, keep their initial probability unchanged to avoid irrelevant interference.

[0071] To adjust the strength of the correlation coefficient, a probability adjustment function can be defined to correct the probability of each classification result. Specifically, linear or non-linear adjustment methods can be used; this disclosure does not limit the adjustment method.

[0072] In some embodiments, step S22 can be implemented as follows: the probability distribution of the initial classification features is corrected to obtain correction parameters; the initial classification features are adjusted based on the correction parameters to obtain target classification features.

[0073] For example, the initial classification features are P = [p1, p2, ..., pn]. T Let n be the number of categories, and pi be the initial probability distribution of the i-th category. Calculate the covariance matrix Σ of p, and calculate the inverse matrix W = Σ. -1 To obtain the correction parameter W. Let y0 = Σ -1 p is used as the target classification feature.

[0074] If the covariance matrix Σ is not invertible, then a regularization term ∈I is added to obtain an invertible matrix Σ′=Σ+∈I, ensuring the feasibility of matrix inversion, where I is the identity matrix and ∈ is a small positive number.

[0075] In some other embodiments, the correction parameter can be determined by: linearly coupling the probability matrix to obtain an outer product parameter; wherein the outer product parameter is used to enhance the influence of positive sample labels; performing probability regularization on the diagonal matrix of the probability matrix to obtain a balance parameter; the balance parameter is used to eliminate the influence of negative sample labels; and calculating the correction parameter based on the outer product parameter and the balance parameter.

[0076] A = (1 - logp)·pp T

[0077] Where A is the outer product parameter; p is the probability matrix.

[0078] B = logp·diag(p)

[0079] Where B is the balance parameter; diag(p) ​​is the diagonal matrix of the probability matrix.

[0080] Therefore, the target classification feature y0≈(I+γ(1-logp)·pp) T -γlogp·diag(p)) -1 p

[0081] Where I is the identity matrix; γ is the first preset weight hyperparameter, which is usually set to 4.

[0082] In step S23, the loss value of the model is calculated based on the target classification features and preset parameters.

[0083] For example, the Focal Loss mechanism can be used to calculate the loss value of the model in the embodiments of this disclosure.

[0084] L Focal = -α(1-p) γ ·y0log(p)

[0085] Where L Focal Let represent the loss value, where α is the second preset weight hyperparameter, usually set to 2; γ is the first preset weight hyperparameter, usually set to 4; y0 is the target classification feature; and p is the initial classification feature.

[0086] Substituting y0 into the equation yields:

[0087]

[0088] In step S24, the model parameters are adjusted based on the loss value to optimize the model. This can be achieved through backpropagation, which calculates the partial derivatives of the parameters at each layer, starting from the loss function and working up to the input layer, thus updating the model online. The updated model is then used to recognize the next image to be identified.

[0089] In practical applications, the execution flow of this image recognition model adaptive adjustment method can be as follows:

[0090] Obtain the image of the target region to be identified, and the pre-trained model; input the image to be identified into the pre-trained model for recognition, and obtain the probability matrix p corresponding to each classification result of each candidate target; based on p, use the above formula y0≈(I+γ(1-logp)·pp T -γlogp·diag(p)0 -1 p calculates the classification feature y0 for each target. L is constructed using p and y0. Focal The model parameters are then adjusted based on the calculated loss value.

[0091] The above process is repeated for each new input image to be recognized, achieving dynamic adaptation. It can be embedded into existing inference processes without requiring modifications to the model structure or additional supervision signals, exhibiting good versatility and deployment friendliness.

[0092] The image recognition model adaptive adjustment method provided in this disclosure, in response to acquiring the image to be recognized of the target region, uses a pre-trained model to predict the initial classification features of the image to be recognized; the initial classification features include the probability matrix of each classification result; the classification result represents the category of the target object included in the image to be recognized; by analyzing the correlation between the initial classification features, the probability of each classification result is corrected to obtain the target classification features, which improves the accuracy of the target classification features to a certain extent. Then, the loss value of the model can be calculated based on the target classification features and preset parameters, and the model parameters can be adjusted based on the loss value to optimize the model, improve the accuracy of the model in different usage environments, and enhance the robustness of the model.

[0093] The image recognition model adaptive adjustment method of this disclosure enables model optimization without label input, broadening the applicability of the model in practical deployments and making it suitable for scenarios with missing data labels or high uncertainty. By calculating the correlation between initial classification features, the training focus area is automatically adjusted to enhance the optimization effect of the initial classification features, reduce the false negative rate, and improve the detection integrity and stability of the model in complex scenarios. The image recognition model adaptive adjustment method of this disclosure does not require adjustment of the model structure and can achieve adaptive updates of the model during use, enhancing model compatibility and improving model deployment efficiency.

