Shielding object identification method and device, electronic equipment and storage medium
By acquiring the target object and occlusion labeling images, performing occlusion processing and depth separation convolution processing, the problem of low occlusion recognition accuracy is solved, the recognition accuracy and efficiency are improved, and the accurate identification of occlusion is ensured.
Patent Information
- Application Number
- CN202411254282.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-06
- Publication Date
- 2024-12-03
AI Technical Summary
The existing occlusion recognition technology has the problem of low recognition accuracy in the agricultural insurance field. It is mainly due to insufficient training samples and incorrect key point labeling, which leads to the model being prone to errors when identifying occlusions.
A method of occlusion recognition is proposed. By obtaining the target object image and occlusion label image, occlusion processing is performed to generate the object occlusion image, and using depth separation convolution processing and attention weight calculation, the model's feature extraction ability of occlusion is improved and identification accuracy.
It improves the accuracy and efficiency of occlusion recognition, enhances the feature extraction capability of the model, reduces the computational complexity, and ensures the accurate identification of occlusions in the target image.
Smart Images

Figure CN119068512A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the fields of financial technology and artificial intelligence technology, and in particular to a method and device for identifying obstructions, an electronic device, and a storage medium. Background Art
[0002] Occluder recognition refers to the identification of objects that obstruct the target object in an image. For example, in the insurance claims scenario of farmed animals in the field of agricultural insurance, in order to ensure the safety of the photographers photographing farmed animals, the photographers often photograph farmed animals outside the fence. Therefore, the images of farmed animals captured often contain occluders such as fences.
[0003] Currently, common occlusion recognition is mostly performed by judging the occlusion of the target image through a conventional binary classification model. However, there are few training samples containing occlusions and target objects for the binary classification model, or the key points of the target object are incorrectly labeled, resulting in low model accuracy, which easily leads to occlusion recognition errors. Therefore, how to improve the accuracy of occlusion recognition has become a technical problem that needs to be solved urgently. Summary of the invention
[0004] The main purpose of the embodiments of the present application is to provide a method and device for identifying an obstruction, an electronic device and a storage medium, aiming to improve the accuracy of identifying an obstruction.
[0005] To achieve the above objective, a first aspect of an embodiment of the present application provides a method for identifying an obstruction, the method comprising:
[0006] Acquire a target object image containing a target recognition object;
[0007] Acquire an occluder annotated image including a reference occluder and an occluder label, wherein the occluder label is used to indicate a category of the reference occluder;
[0008] Performing occlusion processing on the target object image based on the occlusion object annotation image to obtain an object occlusion image, wherein the object occlusion image includes the target recognition object and the reference occlusion object, and the target recognition object in the object occlusion image is occluded by the reference occlusion object;
[0009] Based on a preset initial occluder recognition model, a depth-separable convolution process is performed on the object occlusion image to obtain image features;
[0010] Based on the image features and the initial occluder recognition model, perform occluder prediction on the object occluded image to obtain an occluder prediction result and a prediction label, wherein the object occluded image includes an occluder prediction region, the occluder prediction result indicates whether the occluder prediction region contains the predicted occluder or does not contain the predicted occluder, and the prediction label is used to indicate the category of the predicted occluder;
[0011] Based on the occlusion prediction result, the occluder prediction area and the prediction label, performing loss calculation on the initial occluder recognition model to obtain model loss data;
[0012] Adjusting parameters of the initial occluder recognition model based on the model loss data to obtain a target occluder recognition model;
[0013] A target image is acquired, and occluders are identified on the target image using the target occluder identification model.
[0014] In some embodiments, performing occluder prediction on the object occluded image based on the image features and the initial occluder recognition model to obtain an occlusion prediction result and a prediction label includes:
[0015] Based on the initial occluder recognition model and the image features, performing occluder detection on the object occluded image to obtain the occluder prediction area;
[0016] Calculating the attention weight of the predicted occluder area to obtain a regional attention weight;
[0017] Calculating the occluder probability of the occluder prediction region based on the regional attention weight, wherein the occluder probability refers to the probability that the occluder prediction region is the predicted occluder;
[0018] Determining the occlusion prediction result based on the occlusion probability;
[0019] Calculating, based on the regional attention weight, an occluder label probability of the predicted occluder in the occluder prediction region;
[0020] Based on the occlusion label probability, a label prediction is performed on the object occlusion image to obtain the predicted label.
[0021] In some embodiments, the performing loss calculation on the initial occluder recognition model based on the occlusion prediction result, the occluder prediction area and the prediction label to obtain model loss data includes:
[0022] Acquire occlusion data of the object occlusion image, wherein the occlusion data includes object occlusion data, occlusion object marking area data and occlusion object label data;
[0023] Calculating occlusion classification loss data of the initial occluder recognition model based on the object occlusion data and the occlusion prediction result;
[0024] Calculating the occluder segmentation loss data of the initial occluder recognition model based on the occluder marking area data and the occluder prediction area;
[0025] Calculating occluder label loss data of the initial occluder recognition model based on the occluder label data and the predicted label;
[0026] The occlusion classification loss data, the occlusion segmentation loss data and the occlusion label loss data are merged to obtain the model loss data.
[0027] In some embodiments, the performing of depth-separable convolution processing on the object occlusion image based on a preset initial occlusion recognition model to obtain image features includes:
[0028] Performing deep convolution processing on the object occlusion image based on the initial occlusion recognition model to obtain deep convolution image features;
[0029] The deep convolution image feature is subjected to point-by-point convolution processing based on the initial occluder recognition model to obtain the image feature.
