Image recognition method and device, electronic equipment, storage medium and product

Abstract

Embodiments of the present disclosure disclose an image recognition method and device, electronic equipment, storage medium and product, wherein the method comprises: obtaining at least one data paradigm of a target object to be identified image; input the to-be-identified image into a preset image recognition model to obtain a recognition result; wherein the preset image recognition model is generated based on a plurality of preset paradigm sample images and a dynamic fusion feature corresponding to each preset paradigm and a dynamic target loss function. The technical scheme of the embodiments of the present disclosure solves the problem that when a plurality of image recognition models for identifying single paradigm images are combined for use, the accuracy of the recognition result is not high, and can improve the accuracy of the recognition result of the image recognition model, and is safer in application.

Description

Technical Field

[0001] This disclosure relates to the field of image recognition technology, and in particular to an image recognition method, apparatus, electronic device, storage medium, and product. Background Technology

[0002] In scenarios requiring user authentication, deep learning models can be used for image recognition to determine whether a face in an image is a real face, i.e., facial liveness detection. Currently, the input data for algorithms that identify silent liveness images are all based on a single image paradigm, such as a face image with a fixed cropping ratio, a frequency domain image after wavelet transform, a depth image, or an infrared image.

[0003] During application, it was found that models trained with different input paradigms and structures can capture different non-liveness cue features. To identify cue features from various attack methods in facial liveness detection applications, models with different algorithm structures can be combined. However, each algorithm model can falsely intercept genuine liveness images, and the image recognition accuracy of simple multi-structure model combinations still needs improvement. Summary of the Invention

[0004] This disclosure provides an image recognition method, apparatus, electronic device, storage medium, and product that can recognize target objects from images of various data formats, thereby improving the accuracy of image recognition.

[0005] In a first aspect, embodiments of this disclosure provide an image recognition method, the method comprising:

[0006] Obtain the image to be recognized corresponding to at least one data paradigm of the target object;

[0007] The image to be recognized is input into a preset image recognition model to obtain the recognition result of the image to be recognized;

[0008] The preset image recognition model is generated based on sample images of multiple preset paradigms, as well as the dynamic fusion features and dynamic target loss functions corresponding to each preset paradigm.

[0009] Secondly, embodiments of this disclosure also provide an image recognition device, the device comprising:

[0010] The image acquisition module is used to acquire the image to be recognized corresponding to at least one data paradigm of the target object;

[0011] An image recognition module is used to input the image to be recognized into a preset image recognition model to obtain the recognition result of the image to be recognized;

[0012] The preset image recognition model is generated based on sample images of multiple preset paradigms, as well as the dynamic fusion features and dynamic target loss functions corresponding to each preset paradigm.

[0013] Thirdly, embodiments of this disclosure also provide an electronic device, the electronic device comprising:

[0014] One or more processors;

[0015] Storage device for storing one or more programs.

[0016] When the one or more programs are executed by the one or more processors, the one or more processors implement the image recognition method as described in any of the embodiments of this disclosure.

[0017] Fourthly, embodiments of this disclosure also provide a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform the image recognition method as described in any of the embodiments of this disclosure.

[0018] Fifthly, this disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements the image recognition method as described in any of the embodiments of the present invention.

[0019] This disclosure embodiment obtains an image to be recognized corresponding to at least one data paradigm of a target object; and inputs the image to be recognized of at least one paradigm into a preset image recognition model to obtain the recognition result of the image to be recognized; wherein, the preset image recognition model is trained and generated based on sample images of multiple preset paradigms and the dynamic fusion features and dynamic target loss functions corresponding to each preset paradigm. The technical solution of this disclosure embodiment solves the problem in the prior art that the image recognition accuracy is not high when multiple image recognition models trained based on a single data paradigm sample are simply combined and used. It enables the image recognition model to perform image recognition on images of a single data paradigm, as well as to comprehensively analyze and learn images of the same target object from multiple data paradigms, thereby improving the accuracy of the recognition result of the image recognition model and making it safer in application. Attached Figure Description

[0020] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0021] Figure 1This is a schematic flowchart of an image recognition method provided in an embodiment of this disclosure;

[0022] Figure 2 This is an image to be identified corresponding to different cropping ratios provided in the embodiments of this disclosure;

[0023] Figure 3 This is a schematic diagram of an image recognition model training method provided in an embodiment of this disclosure;

[0024] Figure 4 This is a schematic diagram of an image recognition model training process provided in an embodiment of this disclosure;

[0025] Figure 5 This is a schematic diagram of the structure of an image recognition device provided in an embodiment of this disclosure;

[0026] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure. Detailed Implementation

[0027] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0028] It should be understood that the steps described in the method embodiments of this disclosure may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of this disclosure is not limited in this respect.

[0029] The term "comprising" and its variations as used herein are open-ended inclusions, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the description below.

[0030] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0031] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0032] It is understood that before using the technical solutions disclosed in the various embodiments of this disclosure, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in this disclosure in an appropriate manner in accordance with relevant laws and regulations, and user authorization should be obtained.

[0033] For example, upon receiving a user's active request, a prompt message is sent to the user to explicitly inform them that the requested operation will require the acquisition and use of the user's personal information. This allows the user to independently choose whether to provide personal information to the software or hardware, such as the electronic device, application, server, or storage medium performing the operations of this disclosed technical solution, based on the prompt message.

[0034] As an optional but non-limiting implementation, in response to a user's active request, sending a prompt message to the user can be done via a pop-up window, where the prompt message can be presented in text format. Furthermore, the pop-up window can also include a selection control allowing the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0035] It is understood that the above notification and user authorization process are merely illustrative and do not constitute a limitation on the implementation of this disclosure. Other methods that comply with relevant laws and regulations may also be applied to the implementation of this disclosure.

