Target detection method, system and device for adaptively selecting attention mechanism

By introducing the attention mechanism adaptive selection black box in the YOLOv7 model and independently selecting the appropriate attention mechanism, the detection accuracy problem in the existing technology is solved in small and medium-sized goals and complex backgrounds, and the adaptability and detection accuracy of the model are improved.

CN119091125APending Publication Date: 2024-12-06GUANGDONG UNIV OF TECH
View PDF 10 Cites 0 Cited by

Patent Information

Application Number
CN202411232879.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-04
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

Existing object detection methods have problems with performance degradation when dealing with small targets and complex backgrounds, and lack adaptability to select the best attention mechanism, which affects detection accuracy.

Method used

Based on the YOLOv7 model, the adaptive selection black box of attention mechanism is introduced. By designing a variety of attention mechanism types and learnable parameters, the appropriate attention mechanism scheme is independently selected, and the object detection model of the adaptive selection attention mechanism is constructed.

Benefits of technology

The adaptability and detection accuracy of the model are improved, the human parameter adjustment intervention is reduced, and it can better adapt to different object detection tasks, and the model's fault tolerance and expression ability are enhanced.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119091125A_ABST
    Figure CN119091125A_ABST
Patent Text Reader

Abstract

The invention discloses a target detection method, system and device for adaptively selecting an attention mechanism. The method comprises the following steps: acquiring a to-be-detected image; based on a basic framework of a YOLOv7 model, an attention mechanism is introduced into a backbone network to adaptively select a black box, and a target detection model is constructed; training the detection model to obtain a trained detection model; and inputting the to-be-detected image into the trained detection model to obtain a detection result. The system comprises an image acquisition unit, a model construction unit, a model training unit and a detection unit. The device comprises a memory and a processor used for executing the target detection method of the adaptive selection attention mechanism. According to the invention, attention mechanism selection is carried out for different tasks, and the target detection precision is improved. The method can be widely applied to the field of image processing.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of image processing, and in particular to a target detection method, system and device with an adaptive selection attention mechanism. Background Art

[0002] Object detection is one of the important tasks in the field of computer vision and is widely used in image recognition, autonomous driving, smart security and other fields. Traditional one-stage object detection algorithms have certain challenges in dealing with small targets, target positioning accuracy and robustness to complex backgrounds. With the in-depth research of the YOLO series of one-stage object detection algorithms, YOLOv7 stands out with its simplicity and speed, and is widely used in real-time applications and resource-constrained scenarios. However, the algorithm may face some performance degradation when dealing with small targets and complex backgrounds. Small targets may be ignored or misjudged as background due to their small size, and interfering objects in complex backgrounds may cause positioning deviations of the target.

[0003] With the continuous development of attention mechanism technology, the YOLOv7 algorithm equipped with attention mechanism has broken through many limitations in small target detection and complex backgrounds. However, for different tasks, the use of attention and the type of attention (such as direct attention mechanism, spatial attention mechanism, etc.) will directly affect the final detection effect. However, the existing integrated model is relatively fixed and lacks certain adaptive capabilities. Summary of the invention

[0004] In view of this, in order to solve the technical problem that the existing target detection method cannot select the attention mechanism for different tasks, thereby resulting in low detection accuracy, the present invention proposes a target detection method for adaptively selecting the attention mechanism, the method comprising the following steps:

[0005] Acquire the image to be tested;

[0006] Based on the basic framework of the YOLOv7 model, the attention mechanism is introduced into the backbone network to adaptively select black boxes and build a target detection model;

[0007] Training the detection model to obtain a trained detection model;

[0008] The image to be tested is input into the trained detection model to obtain a detection result.

[0009] In some embodiments, the target detection model includes a backbone network, a neck network, and a head network, wherein:

[0010] The backbone network includes an attention mechanism adaptive multi-layer convolution structure, a convolution pooling structure, an attention mechanism adaptive selection black box and a composite convolution layer;

[0011] The neck network ELAN-A structure, SPPCSPC-A structure and composite convolutional layer.