[0030] In some embodiments, performing point-by-point convolution processing on the deep convolution image feature based on the initial occluder recognition model to obtain the image feature includes:
[0031] Obtaining the number of output channels of the initial occluder recognition model;
[0032] Determine the number of kernels of a preset target convolution kernel based on the number of output channels;
[0033] Based on the target convolution kernel and the number of kernels, cross-channel convolution is performed on the deep convolution image features to obtain the image features.
[0034] In some embodiments, performing occlusion processing on the target object image based on the occlusion annotated image to obtain the object occlusion image includes:
[0035] Performing target segmentation on the target object image to obtain an object image without background;
[0036] Performing target segmentation on the occluder annotated image to obtain an occluder-free background image;
[0037] The object-without-background image and the occlusion-without-background image are fused to obtain the object-occlusion image.
[0038] In some embodiments, the step of fusing the object without background image with the occlusion object without background image to obtain the object occlusion image includes:
[0039] Performing edge detection on the image without background of the occluder to obtain the edge contour of the occluder;
[0040] Based on the edge contour of the occluder, the image without background of the occluder is cropped to obtain an occluder cropped image;
[0041] The occlusion object cropped image is Laplace-fused with the object background-free image to obtain the object occlusion image.
[0042] To achieve the above-mentioned purpose, a second aspect of an embodiment of the present application provides a device for identifying an obstruction, the device comprising:
[0043] An object image acquisition module, used to acquire a target object image containing a target recognition object;
[0044] An occluder image acquisition module, used to acquire an occluder annotated image including a reference occluder and an occluder label, wherein the occluder label is used to indicate the category of the reference occluder;
[0045] A sample image generation module, configured to perform occlusion processing on the target object image based on the occluder annotated image to obtain an object occlusion image, wherein the object occlusion image includes the target recognition object and the reference occluder, and the target recognition object in the object occlusion image is occluded by the reference occluder;
[0046] An image feature extraction module, used to perform a depth-separable convolution process on the object occlusion image based on a preset initial occlusion recognition model to obtain image features;
[0047] an occluder prediction module, configured to perform occluder prediction on the object occluded image based on the image features and the initial occluder recognition model, and obtain an occluder prediction result and a prediction label, wherein the object occluded image includes an occluder prediction region, the occluder prediction result indicates whether the occluder prediction region contains the predicted occluder or does not contain the predicted occluder, and the prediction label is used to indicate the category of the predicted occluder;
[0048] An occlusion loss calculation module, configured to perform loss calculation on the initial occlusion recognition model based on the occlusion prediction result, the occlusion prediction area and the prediction label to obtain model loss data;
[0049] A model parameter adjustment module, used to adjust the parameters of the initial occluder recognition model based on the model loss data to obtain a target occluder recognition model;
[0050] The occluder recognition module is used to acquire a target image and perform occluder recognition on the target image using the target occluder recognition model.
[0051] To achieve the above objectives, a third aspect of an embodiment of the present application proposes an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method described in the first aspect is implemented.
[0052] To achieve the above-mentioned purpose, the fourth aspect of an embodiment of the present application proposes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described in the first aspect is implemented.
[0053] The occlusion recognition method and device, electronic device and storage medium proposed in the present application separately obtain a target object image containing a target recognition object and an occlusion annotated image containing an occlusion reference and an occlusion label, and then use the occlusion annotated image to perform occlusion processing on the target object image to obtain an object occlusion image, thereby generating an object occlusion image containing the target recognition object and the occlusion reference. This can increase the number of samples for model training, thereby improving the accuracy of the model. Furthermore, according to a preset initial occlusion recognition model, a deep separable convolution processing is performed on the object occlusion image to obtain image features, which can enhance the model's feature extraction capability for the object occlusion image and reduce the complexity of the model calculation. , thereby improving the efficiency of occlusion recognition. Secondly, according to the image features and the initial occlusion recognition model, the occlusion prediction is performed on the object occlusion image to obtain the occlusion prediction result and the prediction label. According to the occlusion prediction result, the occlusion prediction area and the prediction label, the loss of the initial occlusion recognition model is calculated to obtain the model loss data. Then, according to the model loss data, the parameters of the initial occlusion recognition model are adjusted to obtain the target occlusion recognition model, which improves the diversity of the model's occlusion recognition results and thus improves the accuracy of occlusion recognition. Finally, the target occlusion recognition model is used to perform occlusion recognition on the acquired target image to ensure the accuracy of occlusion recognition of the target image. BRIEF DESCRIPTION OF THE DRAWINGS
[0054] Figure 1 is a flow chart of the obstruction recognition method provided by an embodiment of the present application;
[0055] Figure 2 yes Figure 1Flow chart of step S103 in FIG.
[0056] Figure 3 yes Figure 2 Flow chart of step S203 in FIG.
[0057] Figure 4 yes Figure 1 Flow chart of step S104 in FIG.
[0058] Figure 5 yes Figure 4 Flow chart of step S402 in FIG.
[0059] Figure 6 yes Figure 1 Flow chart of step S105 in FIG.
[0060] Figure 7 yes Figure 1 Flow chart of step S106 in FIG.
[0061] Figure 8 is a schematic diagram of the structure of an obstruction recognition device provided in an embodiment of the present application;
[0062] Fig. 9 It is a schematic diagram of the hardware structure of the electronic device provided in the embodiment of the present application. DETAILED DESCRIPTION
[0063] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0064] It should be noted that, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first", "second", etc. in the specification, claims and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.
[0065] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.