[0036] It is understood that the data involved in this technical solution (including but not limited to the data itself, the acquisition or use of the data) shall comply with the requirements of relevant laws, regulations and related provisions.

[0037] Figure 1 This is a flowchart illustrating an image recognition method provided in an embodiment of the present disclosure. This embodiment is applicable to situations where identity authentication is achieved based on facial recognition, particularly in scenarios where multiple paradigms of images of the target object can be obtained for image recognition. The method can be executed by an image recognition device, which can be implemented in software and / or hardware, or optionally, by an electronic device, such as a mobile terminal, a PC, or a server.

[0038] like Figure 1 As shown, the image recognition method includes:

[0039] S110. Obtain the image to be recognized corresponding to at least one data paradigm of the target object.

[0040] The target object can be an object that needs to be identified through facial recognition. Correspondingly, the image to be identified can be an image containing the face of the target object. Identity verification can be achieved by determining whether the face of the target object in the image to be identified is a live face image.

[0041] The image to be identified can be an image captured in scenarios requiring identity recognition and authentication, such as account login or transaction information verification, where facial recognition verification is required, and the user's face image is captured in real time through the camera of the application client terminal. Of course, the image to be identified could also be a fake live face image that could attack a preset image recognition algorithm; therefore, it is necessary to identify whether the captured image is a genuine live face image. Alternatively, the technical solution of this embodiment can also be applied to other scenarios requiring live face image detection for image recognition.

[0042] Data paradigms can be understood as the representation of an image to be recognized. For example, the time-domain and frequency-domain representations of the image to be recognized, or multiple data paradigm images to be recognized obtained by cropping an image of one paradigm according to different cropping ratios. Figure 2 In the images shown, after processing the images according to different cropping ratios, the proportion of the facial region of the target object to the overall image region varies in different image paradigms. Furthermore, imaging devices using different imaging principles acquire images of the target object, resulting in different data paradigms for the images to be identified. For example, there are different modalities such as RGB images obtained from conventional cameras and near-infrared images obtained from infrared cameras.

[0043] The purpose of recognizing images from multiple paradigms is based on the discovery in the design of deep learning-based silent liveness detection algorithms that image recognition models derived from different input image paradigms exhibit feature selectivity. For example, when the face in an image is tightly cropped (the face region occupies a large proportion of the total image area), the image recognition model tends to capture moiré pattern information; when the face is loosely cropped (the face region occupies a small proportion of the total image area), the image recognition model tends to capture edge information of the reproduced medium; and when using the frequency domain information of the image to be recognized as input, the image recognition model tends to capture the material information of the printed paper. Therefore, by using images from multiple data paradigms as input, the image recognition model can capture more image features, thereby improving the accuracy of image recognition.

[0044] Specifically, in the process of acquiring the image to be recognized corresponding to at least one data paradigm of the target object, image acquisition can be performed using at least one image acquisition device. Images acquired by different image acquisition devices correspond to different data paradigms. For example, the image to be recognized, acquired in at least one data paradigm of the target object, can be a conventional RGB image, a near-infrared image, or a depth representation of image information.

[0045] Furthermore, at least one preset paradigm of the target object image to be recognized can be obtained, such as the RGB image paradigm; then, the preset paradigm image to be recognized is cropped according to at least one preset cropping ratio to obtain RGB images with multiple cropping paradigms. Alternatively, the preset paradigm image to be recognized can be transformed to obtain an image to be recognized with at least one data paradigm of the target object, such as obtaining the image representation of the corresponding data paradigm through wavelet transform or Fourier transform.

[0046] S120. Input the image to be recognized into a preset image recognition model to obtain the recognition result of the image to be recognized.

[0047] The image to be recognized input into the preset image recognition model can be an image corresponding to one data paradigm, or an image corresponding to two or more data paradigms. Ultimately, the data paradigm of the image to be recognized input into the preset image recognition model can be one or more data paradigms randomly selected from a variety of preset, available data paradigms.

[0048] The preset image recognition model is trained and generated based on sample images of multiple preset paradigms, as well as the dynamic fusion features and dynamic objective loss functions corresponding to each preset paradigm. Under the constraints of the dynamic fusion features and dynamic objective functions of sample images of multiple paradigms, the preset image recognition model can not only recognize images of a single paradigm, but also recognize images of multiple paradigms.

[0049] Specifically, the training process of the preset image recognition model includes:

[0050] Step 1: Input each sample image in the preset model training sample set, and / or at least one preset paradigm sample image from all preset paradigm sample images associated with each sample image, into the image recognition model to be trained.

[0051] Typically, during the learning process of a model algorithm, the sample images in the existing pre-set model training sample set are usually RGB images. During the sample construction process, the sample images in the pre-set model training sample set can be preprocessed to obtain multiple sets of extended sample images with different data paradigms corresponding to each sample image.

[0052] During the image sample input process, a single paradigm sample image or multiple paradigm sample images can be input. The types of paradigms are not limited, and input sample images can be randomly selected from the corresponding sample image set. This allows the image recognition model to learn multiple data paradigms of images to be recognized simultaneously, resulting in a more comprehensive analytical capability and improved resistance to external algorithm attacks.

[0053] Step 2: Based on the recognition results and dynamic weights of each input sample image output by the image recognition model to be trained, determine the dynamic fusion features of each input sample image and the dynamic target loss of the image recognition model to be trained.

[0054] Specifically, when the input sample image is a data paradigm image to be recognized, the image recognition model to be trained first extracts the feature map of the input sample image. Then, the model's classification judgment network structure outputs the recognition result of the sample image and the corresponding dynamic weights, and obtains the corresponding image recognition output loss as the dynamic target loss.