[0012] In some embodiments, in the attention mechanism adaptive selection black box:

[0013] Set parameter threshold interval;

[0014] Combining the current image parameters to be tested and the parameter threshold interval, the input feature map is moved to the corresponding attention mechanism for processing;

[0015] The attention mechanism includes an independent channel attention mechanism, an independent spatial attention mechanism, a serial channel and spatial attention mechanism, and a parallel channel and spatial attention mechanism.

[0016] In this embodiment, considering that the feature extraction process needs to pay attention to the channel information and spatial information of the feature map, a total of four types of attention mechanisms are designed, and the subsequent processing process is determined by the parameter threshold interval.

[0017] In some embodiments, a compression and excitation module is introduced into the channel attention mechanism, and a channel weight is learned on each channel by the compression and excitation module to adaptively adjust the importance of channel features.

[0018] Through this preferred embodiment, the performance and expression ability of the model can be improved.

[0019] In some embodiments, a spatial transformation module is introduced into the spatial attention mechanism, and the target detection model is enabled to adaptively perform spatial transformation on the input through the spatial transformation module.

[0020] In this embodiment, the spatial transformation module can learn adaptive transformation parameters so that the network can better model and locate the target at different input positions and scales, thereby improving the performance of the task.

[0021] In some embodiments, in the SPPCSPC-A structure:

[0022] The feature map undergoes channel transformation and feature extraction through a multi-layer convolution module, and then enters a multi-branch structure consisting of a maximum pooling operation;

[0023] According to the first preset parameter, determining whether the output of the multi-branch structure uses channel attention, and obtaining a determination result;

[0024] The data processing path is executed based on the decision results and the output is fused.

[0025] In this embodiment, considering the task of having a preference for the target scale, a learnable parameter is set to select whether to use the attention mechanism, so that the model has the ability of autonomous learning and adaptation, and can better adapt to different target detection tasks.

[0026] In some embodiments, in the ELAN-A structure, a first branch and a second branch are provided, wherein:

[0027] The first branch adopts a CBS convolution structure;

[0028] The second branch is provided with at least two CBS convolution structures;

[0029] When the second preset parameter is in the first interval, fusing the outputs of the first branch and the second branch;

[0030] When the second preset parameter is in the second interval, the spatial attention mechanism is enabled to pay attention to the spatial information of the outputs of the first branch and the second branch, and then the outputs are fused.

[0031] In this embodiment, the introduced parameters can enable the model to better detect different target detection tasks, improve the overall fault tolerance of the model, increase the learnability of the model itself, and reduce human intervention.

[0032] The present invention also proposes a target detection system with an adaptive selection attention mechanism, the system comprising:

[0033] An image acquisition unit, used for acquiring an image to be tested;

[0034] The model building unit, based on the basic framework of the YOLOv7 model, introduces the attention mechanism in the backbone network to adaptively select black boxes and build a target detection model;

[0035] A model training unit, used to train the detection model to obtain a trained detection model;

[0036] The detection unit is used to input the image to be detected into the trained detection model to obtain a detection result.

[0037] The present invention also proposes a target detection device for adaptively selecting an attention mechanism, comprising:

[0038] at least one processor;

[0039] at least one memory for storing at least one program;

[0040] When the at least one program is executed by the at least one processor, the at least one processor implements the target detection method of adaptively selecting the attention mechanism as described above.