[0066] First, some nouns involved in this application are analyzed:
[0067] Depthwise separable convolution processing: Depthwise separable convolution is an efficient convolution operation that decomposes traditional convolution into two steps: depthwise convolution and pointwise convolution. Depthwise convolution performs convolution operations independently on each input channel, while pointwise convolution uses a 1×1 convolution kernel to linearly combine features from different channels. This method significantly reduces the number of parameters and computational complexity of the model while maintaining good feature extraction capabilities, and is widely used in lightweight neural networks such as MobileNet.
[0068] Occluder recognition refers to the identification of objects that obstruct the target object in an image. For example, in the insurance claims scenario of farmed animals in the field of agricultural insurance, in order to ensure the safety of the photographers photographing farmed animals, the photographers often photograph farmed animals outside the fence. Therefore, the images of farmed animals captured often contain occluders such as fences.
[0069] Currently, common occlusion recognition is mostly performed by judging the occlusion of the target image through a conventional binary classification model. However, there are few training samples containing occlusions and target objects for the binary classification model, or the key points of the target object are incorrectly labeled, resulting in low model accuracy, which easily leads to occlusion recognition errors. Therefore, how to improve the accuracy of occlusion recognition has become a technical problem that needs to be solved urgently.
[0070] Based on this, the embodiments of the present application provide a method and device for identifying an obstruction, an electronic device, and a storage medium, aiming to improve the accuracy of identifying an obstruction.
[0071] The obstruction recognition method and device, electronic device and storage medium provided in the embodiments of the present application are specifically described through the following embodiments. First, the obstruction recognition method in the embodiments of the present application is described.
[0072] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.
[0073] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.
[0074] The obstruction recognition method provided in the embodiment of the present application relates to the fields of financial technology and artificial intelligence technology. The obstruction recognition method provided in the embodiment of the present application can be applied to a terminal, can be applied to a server side, and can also be software running in a terminal or a server side. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc.; the server side can be configured as an independent physical server, or a server cluster or a distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application that implements the obstruction recognition method, etc., but is not limited to the above forms.
[0075] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.
[0076] Figure 1 is an optional flow chart of the method for identifying an obstruction provided in an embodiment of the present application. Figure 1 The method may include but is not limited to steps S101 to S108.
[0077] Step S101, obtaining a target object image containing a target recognition object;
[0078] Step S102, obtaining an occluder annotated image including a reference occluder and an occluder label, wherein the occluder label is used to indicate a category of the reference occluder;
[0079] Step S103, performing occlusion processing on the target object image based on the occlusion object annotation image to obtain an object occlusion image, wherein the object occlusion image includes the target recognition object and the reference occlusion object, and the target recognition object in the object occlusion image is occluded by the reference occlusion object;
[0080] Step S104, based on a preset initial occluder recognition model, performing depth-separable convolution processing on the object occlusion image to obtain image features;
[0081] Step S105, based on the image features and the initial occluder recognition model, perform occluder prediction on the object occluded image to obtain an occluder prediction result and a prediction label, wherein the object occluded image includes an occluder prediction region, the occluder prediction result indicates whether the occluder prediction region contains the predicted occluder or does not contain the predicted occluder, and the prediction label is used to indicate the category of the predicted occluder;
[0082] Step S106, based on the occlusion prediction result, the occlusion prediction area and the prediction label, the initial occlusion recognition model is subjected to loss calculation to obtain model loss data;
[0083] Step S107, adjusting parameters of the initial occluder recognition model based on the model loss data to obtain a target occluder recognition model;
[0084] Step S108, acquiring a target image, and using a target occluder recognition model to perform occluder recognition on the target image.
[0085] In steps S101 to S108 shown in the embodiment of the present application, a target object image including a target recognition object is obtained, and an occlusion annotated image including an occlusion reference and an occlusion label is obtained, and then the target object image is subjected to occlusion processing using the occlusion annotated image to obtain an object occlusion image including the target recognition object and the occlusion reference. Further, according to a preset initial occlusion recognition model, a depth-separable convolution processing is performed on the object occlusion image to obtain image features, and then according to the image features and the initial occlusion recognition model, the object occlusion image is subjected to depth-separable convolution processing to obtain image features. For example, occlusion prediction is performed to obtain an occlusion prediction result indicating whether the occlusion prediction area contains the predicted occlusion or does not contain the predicted occlusion and a prediction label indicating the category of the predicted occlusion. Secondly, based on the above occlusion prediction result, occlusion prediction area and prediction label, the loss of the initial occlusion recognition model is calculated to obtain model loss data. Based on the model loss data, the parameters of the initial occlusion recognition model are adjusted to obtain the target occlusion recognition model. Finally, the target image is obtained, and the target occlusion recognition model is used to perform occlusion recognition on the target image. Therefore, the present application obtains a target object image containing a target recognition object and an occlusion annotated image containing an occlusion reference object and an occlusion label separately, and then uses the occlusion annotated image to perform occlusion processing on the target object image to obtain an object occlusion image, thereby generating an object occlusion image containing a target recognition object and an occlusion reference object. This can increase the number of samples for model training, thereby improving the accuracy of the model. Furthermore, according to a preset initial occlusion recognition model, the object occlusion image is subjected to deep separable convolution processing to obtain image features, which can enhance the model's feature extraction capability for the object occlusion image and reduce the complexity of the model calculation, thereby improving the accuracy of occlusion recognition. Efficiency, secondly, according to the image features and the initial occlusion recognition model, the occlusion prediction is performed on the object occlusion image to obtain the occlusion prediction result and the prediction label, and according to the occlusion prediction result, the occlusion prediction area and the prediction label, the loss calculation of the initial occlusion recognition model is performed to obtain the model loss data, and then according to the model loss data, the parameters of the initial occlusion recognition model are adjusted to obtain the target occlusion recognition model, which improves the diversity of the model's occlusion recognition results, thereby improving the accuracy of occlusion recognition. Finally, the target occlusion recognition model is used to perform occlusion recognition on the acquired target image, which can ensure the accuracy of occlusion recognition of the target image.