[0055] When the number of input sample images is greater than one, the image recognition model to be trained first extracts feature maps from each input sample image. Then, the feature maps are processed by the classification network structure of the image recognition model to output the recognition result, image recognition output loss, and corresponding dynamic weights for each sample image. Furthermore, feature fusion can be performed on the features in each feature map based on the dynamic weight values, and the fused features are classified. Finally, the dynamic target loss is obtained by combining the image recognition output loss of each sample image and the image recognition loss of the fused feature classification result. In other words, the recognition result of the image is determined based on a comprehensive analysis of the features of sample images from different data paradigms of the same target object. In each batch of model training, the data paradigm of the input sample images may be different, so the data paradigm for feature fusion is also different, and the corresponding weight values ​​are also different. Therefore, the obtained fused feature is a dynamic fused feature.

[0056] Step 3: Iteratively update the image recognition model to be trained based on the dynamic target loss to obtain the preset image recognition model.

[0057] That is, based on the comprehensive analysis of the image features of sample images from different data paradigms of the same target object, the recognition result of the image is determined, and the parameters of the image recognition model to be trained are continuously updated until the loss value of the dynamic target loss function converges or the preset model training rounds are reached.

[0058] The technical solution of this disclosure involves obtaining an image to be recognized corresponding to at least one data paradigm of a target object; and inputting the image to be recognized of at least one paradigm into a preset image recognition model to obtain the recognition result of the image to be recognized. The training process of the preset image recognition model involves inputting at least one preset paradigm sample image from each sample image in the preset model training sample set and / or from all preset paradigm sample images associated with each sample image into the image recognition model to be trained. The image recognition model to be trained then extracts feature maps of sample images of different paradigms and performs image recognition. Furthermore, based on the image recognition output loss and dynamic weights of the image recognition model to be trained for each input sample image, the dynamic target loss of the image recognition model to be trained is determined, and the image recognition model to be trained is iteratively updated according to the dynamic target loss to obtain the preset image recognition model. The technical solution of this disclosure solves the problem in the prior art that the image recognition accuracy is not high when multiple image recognition models trained based on a single data paradigm sample are simply combined and used. It enables the image recognition model to perform image recognition on images of a single data paradigm, as well as to perform comprehensive analysis and learning on images of the same target object from multiple data paradigms. This can improve the accuracy of the recognition results of the image recognition model and make it safer in application.

[0059] Figure 3 This is a flowchart illustrating an image recognition model training method provided in this embodiment. The flowchart further describes the training process of a preset image recognition model, enabling training of the model using sample images of various paradigms. This method can be executed by a model training module in an image recognition device, which can be implemented in software and / or hardware, optionally through an electronic device such as a mobile terminal, PC, or server.

[0060] like Figure 3 As shown, the image recognition model training method includes:

[0061] S210. Input each sample image in the preset model training sample set, and / or at least one preset paradigm sample image from all preset paradigm sample images associated with each sample image, into the image recognition model to be trained.

[0062] During model training, the input sample images for model training can be dynamically selected. For each original sample image in the preset model training sample set, a preset extended paradigm sample image with a quantity value less than the total number of preset extended paradigm types can be randomly selected from the preset extended paradigm sample image set. The original sample image and the corresponding randomly selected preset extended paradigm sample image are then input into the image recognition model to be trained.

[0063] For example, in one feasible implementation, for each original sample image in the preset model training set, image processing can be performed to obtain ten corresponding data paradigm sample images, which are then included in the preset extended paradigm sample image set. When inputting sample images, fewer than or equal to ten sample images can be randomly selected from the ten data paradigm sample images and input into the image recognition model to be trained. This is equivalent to using a dropout method, randomly discarding certain data paradigms for training. The paradigm type of the sample images input into the model to be trained may be different each time, which can further improve the reliability of the image recognition algorithm. Because, for malicious users, even if they happen to find a specific vulnerability, the dynamic paradigm input combination can effectively resist malicious attacks, thereby greatly improving the security of the live face recognition algorithm.

[0064] S220. Through the feature extraction network of the image recognition model to be trained, feature extraction is performed on each input sample image to obtain feature maps of sample images of different paradigms.

[0065] The image recognition model to be trained includes one or both of the following structures: residual learning network structure or ShuffleNet convolutional neural network structure. This serves as the feature extraction network, used to extract image features from each input sample image. For multiple input sample images with different data paradigms, the feature extraction network is shared, utilizing the same feature extraction logic to extract features and obtain feature maps for sample images of different paradigms.

[0066] S230. Input each of the feature maps into the feature classification network of the corresponding paradigm to obtain the classification result and dynamic weight value of each feature map.

[0067] The feature classification network (corresponding to the Head in the network structure) is used to further learn the extracted feature maps, extract more features, and classify and predict those features. Each feature classification network can consist of multiple fully connected layers. Specifically, in this embodiment, the number of feature classification networks connected to the feature extraction network is multiple, the same as the number of all data paradigms used for model training; that is, each data paradigm corresponds to one feature classification network. The structure of each feature classification network can be the same or different. Based on this, the trained image recognition model can recognize a single input image from any data paradigm and obtain the corresponding recognition result.

[0068] Each feature classification network outputs two contents, including a floating-point weight value and an image recognition result score. The dynamic weight is the result learned by the to-be-trained image recognition model. In this embodiment, the weight value is referred to as a dynamic weight value. It can be understood that the data paradigm types of the sample images input to the to-be-trained image recognition model each time are different, and the output results obtained by different feature classification networks are dynamically changing.

[0069] S240: Perform feature fusion according to the dynamic weight values to obtain the fused feature corresponding to each of the feature maps and the fused feature classification result.