[0094] This disclosure also proposes an image recognition method, such as... Figure 3 As shown, the image recognition method of this disclosure embodiment may include the following steps:

[0095] Step S31: Obtain the image to be recognized of the target area;

[0096] Step S32: Identify the image to be identified based on the pre-trained model to obtain the pose information and category information of the target object included in the content of the image to be identified. The pre-trained model is optimized according to the above-mentioned image recognition model adaptive adjustment method.

[0097] For example, during model usage, after obtaining the pose and category information of the target object, the model parameters are adjusted based on the loss value to optimize the model. This allows the previously optimized model to be used for recognizing the next image to be recognized, and further optimization can be performed. Specifically, the optimization method is the same as in the above embodiments, and will not be repeated here.

[0098] Corresponding to the above implementation of the image recognition model adaptive adjustment method, this disclosure also provides an image recognition model adaptive adjustment device for performing the above-described method. Figure 2 An image recognition model adaptive adjustment method illustrated in any embodiment. For example... Figure 4 As shown, the image recognition model adaptive adjustment device includes:

[0099] The initial prediction module 401 is used to predict the initial classification features of the image to be identified using a pre-trained model in response to the acquisition of the image to be identified of the target region; the initial classification features include the probability matrix corresponding to each classification result; the classification result represents the category of the target object included in the image to be identified;

[0100] The correction module 402 is used to correct the probability of each classification result based on the correlation between the initial classification features to obtain the target classification features;

[0101] The loss calculation module 403 is used to calculate the loss value of the model based on the target classification features and preset parameters;

[0102] The optimization module 404 is used to adjust the model parameters of the model based on the loss value in order to optimize the model.

[0103] The image recognition model adaptive adjustment device provided in the above embodiments of this disclosure and the image recognition model-based adaptive adjustment method provided in the embodiments of this disclosure are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the applications stored therein.

[0104] Corresponding to the above-described image recognition methods, this disclosure also provides an image recognition device for performing the above-described... Figure 3 Image recognition methods of any of the illustrated embodiments. For example... Figure 5 As shown, the language recognition device includes:

[0105] Image acquisition module 501 is used to acquire the image to be recognized of the target area;

[0106] The recognition module 502 is used to recognize the image to be recognized according to the pre-trained model, and obtain the pose information and category information of the target object included in the content of the image to be recognized, wherein the pre-trained model is optimized according to the above-mentioned image recognition model adaptive adjustment method.

[0107] The language recognition device provided in the above embodiments of this disclosure and the image recognition method provided in the embodiments of this disclosure are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the applications stored therein.

[0108] This disclosure also provides an electronic device for performing the above-described data transmission method. Please refer to... Figure 6 This illustrates a schematic diagram of an electronic device provided by some embodiments of the present disclosure. For example... Figure 6 As shown, the electronic device includes: a processor 600, a memory 601, a bus 602, and a communication interface 603. The processor 600, the communication interface 603, and the memory 601 are connected via the bus 602. The memory 601 stores a computer program that can run on the processor 600. When the processor 600 runs the computer program, it executes the aforementioned provisions of this disclosure. Figure 2 or Figure 3 The method provided by any of the illustrated embodiments.

[0109] The memory 601 may include high-speed random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 603 (which can be wired or wireless), such as the Internet, wide area network, local area network, or metropolitan area network.

[0110] Bus 602 can be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into address bus, data bus, control bus, etc. Memory 601 is used to store programs. After receiving execution instructions, processor 600 executes the program, as described above. Figure 2 or Figure 3 The method disclosed in any of the illustrated embodiments can be applied to or implemented by the processor 600.

[0111] The processor 600 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of the processor 600 or by instructions in software form. The processor 600 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this disclosure. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this disclosure can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules may reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory 601. Processor 600 reads the information in memory 601 and, in conjunction with its hardware, completes the steps of the above method.

[0112] The electronic device and the data transmission method provided in this disclosure are based on the same inventive concept and have the same beneficial effects as the methods they employ, operate, or implement.

[0113] This disclosure also provides a computer-readable storage medium corresponding to the data transmission method provided in the foregoing embodiments. Please refer to... Figure 7 The computer-readable storage medium shown is an optical disc 30, on which a computer program (i.e., a program product) is stored. When the computer program is run by the microprocessor, it executes the data transmission method provided in any of the aforementioned embodiments.

[0114] It should be noted that examples of computer-readable storage media may also include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other optical and magnetic storage media, which will not be elaborated here.

[0115] The computer-readable storage medium provided in the above embodiments of this disclosure and the data transmission method provided in the embodiments of this disclosure are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the applications stored therein.

[0116] It should be noted that:

[0117] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of this disclosure may be practiced without these specific details. In some instances, well-known structures and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0118] Similarly, it should be understood that, in order to simplify this disclosure and aid in understanding one or more of the various inventive aspects, in the above description of exemplary embodiments of this disclosure, various features of this disclosure are sometimes grouped together in a single embodiment, figure, or description thereof. However, this approach to disclosure should not be construed as reflecting a schematic diagram in which the claimed disclosure requires more features than are expressly recited in each claim. Rather, as reflected in the following claims, inventive aspects lie in fewer than all features of a single foregoing disclosed embodiment. Therefore, the claims following the detailed description are hereby expressly incorporated into that detailed description, wherein each claim itself is a separate embodiment of this disclosure.

