Image processing method and device based on deep learning
By building a deep learning-based object detection model and using natural image samples and prompt information for training, the problem of lack of interactive ability in object detection in image processing is solved, and a more efficient image processing effect is achieved.
Patent Information
- Application Number
- CN202411098789.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-12
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2044-08-12
AI Technical Summary
In the prior art, object detection lacks interactive capabilities during image processing, which limits the effect and scope of application of image processing.
By building a deep learning-based object detection model, using natural image samples, prompt information and tag information for training, fusing image encoder, prompt information encoder, feature fusion module and tag matching module, the interactive detection of image and prompt information is realized. .
It significantly improves the interactive ability of the detection model, can effectively realize object detection based on prompt information, and enhances the effect and scope of application of image processing.
Smart Images

Figure CN118747797B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of image processing technology, and in particular to an image processing method and device based on deep learning. Background Art
[0002] Object detection is one of the important means of image processing. For example, in the medical field, traditional object detection uses two-stage object detection models such as RCNN or one-stage object detection models such as YOLO. When detecting objects, two-stage object detection models such as RCNN use the candidate regions generated in the first stage and classify and regress the boundaries of each candidate region in the second stage. One-stage object detection models such as YOLO use a one-stage model structure. In the object detection process, existing technologies lack interactive capabilities when detecting image objects and cannot achieve detection based on prompt objects, which limits the effect and scope of application of image processing. Summary of the invention
[0003] The present invention provides an image processing method and device based on deep learning to solve the technical problems in the prior art that target detection in the image processing process lacks interactive capability and the image processing effect is limited.
[0004] The present invention provides an image processing method based on deep learning, the method comprising:
[0005] Acquire image data to be detected, and merge the image data with input prompt information to obtain input data;
[0006] Inputting the input data into a pre-built target detection model to obtain a target detection result output by the target detection model;
[0007] The target detection model is trained based on a pre-built deep learning network using natural image samples, prompt information corresponding to the natural image samples, and label information.
[0008] In some embodiments, based on a pre-built deep learning network, training is performed using natural image samples, prompt information corresponding to the natural image samples, and label information to obtain the object detection model, specifically including:
[0009] Constructing a data set, the data set including natural image samples, prompt information corresponding to the natural image samples, and label information;
[0010] Dividing the data set into a training set, a validation set, and a test set;
[0011] Inputting the samples in the training set into a pre-built deep learning network for training to obtain an initial detection model;
[0012] The initial detection model is verified and tested using a verification set and a test set, respectively, to obtain the target detection model.
[0013] In some embodiments, constructing a data set specifically includes:
[0014] Collect massive natural image samples to build a library;
[0015] Generating a mask image of equal resolution based on each of the natural image samples to establish a mask library;
[0016] Mark the unmarked objects in the non-masked area on the natural image samples to generate label information;
[0017] A natural image sample, a mask image corresponding to the natural image sample, and label information of its non-mask area are taken as a unit to construct a data set with multiple units.
[0018] In some embodiments, the network architecture of the deep learning network includes:
[0019] An image encoder, wherein the image encoder is used to extract semantic information from an input natural image sample and obtain a feature map of the natural image sample through multiple downsampling;
[0020] A prompt information encoder, which is used to extract semantic information from the input prompt information, and fuse and process the information to obtain a feature map fused with the prompt information;
[0021] A feature fusion module, wherein the feature fusion module is used to perform feature fusion on feature maps of multiple sizes;
[0022] The label matching module determines the label information of the sample using the minimum cost loss and uses contrastive learning to establish the relationship between the image information and the text description.
[0023] In some embodiments, the prompt information includes mask information and text information, and the prompt information encoder includes:
[0024] A mask encoder, wherein the mask encoder is used to extract semantic information from an input mask image and obtain a feature map of the mask image through multiple downsampling;
[0025] A fusion layer, wherein the fusion layer is used to fuse the feature map of the natural image sample with the feature map of the mask image;
[0026] A text encoder, wherein the text encoder is used to extract high-level semantic information of the input text information and establish associations between image features and text features through contrastive learning;
[0027] In some embodiments, the feature fusion module includes same-scale feature fusion and cross-scale feature fusion.
[0028] The present invention also provides an image processing device based on deep learning, the device comprising:
[0029] A data acquisition unit, used to acquire image data to be detected, and merge the image data with input prompt information to obtain input data;
[0030] A result generating unit, used for inputting the input data into a pre-built target detection model to obtain a target detection result output by the target detection model;
[0031] The target detection model is trained based on a pre-built deep learning network using natural image samples, prompt information corresponding to the natural image samples, and label information.