[0086] In step S101 of some embodiments, the target identification object may be a farmed animal, such as a cow, a pig, a sheep, or the like.
[0087] The present application can obtain a target object image that includes a target recognition object and does not include any obstructions through an Internet query.
[0088] In step S102 of some embodiments, the occlusion reference object may be an object that has an obstructing effect on the target, such as a fence, a railing, etc. The occlusion object label indicates the category of the occlusion reference object, for example, the category of the fence may be metal, wood, etc.
[0089] The present application can obtain an occlusion image containing an occlusion reference object by the same method as the above-mentioned target object image, and then label the occlusion reference object in the occlusion image through manual or trained material recognition models to generate an occlusion labeled image containing the occlusion reference object and the occlusion label.
[0090] In one application scenario, taking farmed animals such as cattle as an example, through network query, we can obtain a target object image including the cattle head and an occlusion image including wooden railings or iron fences. Then, we can mark the occlusions of the wooden railings or iron fences in the occlusion image, determine the occlusion labels of the wooden railings or iron fences, and obtain the occlusion labeled image.
[0091] In step S103 of some embodiments, occlusion processing is performed on the target object image based on the occluder annotation image by covering the occluder in the occluder annotation image onto the surface of the target recognition object in the target object image, thereby obtaining an object occlusion image containing the target recognition object and a reference occluder, and the target recognition object in the object occlusion image is occluded by the reference occluder.
[0092] For details, see Figure 2 In some embodiments, step S103 may include but is not limited to steps S201 to S203:
[0093] Step S201, performing target segmentation on the target object image to obtain an object image without background;
[0094] Step S202, performing target segmentation on the occluder annotated image to obtain an occluder-free background image;
[0095] Step S203 , performing image fusion on the object without background image and the occlusion object without background image to obtain the object occlusion image.
[0096] In step S201 of some embodiments, the target recognition object in the target object image can be segmented by using methods such as Mask R-CNN to obtain an object-free background image containing only the target recognition object.
[0097] In step S202 of some embodiments, the segmentation process of the reference occluder in the occluder annotated image may be implemented by using methods such as Mask R-CNN to obtain an occluder-free background image containing only the reference occluder.
[0098] In step S203 of some embodiments, the object occlusion image can be obtained by performing Laplacian fusion on the reference occlusion object in the occlusion object without background image and the object without background image.
[0099] For details, see Figure 3 In some embodiments, step S203 may include but is not limited to steps S301 to S303:
[0100] Step S301, performing edge detection on the image of the occluder without background to obtain the edge contour of the occluder;
[0101] Step S302, based on the edge contour of the occluder, the image without background of the occluder is cropped to obtain a cropped image of the occluder;
[0102] Step S303 , performing Laplacian fusion on the occlusion cropped image and the object background-free image to obtain the object occlusion image.
[0103] In step S301 of some embodiments, the Canny algorithm may be used to perform edge detection on the reference occluder in the occluder-free background image, so as to determine the edge contour of the reference occluder, that is, the occluder edge contour.
[0104] In step S302 of some embodiments, in order to ensure that the occluder without background image can be merged with the object without background image, the occluder without background image needs to be cropped to ensure that the size of the cropped occluder cropped image is the same as that of the object without background image.
[0105] In step S303 of some embodiments, the object occlusion image can be obtained by overlapping the occlusion cropped image with the object background-free image, setting the non-reference occlusion area in the occlusion cropped image to zero, and setting the area occluded by the reference occlusion in the object background-free image to zero.
[0106] In steps S301 to S303 shown in the present embodiment, edge detection is performed on the occluder without background image to obtain the occluder edge contour, and then the occluder without background image is cropped according to the occluder edge contour to obtain the occluder cropped image, ensuring that the occluder cropped image and the object without background image are the same size, thereby facilitating image fusion. Furthermore, the occluder cropped image and the object without background image are Laplace fused to obtain the object occluded image, which can improve the authenticity of the object occluded image, thereby ensuring that the object occluded image is effective during model training.
[0107] In steps S201 to S203 shown in the present embodiment, target segmentation is performed on the target object image to obtain an object background-free image, and target segmentation is performed on the occlusion annotated image to obtain an occlusion background-free image. The object background-free image and the occlusion background-free image are then fused, thereby generating a variety of object occlusion images containing occlusions and target recognition objects, thereby increasing the number of object occlusion image samples and enriching the sample database for model training.
[0108] In step S104 of some embodiments, the preset initial occluder model may be a lightweight classification model to facilitate occluder recognition in a mobile terminal such as a mobile phone, for example, a MobileNetV3-Small model.
[0109] The present application can utilize a pre-built initial occluder recognition model to perform deep convolution processing on the above-mentioned object occlusion image to obtain the object occlusion image features after deep convolution processing, and then perform point-by-point convolution processing on the image features to obtain the target features of the object occluder image, that is, the image features.
[0110] For details, see Figure 4 In some embodiments, step S104 may include but is not limited to steps S401 to S402:
[0111] Step S401, performing deep convolution processing on the object occlusion image based on the initial occlusion recognition model to obtain deep convolution image features;
[0112] Step S402: performing point-by-point convolution processing on the deep convolution image features based on the initial occluder recognition model to obtain image features.