[0119] Furthermore, those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are intended to be within the scope of this disclosure and form different embodiments. For example, in the following claims, any of the claimed embodiments can be used in any combination.

[0120] The above are merely preferred embodiments of this disclosure, but the scope of protection of this disclosure is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this disclosure should be included within the scope of protection of this disclosure. Therefore, the scope of protection of this disclosure should be determined by the scope of the claims.

Claims

1. An adaptive adjustment method for an image recognition model, characterized in that, The method includes: In response to acquiring an image of the target region to be identified, a pre-trained model is used to predict the initial classification features of the image to be identified; the initial classification features include a probability matrix corresponding to each classification result; the classification result represents the category of the target object included in the image to be identified; Based on the correlation between the initial classification features, the probabilities of each classification result are corrected to obtain the target classification features; The loss value of the model is calculated based on the target classification features and preset parameters; The model parameters are adjusted based on the loss value to optimize the model; The step of correcting the probabilities of each classification result based on the correlation between the initial classification features to obtain the target classification features includes: The probability distribution of the initial classification features is corrected to obtain correction parameters; The initial classification features are adjusted based on the correction parameters to obtain the target classification features; The process of correcting the probability distribution of the initial classification features to obtain correction parameters includes: The probability matrix is ​​linearly coupled to obtain the outer product parameter; wherein the outer product parameter is used to enhance the influence of the positive sample label. The diagonal matrix of the probability matrix is ​​subjected to probability regularization to obtain a balance parameter; the balance parameter is used to eliminate the influence of negative sample labels. The correction parameter is calculated based on the outer product parameter and the balance parameter; The outer product parameter is calculated in the following way: A=(1 logp) pp T Where A is the outer product parameter; p is the probability matrix; The balance parameters are calculated in the following manner: B=logp diag(p) Where B is the balance parameter; diag(p) ​​is the diagonal matrix of the probability matrix.

2. The method according to claim 1, characterized in that, The step of calculating the loss value of the model based on the target classification features and preset parameters includes: The target classification features are used as the classification features of the model, and the loss value is determined using the model's preset loss function.

3. An image recognition method, characterized in that, The method includes: Acquire the image of the target region to be identified; The image to be identified is identified based on a pre-trained model, and the pose information and category information of the target object included in the content of the image to be identified are obtained, wherein the pre-trained model is optimized according to the image recognition model adaptive adjustment method of claim 1 or 2.

4. An adaptive adjustment device for an image recognition model, characterized in that, The device includes: An initial prediction module is used to predict the initial classification features of the image to be identified using a pre-trained model in response to the acquisition of the image to be identified of the target region; the initial classification features include the probability matrix corresponding to each classification result; the classification result represents the category of the target object included in the image to be identified; The correction module is used to correct the probability of each classification result based on the correlation between the initial classification features to obtain the target classification features; The loss calculation module is used to calculate the loss value of the model based on the target classification features and preset parameters; An optimization module is used to adjust the model parameters of the model based on the loss value in order to optimize the model; The correction module is further configured to: correct the probability distribution of the initial classification features to obtain correction parameters; The initial classification features are adjusted based on the correction parameters to obtain the target classification features; The process of correcting the probability distribution of the initial classification features to obtain correction parameters includes: The probability matrix is ​​linearly coupled to obtain the outer product parameter; wherein the outer product parameter is used to enhance the influence of the positive sample label. The diagonal matrix of the probability matrix is ​​subjected to probability regularization to obtain a balance parameter; the balance parameter is used to eliminate the influence of negative sample labels. The correction parameter is calculated based on the outer product parameter and the balance parameter; The outer product parameter is calculated in the following way: A=(1 logp) pp T Where A is the outer product parameter; p is the probability matrix; The balance parameters are calculated in the following manner: B=logp diag(p) Where B is the balance parameter; diag(p) ​​is the diagonal matrix of the probability matrix.

5. An image recognition device, characterized in that, The device includes: The image acquisition module is used to acquire the image to be identified of the target area; The recognition module is used to recognize the image to be recognized according to a pre-trained model, and to obtain the pose information and category information of the target object included in the content of the image to be recognized, wherein the pre-trained model is optimized according to the image recognition model adaptive adjustment method of claim 1 or 2.

6. An image recognition system, characterized in that, The system includes a terminal device and a processor; the terminal device includes an image acquisition component and a transmission component. The image acquisition component is used to acquire the image to be identified in the target area; The transmission component is used to transmit the image to be identified to the processor; The processor is configured to invoke the method of claim 3 to identify the pose information and category information of the target object included in the content of the image to be identified, or to invoke the method of claim 1 or 2 to optimize the model.

Images