[0032] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described above when executing the program.
[0033] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the method described above when executed by a processor.
[0034] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the method described above is implemented.
[0035] The deep learning-based image processing method and device provided by the present invention obtains the image data to be detected, fuses the image data with the input prompt information to obtain the input data; inputs the input data into a pre-constructed target detection model to obtain the target detection result output by the target detection model; wherein the target detection model is based on a pre-constructed deep learning network, trained using natural image samples, prompt information corresponding to the natural image samples, and label information. The method and device provided by the present invention, by fusing prompt information in model training, can significantly improve the ability of the detection model to interact with humans based on prompt information, has certain practical use value, and solves the technical problems of the lack of interactive ability of target detection and limited image processing effect in the image processing process in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0037] Figure 1 It is one of the flow charts of the image processing method based on deep learning provided by the present invention;
[0038] Figure 2 This is the second flowchart of the image processing method based on deep learning provided by the present invention;
[0039] Figure 3 This is the third flowchart of the image processing method based on deep learning provided by the present invention;
[0040] Figure 4 is a structural block diagram of an image processing device based on deep learning provided by the present invention;
[0041] Figure 5 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION
[0042] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0043] With the rapid development of deep learning networks, the performance of target detection models has been greatly improved. However, with the growing demand for interactive artificial intelligence systems, there are few studies on visual understanding tasks in the field of target detection technology. Recently, the success of large-scale language models has demonstrated the importance of modern artificial intelligence models in human interaction; the introduction of interactive segmentation models has provided a paradigm for the study of visual understanding tasks; some target detection technologies avoid manually designed components, such as anchor definition, label assignment, prediction box post-processing, etc., and complete target detection tasks end-to-end; and some multimodal models associate text and image features. Accordingly, the present invention proposes a deep learning-based and interactive target detection method, which detects the prompted target through prompts such as points, boxes, text, and masks, thereby improving the interactive ability of the model and enhancing its practical use value.
[0044] In a specific embodiment, the image processing method based on deep learning provided by the present invention comprises the following steps:
[0045] S110: Acquire image data to be detected, and merge the image data with input prompt information to obtain input data;
[0046] S120: Inputting the input data into a pre-built target detection model to obtain a target detection result output by the target detection model;
[0047] The target detection model is trained based on a pre-built deep learning network using natural image samples, prompt information corresponding to the natural image samples, and label information.
[0048] In some embodiments, Figure 2 As shown, based on a pre-built deep learning network, training is performed using natural image samples, prompt information corresponding to the natural image samples, and label information to obtain the target detection model, which specifically includes the following steps:
[0049] S210: constructing a data set, the data set including natural image samples, prompt information corresponding to the natural image samples, and label information;
[0050] S220: Divide the data set into a training set, a validation set, and a test set;
[0051] S230: Input the samples in the training set into a pre-built deep learning network for training to obtain an initial detection model; Network model: Based on the DINO framework, by introducing a mask encoding branch and a text encoding branch, the model's interactive ability for prompts such as points, boxes, texts, and masks is improved;
[0052] S240: Utilize the validation set and the test set to respectively validate and test the initial detection model to obtain the target detection model.
[0053] In step S210, constructing a data set specifically includes:
[0054] Collect massive natural image samples to build a library;
[0055] Generating a mask image of equal resolution based on each of the natural image samples to establish a mask library;
[0056] Mark the unmarked objects in the non-masked area on the natural image samples to generate label information;
[0057] A natural image sample, a mask image corresponding to the natural image sample, and label information of its non-mask area are taken as a unit to construct a data set with multiple units.
[0058] Specifically, in the process of constructing the dataset, we first collect a large number of natural images (including natural images included in datasets such as COCO and VOC) and build a library; generate mask images of equal resolution for natural images, where the mask images are single-channel grayscale images, and build a mask library; annotate the unannotated objects in the non-mask areas of natural images (including categories, bounding boxes, and high-quality text descriptions) and generate label information; take the natural image, the mask image corresponding to the natural image, and the label information of its non-mask area as a basic unit, and divide it into training set, validation set, and test set according to the numerical ratio of 8:1:1.