[0113] In step S401 of some embodiments, a deep convolution kernel for deep convolution processing pre-set in the initial occluder recognition model may be used to perform convolution processing on the pixels in the above-mentioned object occlusion image to obtain a reduced-dimensional deep convolution image feature.
[0114] It is important to know that when performing deep convolution processing on an object occluded image, any deep convolution kernel can only process the object occluded image in its own channel.
[0115] In step S402 of some embodiments, the target convolution kernel for point-by-point convolution processing pre-set in the initial occluder recognition model can be used to perform cross-channel processing on the above-mentioned deep convolution image features to improve the correlation between pixel points in different input channels, thereby improving the accuracy of occluder recognition.
[0116] For details, see Figure 5In some embodiments, step S402 may include but is not limited to steps S501 to S503:
[0117] Step S501, obtaining the number of output channels of the initial occluder recognition model;
[0118] Step S502, determining the number of kernels of a preset target convolution kernel based on the number of output channels;
[0119] Step S503: Based on the target convolution kernel and the number of kernels, cross-channel convolution is performed on the deep convolution image features to obtain image features.
[0120] In step S501 of some embodiments, the number of output channels of the initial occluder recognition model may be the number of channels of image features output by the initial occluder recognition model.
[0121] In the present application, the initial occluder recognition model is pre-constructed, so the number of output channels of the initial occluder recognition model can be obtained by querying parameter information of the initial occluder recognition model.
[0122] In step S502 of some embodiments, the preset target convolution kernel may be a unit convolution kernel, that is, a convolution kernel with a dimension of 1*1, and the number of channels of the target convolution kernel is the same as the number of input channels of the initial occluder recognition model.
[0123] In order to convert the number of channels of deep convolution image features into the number of output channels of the initial occluder recognition model, the present application needs to set the same number of target convolution kernels as the number of output channels. Therefore, the number of target convolution kernels can be determined according to the number of output channels of the initial occluder recognition model.
[0124] In step S503 of some embodiments, the target convolution kernel is used to perform convolution processing on the deep convolution image features at the same position in different channels, and the number of convolution processing at the same position is the same as the number of kernels, so that the image features can be obtained.
[0125] In steps S501 to S503 shown in this embodiment, the number of kernels of the preset target convolution kernel is determined according to the number of output channels of the initial occluder recognition model, and then the target convolution kernel is used to perform cross-channel convolution on the deep convolution image features at the same position in different channels as many times as the number of kernels, which can achieve the fusion of deep convolution image features in different channels, thereby enhancing the feature extraction capability of the initial occluder recognition model.
[0126] In steps S401 to S402 shown in the present embodiment, deep convolution processing is performed on the object occlusion image to obtain deep convolution image features, and then point-by-point convolution processing is performed on the deep convolution image features to obtain image features, which can reduce the computational complexity of the initial occlusion recognition model and improve the performance of the initial occlusion recognition model. In addition, the deep convolution processing can effectively extract the spatial features of the object occlusion image, and the point-by-point convolution can perform feature fusion on the spatial features extracted by the deep convolution processing in the channel dimension, thereby enhancing the feature extraction capability of the initial occlusion recognition model.
[0127] In step S105 of some embodiments, the possible occluder region in the object occlusion image is determined based on the above-mentioned image features to obtain the occluder prediction region. Furthermore, occluder detection and occluder recognition are performed on the occluder prediction region to obtain an occlusion prediction result indicating whether the occluder prediction region contains the predicted occluder or does not contain the predicted occluder and a prediction label indicating the category of the predicted occluder.
[0128] For details, see Figure 6 In some embodiments, step S105 may include but is not limited to steps S601 to S606:
[0129] Step S601, based on the initial occluder recognition model and image features, perform occluder detection on the object occluded image to obtain an occluder prediction area;
[0130] Step S602, calculating the attention weight of the occluder prediction area to obtain the regional attention weight;
[0131] Step S603, calculating the occluder probability of the occluder prediction region based on the regional attention weight, wherein the occluder probability refers to the probability that the occluder prediction region is an occluder reference object;
[0132] Step S604, determining an occlusion prediction result based on the occlusion probability;
[0133] Step S605, calculating the probability of an occluder label of an occluder reference object in the occluder prediction region based on the regional attention weight;
[0134] Step S606: Based on the occlusion label probability, label prediction is performed on the object occlusion image to obtain a predicted label.
[0135] In step S601 of some embodiments, based on the image features of the object occlusion image obtained by the above-mentioned convolution processing and utilizing the target detection algorithm in the initial occlusion object recognition model, the area in the object occlusion image that may be the occlusion reference object is framed to obtain the occlusion object prediction area.
[0136] In step S602 and step S603 of some embodiments, the initial occluder recognition model is used to perform residual connection, channel attention calculation and other steps on the occluder prediction area to obtain the regional attention weight, and then based on the regional attention weight and the preset hard-swish activation function, the probability that the occluder prediction area is an occlusion reference object, that is, the occluder probability of the occluder prediction area, can be calculated.
[0137] In step S604 of some embodiments, it is possible to determine whether there is an occlusion reference object in the occlusion prediction area based on the occlusion probability of the occlusion prediction area, thereby generating an occlusion prediction result.
[0138] In steps S605 and S606 of some embodiments, based on the regional attention weight, an activation function different from the hard-swish activation function can be used to calculate the occluder label probability of the predicted occluder in the occluder prediction area, and then based on the occluder label probability, the predicted label of the occluder prediction area can be determined.