[0070] In this step, based on the dynamic weight values, the features of the sample images of different data paradigms are subjected to feature fusion, which can comprehensively utilize the features of the sample images of multiple data paradigms, realize the complementary advantages of multiple features, and enable the trained image recognition model to obtain more robust and accurate recognition results. Therefore, the use effect of the image recognition model in this embodiment is better than the use effect of a simple combination of multiple single-paradigm image recognition models obtained by separately training sample images of different paradigms.

[0071] Specifically, in the process of performing feature fusion, first, each dynamic weight value is input into a preset classifier to obtain a corresponding dynamic weight value adjustment value. The preset classifier is a Softmax classifier, which processes each dynamic weight value so that the sum of the dynamic weight values corresponding to the sample images of each data paradigm in the current training batch is 1. Then, the weighted sum of each dynamic weight value adjustment value and the to-be-classified features input to the classifier in the feature classification network of the corresponding paradigm is calculated to obtain the fused feature. The to-be-classified features may be the result of further feature extraction on the feature map input into the feature classification network by the feature classification network.

[0072] Further, by inputting the fused feature into the preset classifier (Softmax classifier), the fused feature classification result and the corresponding image recognition output loss (such as Cross Entropy Loss, CE Loss) can be obtained.

[0073] It can be understood that during the model training process, there are N weights for N feature classification networks. During the model training or application process, according to the computing capability of the processor, the actual model input may be input images corresponding to M data paradigms (M<N), and the sum of the M weights after the softmax operation is still 1, that is, for the same image, if different paradigms are selected, the finally obtained fused features are also different.

[0074] S250: Determine the dynamic target loss based on the image recognition output loss corresponding to the classification result of each of the feature maps and the image recognition output loss corresponding to the fused feature classification result.

[0075] For each data paradigm's sample image, its corresponding classification result will have a corresponding image recognition output loss, such as CE Loss. The dynamic target loss can be determined by the weighted sum of the image recognition output loss corresponding to the feature map of the input image of each data paradigm and the image recognition output loss corresponding to the fused feature classification result; where the weight values ​​of each image recognition output loss are preset hyperparameters.

[0076] S260. Iteratively update the image recognition model to be trained according to the dynamic target loss to obtain the preset image recognition model.

[0077] The image recognition model trained through the above steps can adaptively extract features based on the data paradigm of the input image to complete the image recognition task.

[0078] In a feasible model training example, you can refer to Figure 4 The model training process is shown. During model training, sample images corresponding to any optional data paradigm can be used as input to the model, such as sample images with different cropping ratios of the original RGB sample images, frequency domain information after wavelet transform, or even multimodal input information (such as depth information, near-infrared information, etc.). Figure 4 The image only shows the sample images corresponding to the two cropping ratios and the frequency domain information corresponding to the original sample images.

[0079] The input sample images of each data paradigm share the same feature extraction network. This feature extraction network can be any convolutional neural network used for feature extraction, such as ResNet, ShuffleNet, or other network structures.

[0080] After feature extraction by the feature extraction network, the feature extraction results of input sample images from different data paradigms will be further classified and recognized by the corresponding feature classification network. After passing through the feature classification network, each input sample image of each data paradigm will obtain an image recognition classification result and a dynamic weight. Figure 4 The dynamic weights corresponding to the input sample images of the three data paradigms are represented as dynamic weight A, dynamic weight B, and dynamic weight C. The image classification result is the result obtained after further feature extraction in the feature classification network is input into the classifier (Softmax classifier), and the corresponding CE Loss will also be obtained.

[0081] Furthermore, the dynamic weights A, B, and C are processed using the Softmax function to obtain the dynamic weight adjustment values ​​corresponding to each dynamic weight, ensuring that the sum of all weights is 1. Then, for the feature layer before the classifier's Softmax function ( Figure 4The features represented by line filling are multiplied by the dynamically adjusted weights after Softmax processing, and then summed to obtain the final fused features. Figure 4 (Features represented by black fill in the middle). The fused features are then input into the Softmax classifier to obtain the recognition result of the fused features.

[0082] Finally, the dynamic target loss of the model to be trained is calculated by weighting the cross-entropy losses of the input sample images from multiple data paradigms and the cross-entropy loss of the fused features as the dynamic target loss. The weight parameters of each cross-entropy can be preset hyperparameters, and each hyperparameter can be assigned a value in a 1:1 ratio.

[0083] The image recognition model obtained through the above training process can comprehensively utilize multiple data paradigms to effectively improve algorithm security. These data paradigms can be dynamically combined.

[0084] The technical solution of this disclosure involves inputting at least one preset paradigm sample image from each sample image in the preset model training sample set and / or from all preset paradigm sample images associated with each sample image into an image recognition model to be trained. The image recognition model then extracts feature maps from the sample images of different paradigms and performs image recognition. Furthermore, the features of the extracted sample images from different data paradigms are fused. Based on the image recognition output loss of the image recognition model for each input sample image and the image recognition output loss corresponding to the classification result of the fused features, the dynamic target loss of the image recognition model to be trained is determined. The image recognition model to be trained is then iteratively updated according to the dynamic target loss to obtain the preset image recognition model. This technical solution solves the problem in the prior art where the image recognition accuracy is low when multiple image recognition models trained based on a single data paradigm are simply combined. It enables the image recognition model to perform image recognition on images of a single data paradigm or to comprehensively analyze and learn multiple data paradigm images of the same target object, thereby improving the accuracy of the recognition results and making the application safer.

[0085] Figure 5 This is a schematic diagram of the structure of an image recognition device provided in an embodiment of the present disclosure. The device is suitable for identity authentication based on facial recognition. The image recognition device can be implemented in the form of software and / or hardware and can be configured in an electronic device, such as a mobile terminal, a PC, or a server.