[0059] In some embodiments, the network architecture of the deep learning network includes:
[0060] The image encoder is used to extract semantic information from the input natural image sample and obtain the feature map of the natural image sample after multiple downsampling. Specifically, the image encoder, the MAE pre-trained VisionTransformer (ViT) is used as the image encoder to extract the high-level semantic information of the input natural image and obtain the feature map of the input natural image downsampled by 8 times, 16 times, and 32 times. The feature maps of different levels can be abstractly represented as ,in , is the number of feature maps of different levels;
[0061] A prompt information encoder, which is used to extract semantic information from the input prompt information, and fuse and process the information to obtain a feature map fused with the prompt information;
[0062] Feature fusion module, the feature fusion module is used to fuse the features of multi-scale feature maps, and the feature fusion module includes same-scale feature fusion and cross-scale feature fusion. The feature fusion module aims to improve the model's detection ability for targets of different sizes through multi-scale feature fusion; specifically, in order to reduce the amount of calculation of the model, and considering the lack of semantic information of low-level feature maps and the risk of confusion with semantic information of high-level feature maps, the feature fusion module is decoupled into same-scale feature fusion and cross-scale feature fusion. Same-scale feature fusion can be expressed as:
[0063] ;
[0064] In the above formula, Indicates The feature maps of different levels after the natural image features and the mask image features are fused. Indicates The query corresponding to the feature map of different levels after the natural image features and mask image features are fused, Indicates The key corresponding to the feature map of different levels after the natural image features and the mask image features are fused, Indicates The value corresponding to the feature map of different levels after the natural image features and the mask image features are fused. represents the flattening operation, represents a multi-head self-attention operation, Indicates recovery Dimensions before operation;
[0065] Cross-scale feature fusion adopts the structure of feature pyramid, top-down and bottom-up feature fusion paths, which enriches the position information of high-level feature maps and the semantic information of low-level feature maps, thereby improving the model's detection ability for objects of different sizes.
[0066] The label matching module uses the minimum cost loss to determine the label information of the sample, and establishes classification and regression losses to achieve target detection. In addition, in order to further improve the interactive ability of the model, contrastive learning is used to establish the relationship between image information and text description. The relationship between image information and text description is established by contrastive learning, which can be expressed as:
[0067] ;
[0068] In the above formula, and Represent the loss of visual and textual perspectives, respectively. represents the total loss; Represents the collection of text features within batchsize; Represents the set of image features within the batch size; and for and A subset of Represents a set of text features corresponding to the same category label based on image features; Represents a set of image features corresponding to the same category label based on text features; Represents the cosine similarity between image features and text features.
[0069] In some embodiments, the prompt information includes mask information and text information, and the prompt information encoder includes:
[0070] The mask encoder is used to extract semantic information from the input mask image and obtain the feature map of the mask image after multiple downsampling. Specifically, the mask encoder COCO pre-trained Resnet50 is used as a mask encoder to extract high-level semantic information of the input mask image and obtain feature maps of the input mask image downsampled by 8 times, 16 times, and 32 times. The feature maps of different levels can be abstractly represented as ,in ; The operation formula for fusing natural image features with mask image features can be expressed as:
[0071] ;
[0072] in represents a fully connected layer; represents the feature maps of different levels after fusion, and ;
[0073] A fusion layer, wherein the fusion layer is used to fuse the feature map of the natural image sample with the feature map of the mask image;
[0074] The text encoder is used to extract high-level semantic information of input text information and establish associations between image features and text features through contrastive learning.
[0075] The text encoder uses the CLIP pre-trained text encoder to extract high-level semantic information of the input text, aiming to establish the connection between image features and text features through contrastive learning, thereby improving the model's ability to retrieve target locations based on text prompts; specifically, first randomly sample a batch of natural images , matching the mask image corresponding to it and text description in tag information , Indicates Natural image No. A text description of an object, represents the batch size; then, the natural image , mask image and text description are sent to the image encoder , mask encoder and text encoder , we get image features, mask features and text features, which can be expressed as:
[0076] ;
[0077] In the above formula, and Represents natural images and the mask image The corresponding feature maps of different levels, Indicates Natural image No. The text features of the text description corresponding to the object, and and The number of channels is dimension, The dimension is C dimension;
[0078] The operational formula for fusing natural image features with mask image features can be expressed as:
[0079] ;
[0080] In the above formula, Indicates The feature maps of different levels after the fusion of natural image features and mask image features have C dimensions in number of channels.