[0139] In steps S601 to S606 shown in the present embodiment, occlusion detection is performed on the object occlusion image through the initial occlusion recognition model and image features to obtain the occlusion prediction area, and then the regional attention weight of each occlusion prediction area is calculated, and the occlusion probability and occlusion label probability are calculated respectively through the regional attention weight, so as to obtain the occlusion prediction result and the prediction label, thereby determining the output data of the initial occlusion recognition model, improving the diversity of the output data of the initial occlusion recognition model, and improving the accuracy of occlusion recognition in the object occlusion image.
[0140] In step S106 of some embodiments, the occlusion prediction results, occlusion prediction areas and prediction labels are subjected to loss calculation with actual data of the object occlusion image to obtain a variety of loss data, which are then merged to obtain model loss data.
[0141] For details, see Figure 7 In some embodiments, step S106 may include but is not limited to steps S701 to S705:
[0142] Step S701, obtaining occlusion data of the object occlusion image, wherein the occlusion data includes object occlusion data, occlusion object marking area data and occlusion object label data;
[0143] Step S702, calculating occlusion classification loss data of the initial occlusion object recognition model based on the object occlusion data and the occlusion prediction result;
[0144] Step S703, calculating the occluder segmentation loss data of the initial occluder recognition model based on the occluder marking area data and the occluder prediction area;
[0145] Step S704, calculating the occluder label loss data of the initial occluder recognition model based on the occluder label data and the predicted label;
[0146] Step S705 , merging the occlusion classification loss data, the occlusion segmentation loss data and the occlusion label loss data to obtain model loss data.
[0147] In step S701 of some embodiments, the object occlusion data indicates whether the occluder mark region data contains a reference occluder or does not contain a reference occluder, and the occluder label data is used to indicate the category of the reference occluder.
[0148] The present application can determine the object occlusion data and occlusion label data of the object occlusion image by checking the occlusion label of the reference occlusion in the object occlusion image, and then mark the area of the reference occlusion in the object occlusion image to determine the occlusion mark area data of the object occlusion image.
[0149] In step S702 of some embodiments, by comparing the occlusion prediction results with the object occlusion data in the occlusion data, the occlusion classification loss function of the initial occlusion recognition model can be calculated, wherein the occlusion classification loss function is used to represent the gap between the occlusion prediction results output by the initial occlusion recognition model and the marked object occlusion data.
[0150] In step S703 of some embodiments, by comparing the above-mentioned occlusion prediction area with the occlusion marking area data in the occlusion data, the occlusion label loss data of the initial occlusion recognition model can be calculated, wherein the occlusion label loss data is used to represent the gap between the occlusion prediction area output by the initial occlusion recognition model and the marked occlusion marking area data.
[0151] In step S704 of some embodiments, by comparing the above-mentioned predicted label with the occlusion label in the occlusion data, the occlusion segmentation loss data of the initial occlusion recognition model can be calculated, wherein the occlusion segmentation loss data is used to represent the gap between the predicted label output by the initial occlusion recognition model and the marked occlusion label.
[0152] In step S705 of some embodiments, the occlusion classification loss data, the occlusion segmentation loss data and the occlusion label loss data are summed up to obtain the model loss data of the initial occlusion recognition model.
[0153] In steps S701 to S705 shown in the present embodiment, the occlusion data of the object occlusion image is obtained by marking the occlusion objects, and then the occlusion data is compared with the occlusion prediction results, the occlusion prediction area and the prediction label to obtain the occlusion classification loss data, the occlusion segmentation loss data and the occlusion label loss data of the initial occlusion recognition model. The gap between the occlusion prediction results, the occlusion prediction area and the prediction label output by the initial occlusion recognition model and the actual occlusion data can be clearly determined. Furthermore, the occlusion classification loss data, the occlusion segmentation loss data and the occlusion label loss data are merged to obtain the model loss data, which can determine the overall gap between the output of the initial occlusion recognition model and the occlusion data of the object occlusion image, thereby improving the precision of the initial occlusion recognition model and the accuracy of occlusion recognition.
[0154] In step S107 of some embodiments, after obtaining the model loss data of the initial occluder recognition model, the parameters in the initial occluder recognition model are adjusted according to the model loss data to obtain the target occluder recognition model, so as to ensure that the target occluder recognition model can be applied to occluder recognition of the object occluded image, thereby improving the precision of the initial occluder recognition model and the accuracy of occluder recognition.
[0155] In step S108 of some embodiments, the target image refers to an image that requires occlusion recognition.
[0156] After acquiring the target image, the present application can input the target image into the above-mentioned trained target occluder recognition model to detect the occluder and recognize the occluder label.
[0157] The present application obtains a target object image containing a target recognition object and an occlusion object annotated image containing an occlusion reference object and an occlusion object label separately, and then uses the occlusion object annotated image to perform occlusion processing on the target object image to obtain an object occlusion image, thereby generating an object occlusion image containing a target recognition object and an occlusion reference object. This can increase the number of samples for model training, thereby improving the accuracy of the model. Furthermore, according to a preset initial occlusion object recognition model, a deep separable convolution processing is performed on the object occlusion image to obtain image features, which can enhance the model's feature extraction capability for the object occlusion image and reduce the complexity of the model calculation, thereby improving the efficiency of occlusion object recognition. rate, and secondly, according to the image features and the initial occlusion recognition model, the occlusion prediction is performed on the object occlusion image to obtain the occlusion prediction result and the prediction label, and according to the occlusion prediction result, the occlusion prediction area and the prediction label, the loss calculation is performed on the initial occlusion recognition model to obtain the model loss data, and then according to the model loss data, the parameters of the initial occlusion recognition model are adjusted to obtain the target occlusion recognition model, which improves the diversity of the model's occlusion recognition results, thereby improving the accuracy of occlusion recognition. Finally, the target occlusion recognition model is used to perform occlusion recognition on the acquired target image, which can ensure the accuracy of occlusion recognition of the target image.