[0086] like Figure 5 As shown, the image recognition device includes an image acquisition module 310 and an image recognition module 320.

[0087] The image acquisition module 310 is used to acquire an image to be recognized corresponding to at least one data paradigm of the target object; the image recognition module 320 is used to input the image to be recognized into a preset image recognition model to obtain the recognition result of the image to be recognized, wherein the preset image recognition model is trained and generated based on sample images of multiple preset paradigms and dynamic fusion features and dynamic target loss functions corresponding to each preset paradigm.

[0088] Optionally, based on any optional technical solution in the embodiments of this disclosure, the image acquisition module 310 is specifically used for:

[0089] An image of the target object in at least one data paradigm is acquired using at least one image acquisition device.

[0090] Based on any optional technical solution in the embodiments of this disclosure, optionally, the images to be identified in different data paradigms correspond to images acquired by different image acquisition devices.

[0091] Optionally, based on any of the optional technical solutions in the embodiments of this disclosure, the image acquisition module 310 may also be used for:

[0092] Obtain at least one preset paradigm image of the target object to be recognized;

[0093] The image to be identified is cropped according to at least one preset cropping ratio, and / or the image to be identified according to the preset paradigm is converted into data to obtain an image to be identified in at least one data paradigm of the target object.

[0094] Based on any optional technical solution in the embodiments of this disclosure, the image recognition device may optionally further include: a model training module, which includes a sample input submodule, a sample learning submodule, and a model update submodule;

[0095] The sample input submodule is used to input each sample image in the preset model training sample set, and / or at least one preset paradigm sample image from all preset paradigm sample images associated with each sample image, into the image recognition model to be trained.

[0096] The sample learning submodule is used to determine the dynamic fusion features of each input sample image and the dynamic target loss of the image recognition model to be trained based on the recognition results and dynamic weights of each input sample image output by the image recognition model to be trained.

[0097] The model update submodule is used to iteratively update the image recognition model to be trained based on the dynamic target loss to obtain the preset image recognition model.

[0098] Optionally, based on any optional technical solution in the embodiments of this disclosure, the sample input submodule is specifically used for:

[0099] For each original sample image in the preset model training sample set, a preset extended paradigm sample image with a quantity value less than the total number of preset extended paradigm types is randomly selected from the preset extended paradigm sample image set.

[0100] The original sample image and / or the corresponding randomly selected preset extended paradigm sample image are input into the image recognition model to be trained.

[0101] Optionally, based on any optional technical solution in the embodiments of this disclosure, the sample learning submodule is specifically used for:

[0102] The feature extraction network of the image recognition model to be trained is used to extract features from each input sample image to obtain feature maps of sample images of different paradigms.

[0103] Each of the aforementioned feature maps is input into a feature classification network of the corresponding paradigm to obtain the classification result and dynamic weight value of each feature map;

[0104] Based on the dynamic weight values, feature fusion is performed to obtain the fused features corresponding to each feature map, as well as the fused feature classification results;

[0105] The dynamic target loss is determined based on the image recognition output loss corresponding to the classification results of each feature map and the image recognition output loss corresponding to the classification results of the fused features.

[0106] Optionally, based on any of the optional technical solutions in the embodiments of this disclosure, the sample learning submodule may also be used for:

[0107] Each of the aforementioned dynamic weight values ​​is input into a preset classifier to obtain the corresponding dynamic weight value adjustment value;

[0108] The fused feature is obtained by calculating the weighted sum of the adjustment values ​​of each dynamic weight and the features to be classified in the feature classification network of the corresponding paradigm.

[0109] The fused features are input into the preset classifier to obtain the classification result of the fused features.

[0110] Optionally, based on any of the optional technical solutions in the embodiments of this disclosure, the sample learning submodule may also be used for:

[0111] The weighted sum of the image recognition output loss corresponding to the classification result of each feature map and the image recognition output loss corresponding to the classification result of the fused feature map is determined as the dynamic target loss;

[0112] The weight values ​​for the output loss of each image recognition are preset hyperparameters.

[0113] Based on any optional technical solution in the embodiments of this disclosure, the image recognition model to be trained may optionally include a residual learning network structure and / or a ShuffleNet convolutional neural network structure.

[0114] The technical solution provided in this disclosure obtains an image to be recognized corresponding to at least one data paradigm of a target object; and inputs the image to be recognized of at least one paradigm into a preset image recognition model to obtain the recognition result of the image to be recognized; wherein, the training process of the preset image recognition model is to input at least one preset paradigm sample image from each sample image in the preset model training sample set and / or all preset paradigm sample images associated with each sample image into the image recognition model to be trained, and use the image recognition model to extract feature maps of sample images of different paradigms and perform image recognition; then, based on the image recognition output loss and dynamic weight of the image recognition model to be trained for each input sample image, the dynamic target loss of the image recognition model to be trained is determined, and the image recognition model to be trained is iteratively updated according to the dynamic target loss to obtain the preset image recognition model. The technical solution of this disclosure solves the problem in the prior art that the image recognition accuracy is not high when multiple image recognition models trained based on a single data paradigm sample are simply combined and used. It enables the image recognition model to perform image recognition on images of a single data paradigm, as well as to perform comprehensive analysis and learning on images of the same target object from multiple data paradigms. This can improve the accuracy of the recognition results of the image recognition model and make it safer in application.

[0115] The apparatus provided in this disclosure can execute the method provided in any embodiment of this disclosure, and has the corresponding functional modules and beneficial effects for executing the method.

[0116] It is worth noting that the various units and modules included in the above-mentioned device are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of each functional unit are only for easy differentiation and are not used to limit the protection scope of the embodiments of this disclosure.