[0081] During the model optimization process, a deep learning workstation was established. The operating system used was Ubuntu 20.04, and the hardware configuration was Intel Core i9 processor, 64GB memory and 1 NVIDIA GeForce RTX 3090. The configuration file of the model was based on the DINO structure and retained the contrast denoising module. It avoided the unstable label matching phenomenon of the Hungarian algorithm in the early stage of model training by solving the prior positive and negative query matching problems. The decoder introduced a cross-attention module based on multi-scale deformable convolution, which improved the model's ability to obtain relevant content information based on location information.
[0082] In actual usage scenarios, such as Figure 3 As shown in the figure, the DINO structure provides convenient conditions for the introduction of prompts such as points, boxes, texts, and masks to realize the detection of the prompted targets; specifically, in the test phase, the query position provides an interface for the initialization of physical size information such as points and boxes, and relies on the position information to obtain the image features of interest through the cross-attention module; the mask encoder provides objective conditions for the introduction of mask information, and the supervised learning in the training phase enables the model to autonomously shield the image features of the mask area, so that in the test phase, the query obtains the image features of interest in the non-mask area; the text encoder relies on contrastive learning to establish the connection between text features and image features.
[0083] In the above specific implementation, the image processing method based on deep learning provided by the present invention obtains the image data to be detected, fuses the image data with the input prompt information to obtain the input data; the input data is input into a pre-constructed target detection model to obtain the target detection result output by the target detection model; wherein the target detection model is based on a pre-constructed deep learning network, and is trained using natural image samples, prompt information corresponding to the natural image samples, and label information. The method provided by the present invention, by fusing prompt information in model training, can significantly improve the ability of the detection model to interact with humans based on prompt information, has certain practical use value, and solves the technical problems in the prior art of the lack of interactive capabilities of target detection in the image processing process and the limited image processing effect.
[0084] In addition to the above method, the present invention also provides an image processing device based on deep learning, such as Figure 4 As shown, the device comprises:
[0085] The data acquisition unit 410 is used to obtain image data to be detected, and merge the image data with the input prompt information to obtain input data;
[0086] A result generating unit 420 is used to input the input data into a pre-built target detection model to obtain a target detection result output by the target detection model;
[0087] The target detection model is trained based on a pre-built deep learning network using natural image samples, prompt information corresponding to the natural image samples, and label information.
[0088] In some embodiments, based on a pre-built deep learning network, training is performed using natural image samples, mask image samples corresponding to the natural image samples, and label information to obtain the object detection model, specifically including:
[0089] Constructing a data set, the data set including natural image samples, prompt information corresponding to the natural image samples, and label information;
[0090] Dividing the data set into a training set, a validation set, and a test set;
[0091] Inputting the samples in the training set into a pre-built deep learning network for training to obtain an initial detection model;
[0092] The initial detection model is verified and tested using a verification set and a test set, respectively, to obtain the target detection model.
[0093] In some embodiments, constructing a data set specifically includes:
[0094] Collect massive natural image samples to build a library;
[0095] Generating a mask image of equal resolution based on each of the natural image samples to establish a mask library;
[0096] Mark the unmarked objects in the non-masked area on the natural image samples to generate label information;
[0097] A natural image sample, a mask image corresponding to the natural image sample, and label information of its non-mask area are taken as a unit to construct a data set with multiple units.
[0098] In some embodiments, the network architecture of the deep learning network includes:
[0099] An image encoder, wherein the image encoder is used to extract semantic information from an input natural image sample and obtain a feature map of the natural image sample through multiple downsampling;
[0100] A prompt information encoder, which is used to extract semantic information from the input prompt information, and fuse and process the information to obtain a feature map fused with the prompt information;
[0101] A feature fusion module, wherein the feature fusion module is used to perform feature fusion on feature maps of multiple sizes;
[0102] The label matching module determines the label information of the sample using the minimum cost loss and uses contrastive learning to establish the relationship between the image information and the text description.
[0103] In some embodiments, the prompt information includes mask information and text information, and the prompt information encoder includes:
[0104] A mask encoder, wherein the mask encoder is used to extract semantic information from an input mask image and obtain a feature map of the mask image through multiple downsampling;
[0105] A fusion layer, wherein the fusion layer is used to fuse the feature map of the natural image sample with the feature map of the mask image;
[0106] A text encoder, wherein the text encoder is used to extract high-level semantic information of the input text information and establish associations between image features and text features through contrastive learning;
[0107] In some embodiments, the feature fusion module includes same-scale feature fusion and cross-scale feature fusion.