[0158] See also Figure 8 The embodiment of the present application further provides an obstruction recognition device, which can implement the above-mentioned obstruction recognition method, and the device includes:
[0159] The object image acquisition module 801 is used to acquire a target object image including a target recognition object;
[0160] The occluder image acquisition module 802 is used to acquire an occluder annotated image including a reference occluder and an occluder label, wherein the occluder label is used to indicate the category of the reference occluder;
[0161] The sample image generation module 803 is used to perform occlusion processing on the target object image based on the occlusion object annotation image to obtain an object occlusion image, wherein the object occlusion image includes the target recognition object and the reference occlusion object, and the target recognition object in the object occlusion image is occluded by the reference occlusion object;
[0162] An image feature extraction module 804 is used to perform a depth-separable convolution process on the object occlusion image based on a preset initial occlusion recognition model to obtain image features;
[0163] The occluder prediction module 805 is used to perform occluder prediction on the object occluder image based on the image features and the initial occluder recognition model to obtain an occluder prediction result and a prediction label, wherein the object occluder image includes an occluder prediction region, the occluder prediction result indicates whether the occluder prediction region contains the predicted occluder or does not contain the predicted occluder, and the prediction label is used to indicate the category of the predicted occluder;
[0164] The occlusion loss calculation module 806 is used to calculate the loss of the initial occlusion recognition model based on the occlusion prediction result, the occlusion prediction area and the prediction label to obtain model loss data;
[0165] A model parameter adjustment module 807 is used to adjust the parameters of the initial occluder recognition model based on the model loss data to obtain a target occluder recognition model;
[0166] The occluder recognition module 808 is used to obtain a target image and perform occluder recognition on the target image using a target occluder recognition model.
[0167] The specific implementation of the obstruction recognition device is substantially the same as the specific implementation of the above-mentioned obstruction recognition method, and will not be described in detail herein.
[0168] The embodiment of the present application also provides an electronic device, the electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the above-mentioned obstruction recognition method when executing the computer program. The electronic device can be any intelligent terminal including a tablet computer, a car computer, etc.
[0169] See also Fig. 9 , Fig. 9 The hardware structure of an electronic device of another embodiment is illustrated, and the electronic device includes:
[0170] The processor 901 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;
[0171] The memory 902 may be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 902 may store an operating system and other application programs. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 902, and the processor 901 calls and executes the method for identifying an obstruction in the embodiment of this application.
[0172] Input / output interface 903, used to implement information input and output;
[0173] Communication interface 904, used to realize communication interaction between the device and other devices, which can be realized by wired mode (such as USB, network cable, etc.) or wireless mode (such as mobile network, WIFI, Bluetooth, etc.);
[0174] A bus 905 that transmits information between various components of the device (e.g., the processor 901, the memory 902, the input / output interface 903, and the communication interface 904);
[0175] The processor 901 , the memory 902 , the input / output interface 903 and the communication interface 904 are connected to each other in communication within the device via a bus 905 .
[0176] An embodiment of the present application further provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned obstruction recognition method is implemented.
[0177] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.
[0178] The occlusion recognition method, occlusion recognition device, electronic device and storage medium provided in the embodiments of the present application obtain a target object image including a target recognition object, and obtain an occlusion annotated image including an occlusion reference object and an occlusion label, and then use the occlusion annotated image to perform occlusion processing on the target object image to obtain an object occlusion image including the target recognition object and the occlusion reference object. Further, according to a preset initial occlusion recognition model, a depth-separable convolution processing is performed on the object occlusion image to obtain image features, and then according to the image features and the initial occlusion recognition model, Perform occlusion prediction on the object occlusion image to obtain an occlusion prediction result indicating whether the occlusion prediction area contains the predicted occlusion or does not contain the predicted occlusion and a prediction label indicating the category of the predicted occlusion. Secondly, perform loss calculation on the initial occlusion recognition model based on the above occlusion prediction result, occlusion prediction area and prediction label to obtain model loss data. Then, adjust the parameters of the initial occlusion recognition model based on the model loss data to obtain the target occlusion recognition model. Finally, obtain the target image and use the target occlusion recognition model to perform occlusion recognition on the target image.
[0179] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.
[0180] Those skilled in the art will appreciate that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.
[0181] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0182] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.
[0183] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0184] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.
[0185] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. The mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.
[0186] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0187] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.
[0188] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including multiple instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store programs.
[0189] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but the scope of the rights of the present invention is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present invention should be within the scope of the rights of the present invention.