[0117] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure. Reference is made below. Figure 6It illustrates an electronic device suitable for implementing embodiments of the present disclosure (e.g., Figure 6 The diagram below shows the structure of the terminal device or server 400. The terminal device in this embodiment may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), and vehicle terminals (e.g., vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. Figure 6 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0118] like Figure 6 As shown, electronic device 400 may include a processing unit (e.g., central processing unit, graphics processor, etc.) 401, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 402 or a program loaded from storage device 408 into random access memory (RAM) 403. The RAM 403 also stores various programs and data required for the operation of electronic device 400. The processing unit 401, ROM 402, and RAM 403 are interconnected via bus 404. An edit / output (I / O) interface 405 is also connected to bus 404.

[0119] Typically, the following devices can be connected to I / O interface 405: input devices 406 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 407 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 408 including, for example, magnetic tapes, hard disks, etc.; and communication devices 409. Communication device 409 allows electronic device 400 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 6 An electronic device 400 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively.

[0120] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication device 409, or installed from storage device 408, or installed from ROM 402. When the computer program is executed by processing device 401, it performs the functions defined in the methods of embodiments of this disclosure.

[0121] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0122] The electronic device provided in this embodiment belongs to the same inventive concept as the image recognition method and image recognition model training method provided in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0123] This disclosure provides a computer storage medium storing a computer program that, when executed by a processor, implements the image recognition method and image recognition model training method provided in the above embodiments.

[0124] It should be noted that the computer-readable medium described in this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this disclosure, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0125] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0126] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device.

[0127] The aforementioned computer-readable medium carries one or more programs that, when executed by the electronic device, cause the electronic device to:

[0128] Obtain the image to be recognized corresponding to at least one data paradigm of the target object;

[0129] The image to be recognized is input into a preset image recognition model to obtain the recognition result of the image to be recognized;

[0130] The preset image recognition model is generated based on sample images of multiple preset paradigms, as well as the dynamic fusion features and dynamic target loss functions corresponding to each preset paradigm.

[0131] Computer program code for performing the operations of this disclosure can be written in one or more programming languages ​​or a combination thereof, including but not limited to object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0132] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0133] The units described in the embodiments of this disclosure can be implemented in software or in hardware. The name of a unit does not necessarily limit the unit itself; for example, the first acquisition unit can also be described as "a unit that acquires at least two Internet Protocol addresses".

[0134] The functions described above in this document can be performed, at least in part, by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application Standard Products (ASSPs), System-on-Chip (SoCs), Complex Programmable Logic Devices (CPLDs), and so on.

[0135] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0136] This disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements the image recognition method provided in any embodiment of this disclosure.

[0137] In implementing a computer program product, computer program code for performing the operations of this disclosure can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0138] According to one or more embodiments of this disclosure, [Example 1] provides an image recognition method, the method comprising:

[0139] Obtain the image to be recognized corresponding to at least one data paradigm of the target object;

[0140] The image to be recognized is input into a preset image recognition model to obtain the recognition result of the image to be recognized;

[0141] The preset image recognition model is generated based on sample images of multiple preset paradigms, as well as the dynamic fusion features and dynamic target loss functions corresponding to each preset paradigm.

[0142] According to one or more embodiments of this disclosure, [Example 2] provides an image recognition method, which further includes:

[0143] In some optional implementations, obtaining the image to be recognized corresponding to at least one data paradigm of the target object includes:

[0144] An image of the target object in at least one data paradigm is acquired using at least one image acquisition device.

[0145] According to one or more embodiments of this disclosure, [Example 3] provides an image recognition method, which further includes:

[0146] In some alternative implementations, the images to be identified in different data paradigms correspond to images acquired by different image acquisition devices.

[0147] According to one or more embodiments of this disclosure, [Example 4] provides an image recognition method, further comprising:

[0148] In some optional implementations, obtaining the image to be recognized corresponding to at least one data paradigm of the target object includes:

[0149] Obtain at least one preset paradigm image of the target object to be recognized;

[0150] The image to be identified is cropped according to at least one preset cropping ratio, and / or the image to be identified according to the preset paradigm is converted into data to obtain an image to be identified in at least one data paradigm of the target object.

[0151] According to one or more embodiments of this disclosure, [Example 5] provides an image recognition method, including:

[0152] In some optional implementations, the training process of the preset image recognition model includes:

[0153] Each sample image in the preset model training sample set, and / or at least one preset paradigm sample image from all preset paradigm sample images associated with each sample image, are input into the image recognition model to be trained.

[0154] Based on the recognition results and dynamic weights of each input sample image output by the image recognition model to be trained, the dynamic fusion features of each input sample image and the dynamic target loss of the image recognition model to be trained are determined.

[0155] The image recognition model to be trained is iteratively updated based on the dynamic target loss to obtain the preset image recognition model.

[0156] According to one or more embodiments of this disclosure, [Example Six] provides an image recognition method, further comprising:

[0157] In some optional implementations, the step of inputting each sample image in the preset model training sample set, and / or at least one preset paradigm sample image from all preset paradigm sample images associated with each sample image, into the image recognition model to be trained includes:

[0158] For each original sample image in the preset model training sample set, a preset extended paradigm sample image with a quantity value less than the total number of preset extended paradigm types is randomly selected from the preset extended paradigm sample image set.

[0159] The original sample image and / or the corresponding randomly selected preset extended paradigm sample image are input into the image recognition model to be trained.

[0160] According to one or more embodiments of this disclosure, [Example Seven] provides an image recognition method, further comprising:

[0161] In some optional implementations, determining the dynamic fusion features of each input sample image and the dynamic target loss of the image recognition model to be trained, based on the recognition results and dynamic weights of each input sample image output by the image recognition model to be trained, includes:

[0162] The feature extraction network of the image recognition model to be trained is used to extract features from each input sample image to obtain feature maps of sample images of different paradigms.