[0108] In the above specific implementation, the image processing device based on deep learning provided by the present invention obtains the image data to be detected, fuses the image data with the input prompt information to obtain the input data; the input data is input into a pre-constructed target detection model to obtain the target detection result output by the target detection model; wherein the target detection model is based on a pre-constructed deep learning network, and is trained using natural image samples, prompt information corresponding to the natural image samples, and label information. The device provided by the present invention, by fusing prompt information in model training, can significantly improve the ability of the detection model to interact with humans based on prompt information, has certain practical use value, and solves the technical problems in the prior art of the lack of interactive capabilities of target detection in the image processing process and the limited image processing effect.
[0109] Figure 5 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 5 As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530 and a communication bus 540, wherein the processor 510, the communication interface 520 and the memory 530 communicate with each other via the communication bus 540. The processor 510 may call the logic instructions in the memory 530 to execute the above method.
[0110] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.
[0111] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform the above method.
[0112] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.
[0113] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.
[0114] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. An image processing method based on deep learning, characterized in that: The method comprises: Acquire image data to be detected, and merge the image data with input prompt information to obtain input data; Inputting the input data into a pre-built target detection model to obtain a target detection result output by the target detection model; The target detection model is obtained by training based on a pre-built deep learning network using natural image samples, prompt information corresponding to the natural image samples, and label information; The network architecture of the deep learning network includes: An image encoder, wherein the image encoder is used to extract semantic information from an input natural image sample and obtain a feature map of the natural image sample through multiple downsampling; A prompt information encoder, which is used to extract semantic information from the input prompt information, and fuse and process the information to obtain a feature map fused with the prompt information; A feature fusion module, wherein the feature fusion module is used to perform feature fusion on feature maps of multiple sizes; A label matching module, wherein the label matching module determines the label information of the sample using minimum cost loss and uses contrastive learning to establish the relationship between the image information and the text description; The prompt information includes mask information and text information, and the prompt information encoder includes: A mask encoder, wherein the mask encoder is used to extract semantic information from an input mask image and obtain a feature map of the mask image through multiple downsampling; A fusion layer, wherein the fusion layer is used to fuse the feature map of the natural image sample with the feature map of the mask image; A text encoder, wherein the text encoder is used to extract high-level semantic information of the input text information and establish associations between image features and text features through contrastive learning; The feature fusion module includes same-scale feature fusion and cross-scale feature fusion.
2. The image processing method based on deep learning according to claim 1, characterized in that: Based on a pre-built deep learning network, training is performed using natural image samples, prompt information corresponding to the natural image samples, and label information to obtain the object detection model, specifically including: Constructing a data set, the data set including natural image samples, prompt information corresponding to the natural image samples, and label information; Dividing the data set into a training set, a validation set, and a test set; Inputting the samples in the training set into a pre-built deep learning network for training to obtain an initial detection model; The initial detection model is verified and tested using a verification set and a test set, respectively, to obtain the target detection model.
3. The image processing method based on deep learning according to claim 2, characterized in that: The constructing of the data set specifically includes: Collect massive natural image samples to build a library; Generating a mask image of equal resolution based on each of the natural image samples to establish a mask library; Mark the unmarked objects in the non-masked area on the natural image samples to generate label information; A natural image sample, a mask image corresponding to the natural image sample, and label information of its non-mask area are taken as a unit to construct a data set with multiple units.
4. An image processing device based on deep learning, based on the method according to any one of claims 1 to 3, characterized in that: The device comprises: A data acquisition unit, used to acquire image data to be detected, and merge the image data with input prompt information to obtain input data; A result generating unit, used for inputting the input data into a pre-built target detection model to obtain a target detection result output by the target detection model; The target detection model is trained based on a pre-built deep learning network using natural image samples, prompt information corresponding to the natural image samples, and label information.
5. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the method according to any one of claims 1 to 3 is implemented.
6. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 3 is implemented.
7. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 3 is implemented.
Citation Information
Patent Citations
Target detection method based on data enhancement
CN109063748A
Smoke detection method and system based on lightweight neural network, medium and equipment
CN116434002A
Supervised multi-modal Hash retrieval method and system based on label enhancement
CN117493674A
Target detection method and device based on visual prompt, equipment and storage medium
CN117710644A
Mouse target detection method and system based on adaptive mask, and storage medium
CN118247716A