Claims
1. A method for identifying an obstruction, characterized in that: The method comprises: Acquire a target object image containing a target recognition object; Acquire an occluder annotated image including a reference occluder and an occluder label, wherein the occluder label is used to indicate a category of the reference occluder; Performing occlusion processing on the target object image based on the occlusion object annotation image to obtain an object occlusion image, wherein the object occlusion image includes the target recognition object and the reference occlusion object, and the target recognition object in the object occlusion image is occluded by the reference occlusion object; Based on a preset initial occluder recognition model, a depth-separable convolution process is performed on the object occlusion image to obtain image features; Based on the image features and the initial occluder recognition model, perform occluder prediction on the object occluded image to obtain an occluder prediction result and a prediction label, wherein the object occluded image includes an occluder prediction region, the occluder prediction result indicates whether the occluder prediction region contains the predicted occluder or does not contain the predicted occluder, and the prediction label is used to indicate the category of the predicted occluder; Based on the occlusion prediction result, the occluder prediction area and the prediction label, performing loss calculation on the initial occluder recognition model to obtain model loss data; Adjusting parameters of the initial occluder recognition model based on the model loss data to obtain a target occluder recognition model; A target image is acquired, and occluders are identified on the target image using the target occluder identification model.
2. The method according to claim 1, characterized in that The step of performing occlusion prediction on the object occlusion image based on the image features and the initial occlusion recognition model to obtain an occlusion prediction result and a prediction label includes: Based on the initial occluder recognition model and the image features, performing occluder detection on the object occluded image to obtain the occluder prediction area; Calculating the attention weight of the predicted occluder area to obtain a regional attention weight; Calculating the occluder probability of the occluder prediction region based on the regional attention weight, wherein the occluder probability refers to the probability that the occluder prediction region is the predicted occluder; Determining the occlusion prediction result based on the occlusion probability; Calculating, based on the regional attention weight, an occluder label probability of the predicted occluder in the occluder prediction region; Based on the occlusion label probability, a label prediction is performed on the object occlusion image to obtain the predicted label.
3. The method according to claim 1, characterized in that The performing loss calculation on the initial occluder recognition model based on the occlusion prediction result, the occluder prediction area and the prediction label to obtain model loss data includes: Acquire occlusion data of the object occlusion image, wherein the occlusion data includes object occlusion data, occlusion object marking area data and occlusion object label data; Calculating occlusion classification loss data of the initial occluder recognition model based on the object occlusion data and the occlusion prediction result; Calculating the occluder segmentation loss data of the initial occluder recognition model based on the occluder marking area data and the occluder prediction area; Calculating occluder label loss data of the initial occluder recognition model based on the occluder label data and the predicted label; The occlusion classification loss data, the occlusion segmentation loss data and the occlusion label loss data are merged to obtain the model loss data.
4. The method according to claim 1, characterized in that: The method of performing a depth-separable convolution process on the object occlusion image based on a preset initial occlusion recognition model to obtain image features includes: Performing deep convolution processing on the object occlusion image based on the initial occlusion recognition model to obtain deep convolution image features; The deep convolution image feature is subjected to point-by-point convolution processing based on the initial occluder recognition model to obtain the image feature.
5. The method according to claim 4, characterized in that The step of performing point-by-point convolution processing on the deep convolution image feature based on the initial occluder recognition model to obtain the image feature includes: Obtaining the number of output channels of the initial occluder recognition model; Determine the number of kernels of a preset target convolution kernel based on the number of output channels; Based on the target convolution kernel and the number of kernels, cross-channel convolution is performed on the deep convolution image features to obtain the image features.
6. The method according to any one of claims 1 to 5, characterized in that: The step of performing occlusion processing on the target object image based on the occlusion object annotated image to obtain an object occlusion image includes: Performing target segmentation on the target object image to obtain an object image without background; Performing target segmentation on the occluder annotated image to obtain an occluder-free background image; The object-without-background image and the occlusion-without-background image are fused to obtain the object-occlusion image.
7. The method according to claim 6, characterized in that The step of fusing the object background-free image with the occlusion object background-free image to obtain the object occlusion image includes: Performing edge detection on the image without background of the occluder to obtain the edge contour of the occluder; Based on the edge contour of the occluder, the image without background of the occluder is cropped to obtain an occluder cropped image; The occlusion object cropped image is Laplace-fused with the object background-free image to obtain the object occlusion image.
8. An obstruction recognition device, characterized in that: The device comprises: An object image acquisition module, used to acquire a target object image containing a target recognition object; An occluder image acquisition module, used to acquire an occluder annotated image including a reference occluder and an occluder label, wherein the occluder label is used to indicate the category of the reference occluder; A sample image generation module, configured to perform occlusion processing on the target object image based on the occluder annotated image to obtain an object occlusion image, wherein the object occlusion image includes the target recognition object and the reference occluder, and the target recognition object in the object occlusion image is occluded by the reference occluder; An image feature extraction module, used to perform a depth-separable convolution process on the object occlusion image based on a preset initial occlusion recognition model to obtain image features; an occluder prediction module, configured to perform occluder prediction on the object occluded image based on the image features and the initial occluder recognition model, and obtain an occluder prediction result and a prediction label, wherein the object occluded image includes an occluder prediction region, the occluder prediction result indicates whether the occluder prediction region contains the predicted occluder or does not contain the predicted occluder, and the prediction label is used to indicate the category of the predicted occluder; An occlusion loss calculation module, configured to perform loss calculation on the initial occlusion recognition model based on the occlusion prediction result, the occlusion prediction area and the prediction label to obtain model loss data; A model parameter adjustment module, used to adjust the parameters of the initial occluder recognition model based on the model loss data to obtain a target occluder recognition model; The occluder recognition module is used to acquire a target image and perform occluder recognition on the target image using the target occluder recognition model.
9. An electronic device, characterized in that: The electronic device comprises a memory and a processor, the memory stores a computer program, and the processor implements the obstruction recognition method according to any one of claims 1 to 7 when executing the computer program.
10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the obstruction recognition method according to any one of claims 1 to 7 is implemented.