[0163] Each of the aforementioned feature maps is input into a feature classification network of the corresponding paradigm to obtain the classification result and dynamic weight value of each feature map;

[0164] Based on the dynamic weight values, feature fusion is performed to obtain the fused features corresponding to each feature map, as well as the fused feature classification results;

[0165] The dynamic target loss is determined based on the image recognition output loss corresponding to the classification results of each feature map and the image recognition output loss corresponding to the classification results of the fused features.

[0166] According to one or more embodiments of this disclosure, [Example Eight] provides an image recognition method, further comprising:

[0167] In some optional implementations, the step of performing feature fusion based on the dynamic weight values ​​to obtain fused features corresponding to each feature map, and the fused feature classification result, includes:

[0168] Each of the aforementioned dynamic weight values ​​is input into a preset classifier to obtain the corresponding dynamic weight value adjustment value;

[0169] The fused feature is obtained by calculating the weighted sum of the adjustment values ​​of each dynamic weight and the features to be classified in the feature classification network of the corresponding paradigm.

[0170] The fused features are input into the preset classifier to obtain the classification result of the fused features.

[0171] According to one or more embodiments of this disclosure, [Example Nine] provides an image recognition method, further comprising:

[0172] In some optional implementations, determining the dynamic target loss based on the image recognition output loss of the classification results of each feature map and the image recognition output loss of the fused feature classification results includes:

[0173] The weighted sum of the image recognition output loss corresponding to the classification result of each feature map and the image recognition output loss corresponding to the classification result of the fused feature map is determined as the dynamic target loss;

[0174] The weight values ​​for the output loss of each image recognition are preset hyperparameters.

[0175] According to one or more embodiments of this disclosure, [Example 10] provides an image recognition method, further comprising:

[0176] The image recognition model to be trained includes a residual learning network structure and / or a ShuffleNet convolutional neural network structure.

[0177] According to one or more embodiments of this disclosure, [Example 11] provides an image recognition device, including:

[0178] The image acquisition module is used to acquire an image to be recognized corresponding to at least one data paradigm of the target object;

[0179] An image recognition module is used to input the image to be recognized into a preset image recognition model to obtain the recognition result of the image to be recognized;

[0180] The preset image recognition model is generated based on sample images of multiple preset paradigms, as well as the dynamic fusion features and dynamic target loss functions corresponding to each preset paradigm.

[0181] According to one or more embodiments of this disclosure, [Example Twelve] provides an image recognition device, further comprising:

[0182] In one optional implementation, the image acquisition module is specifically used for:

[0183] An image of the target object in at least one data paradigm is acquired using at least one image acquisition device.

[0184] According to one or more embodiments of this disclosure, [Example Thirteen] provides an image recognition device, further comprising:

[0185] In one alternative implementation, the images to be identified in different data paradigms correspond to images acquired by different image acquisition devices.

[0186] According to one or more embodiments of this disclosure, [Example Fourteen] provides an image recognition device, further comprising:

[0187] In an optional implementation, the image acquisition module may also be used for:

[0188] Obtain at least one preset paradigm image of the target object to be recognized;

[0189] The image to be identified is cropped according to at least one preset cropping ratio, and / or the image to be identified according to the preset paradigm is converted into data to obtain an image to be identified in at least one data paradigm of the target object.

[0190] According to one or more embodiments of this disclosure, [Example Fifteen] provides an image recognition device, further comprising: a model training module, the model training module including a sample input submodule, a sample learning submodule, and a model update submodule;

[0191] The sample input submodule is used to input each sample image in the preset model training sample set, and / or at least one preset paradigm sample image from all preset paradigm sample images associated with each sample image, into the image recognition model to be trained.

[0192] The sample learning submodule is used to determine the dynamic fusion features of each input sample image and the dynamic target loss of the image recognition model to be trained based on the recognition results and dynamic weights of each input sample image output by the image recognition model to be trained.

[0193] The model update submodule is used to iteratively update the image recognition model to be trained based on the dynamic target loss to obtain the preset image recognition model.

[0194] According to one or more embodiments of this disclosure, [Example Sixteen] provides an image recognition device, further comprising:

[0195] In one optional implementation, the sample input submodule is specifically used for:

[0196] For each original sample image in the preset model training sample set, a preset extended paradigm sample image with a quantity value less than the total number of preset extended paradigm types is randomly selected from the preset extended paradigm sample image set.

[0197] The original sample image and / or the corresponding randomly selected preset extended paradigm sample image are input into the image recognition model to be trained.

[0198] According to one or more embodiments of this disclosure, [Example Seventeen] provides an image recognition device, further comprising:

[0199] In one optional implementation, the sample learning submodule is specifically used for:

[0200] The feature extraction network of the image recognition model to be trained is used to extract features from each input sample image to obtain feature maps of sample images of different paradigms.

[0201] Each of the aforementioned feature maps is input into a feature classification network of the corresponding paradigm to obtain the classification result and dynamic weight value of each feature map;

[0202] Based on the dynamic weight values, feature fusion is performed to obtain the fused features corresponding to each feature map, as well as the fused feature classification results;

[0203] The dynamic target loss is determined based on the image recognition output loss corresponding to the classification results of each feature map and the image recognition output loss corresponding to the classification results of the fused features.

[0204] According to one or more embodiments of this disclosure, [Example 18] provides an image recognition device, further comprising:

[0205] In an optional implementation, the sample learning module can also be used for:

[0206] Each of the aforementioned dynamic weight values ​​is input into a preset classifier to obtain the corresponding dynamic weight value adjustment value;

[0207] The fused feature is obtained by calculating the weighted sum of the adjustment values ​​of each dynamic weight and the features to be classified in the feature classification network of the corresponding paradigm.

[0208] The fused features are input into the preset classifier to obtain the classification result of the fused features.

[0209] According to one or more embodiments of this disclosure, [Example Nineteen] provides an image recognition device, further comprising:

[0210] In an optional implementation, the sample learning module can also be used for:

[0211] The weighted sum of the image recognition output loss corresponding to the classification result of each feature map and the image recognition output loss corresponding to the classification result of the fused feature map is determined as the dynamic target loss;

[0212] The weight values ​​for the output loss of each image recognition are preset hyperparameters.

[0213] According to one or more embodiments of this disclosure, [Example 20] provides an image recognition device, further comprising:

[0214] In one alternative implementation, the image recognition model to be trained includes a residual learning network structure and / or a ShuffleNet convolutional neural network structure.

[0215] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features disclosed in this disclosure that have similar functions.

[0216] Furthermore, while the operations are described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in a sequential order. In certain environments, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of this disclosure. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0217] Although the subject matter has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely illustrative examples of implementing the claims.

Claims

1. An image recognition method, characterized in that, include: Obtain the image to be recognized corresponding to at least one data paradigm of the target object; The image to be recognized is input into a preset image recognition model to obtain the recognition result of the image to be recognized; The preset image recognition model is generated based on sample images of multiple preset paradigms and the dynamic fusion features and dynamic target loss functions corresponding to each preset paradigm. The training process of the preset image recognition model includes: For each original sample image in the preset model training sample set, a preset extended paradigm sample image with a quantity value less than the total number of preset extended paradigm types is randomly selected from the preset extended paradigm sample image set. The original sample image and / or the corresponding randomly selected preset extended paradigm sample image are then input into the image recognition model to be trained. The feature extraction network of the image recognition model to be trained is used to extract features from each input sample image to obtain feature maps of sample images of different paradigms. Each of the aforementioned feature maps is input into a feature classification network of the corresponding paradigm to obtain the classification result and dynamic weight value of each feature map; Based on the dynamic weight values, feature fusion is performed to obtain the fused features corresponding to each feature map, as well as the fused feature classification results; The dynamic target loss is determined based on the image recognition output loss corresponding to the classification results of each feature map and the image recognition output loss corresponding to the classification results of the fused features.

2. The method according to claim 1, characterized in that, The acquisition of the image to be identified corresponding to at least one data paradigm of the target object includes: An image of the target object in at least one data paradigm is acquired using at least one image acquisition device.

3. The method according to claim 2, characterized in that, The images to be identified in different data paradigms correspond to images acquired by different image acquisition devices.

4. The method according to any one of claims 1-3, characterized in that, The acquisition of the image to be identified corresponding to at least one data paradigm of the target object includes: Obtain at least one preset paradigm image of the target object to be recognized; The image to be identified is cropped according to at least one preset cropping ratio, and / or the image to be identified according to the preset paradigm is converted into data to obtain an image to be identified in at least one data paradigm of the target object.

5. The method according to claim 1, characterized in that, The training process of the preset image recognition model includes: The image recognition model to be trained is iteratively updated based on the dynamic target loss to obtain the preset image recognition model.

6. The method according to claim 1, characterized in that, The step of fusing features based on the dynamic weight values ​​to obtain fused features corresponding to each feature map, and the fused feature classification result, includes: Each of the aforementioned dynamic weight values ​​is input into a preset classifier to obtain the corresponding dynamic weight value adjustment value; The fused feature is obtained by calculating the weighted sum of the adjustment values ​​of each dynamic weight and the features to be classified in the feature classification network of the corresponding paradigm. The fused features are input into the preset classifier to obtain the classification result of the fused features.

7. The method according to claim 1, characterized in that, The dynamic target loss is determined by the image recognition output loss based on the classification results of each feature map and the image recognition output loss based on the fused feature classification results, including: The weighted sum of the image recognition output loss corresponding to the classification result of each feature map and the image recognition output loss corresponding to the classification result of the fused feature map is determined as the dynamic target loss; The weight values ​​for the loss of each image recognition output are preset hyperparameters.

8. The method according to claim 5, characterized in that, The image recognition model to be trained includes a residual learning network structure and / or a ShuffleNet convolutional neural network structure.

9. An image recognition device, characterized in that, include: The image acquisition module is used to acquire an image to be recognized corresponding to at least one data paradigm of the target object; An image recognition module is used to input the image to be recognized into a preset image recognition model to obtain the recognition result of the image to be recognized; The preset image recognition model is generated based on sample images of multiple preset paradigms and the dynamic fusion features and dynamic target loss functions corresponding to each preset paradigm. The training process of the preset image recognition model includes: for each original sample image in the preset model training sample set, randomly selecting a preset extended paradigm sample image whose quantity is less than the total number of preset extended paradigm types in the preset extended paradigm sample image set, and inputting the original sample image and / or the corresponding randomly selected preset extended paradigm sample image into the image recognition model to be trained. The feature extraction network of the image recognition model to be trained is used to extract features from each input sample image to obtain feature maps of sample images of different paradigms. Each of the aforementioned feature maps is input into a feature classification network of the corresponding paradigm to obtain the classification result and dynamic weight value of each feature map; Based on the dynamic weight values, feature fusion is performed to obtain the fused features corresponding to each feature map, as well as the fused feature classification results; The dynamic target loss is determined based on the image recognition output loss corresponding to the classification results of each feature map and the image recognition output loss corresponding to the classification results of the fused features.

10. An electronic device, characterized in that, The electronic device includes: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the image recognition method as described in any one of claims 1-8.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the image recognition method as described in any one of claims 1-8.

12. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the image recognition method as described in any one of claims 1-8.

Images