[0041] Based on the above scheme, the present invention provides a target detection method, system and device for adaptively selecting an attention mechanism. By introducing an attention mechanism adaptive selection black box, the model can be made more flexible, and can autonomously select the best attention mechanism scheme during training according to different tasks, which improves the adaptive ability of the model and reduces human parameter adjustment intervention. By improving the ELAN structure and SPPCSPC structure in the YOLOv7 model, the module has a certain adaptive ability and can autonomously determine whether to select an attention mechanism to obtain the best detection results. Compared with the existing model, the improved model based on the adaptive selection of the attention mechanism of YOLOv7 can better control the timing of selecting the attention mechanism. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is a flowchart of the steps of a target detection method of an adaptive selection attention mechanism of the present invention;

[0043] Figure 2 It is a structural block diagram of the target detection network of the present invention;

[0044] Figure 3 It is a schematic diagram of the principle of adaptively selecting a black box by the attention mechanism in a specific embodiment of the present invention;

[0045] Figure 4 Schematic diagram of the principle of the channel attention mechanism of a specific embodiment of the present invention;

[0046] Figure 5 It is a schematic diagram of the principle of the spatial attention mechanism of a specific embodiment of the present invention;

[0047] Figure 6 It is a structural block diagram of the SPPCSPC-A module of a specific embodiment of the present invention;

[0048] Figure 7 is a block diagram of the structure of ELAN-A according to a specific embodiment of the present invention;

[0049] Figure 8 It is a structural block diagram of a target detection system of the present invention that adaptively selects attention mechanisms. DETAILED DESCRIPTION

[0050] Currently, the target detection model based on deep learning has a module for deep feature extraction. It generally has an operation of fusing the convolutional feature maps of individual layers to obtain feature information at different depths. However, this structure lacks the expression of the attention to each layer of information. In the latest technical research, feature extraction modules with attention mechanisms have emerged. However, in some detection tasks, not adding the attention mechanism will have better results. Therefore, in order to enable the model to adapt to various situations, it is necessary to give the model adaptive capabilities.

[0051] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0052] It should be noted that, for the convenience of description, only the parts related to the relevant invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other.

[0053] It should be understood that the "system", "device", "unit" and / or "module" used in this application is a method for distinguishing different components, elements, parts, portions or assemblies at different levels. However, if other words can achieve the same purpose, the word can be replaced by other expressions.

[0054] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "a kind" and / or "the" do not refer to the singular, but also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of clearly identified steps and elements, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements. The elements defined by the sentence "includes a..." do not exclude the existence of other identical elements in the process, method, commodity or device that includes the elements.

[0055] In the description of the embodiments of the present application, "plurality" means two or more than two. The following terms "first" and "second" are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features.

[0056] In addition, flow charts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, the various steps may be processed in reverse order or simultaneously. At the same time, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.

[0057] Reference Figure 1 , which is a flow chart of an optional example of a target detection method of adaptively selecting an attention mechanism proposed in the present invention. The method can be applied to a computer device. The target detection method proposed in this embodiment may include but is not limited to the following steps:

[0058] Step S1, obtaining an image to be tested;

[0059] Step S2: Based on the basic framework of the YOLOv7 model, an attention mechanism is introduced into the backbone network to adaptively select black boxes and build a target detection model;

[0060] Step S3: training the detection model to obtain a trained detection model;

[0061] Step S4: input the image to be tested into the trained detection model to obtain a detection result.

[0062] In some feasible embodiments, the target detection model in step S2 is an improved YOLOv7 model, which specifically includes:

[0063] The improved YOLOv7 overall model structure includes three parts: Backbone, Neck and Head, each of which has different functions and roles. Backbone is the basic part of the YOLOv7 model, which is used to extract high-level feature representations from the input image. The backbone network is mainly composed of ELAN-A (attention mechanism adaptive multi-layer convolution) structure, MP-1 (convolution pooling) structure, attention mechanism adaptive selection black box and multiple (CBS) composite convolution layers, which can gradually extract low-level to high-level feature information of the image. In addition, different CBS modules correspond to convolution processing with different parameter settings. Neck is the connection part in the YOLOv7 model, connecting the backbone network and the detection head. Its main function is to perform feature fusion and information transmission between feature maps at different levels to improve the perception ability and accuracy of the target detection algorithm. The connection structure uses the ELAN-A structure, SPPCSPC-A structure and CBS structure to complete bottom-up and top-down sampling fusion. By introducing the connection part, features from different levels can be fused to provide richer contextual information and multi-scale features, thereby improving the target detection algorithm's ability to detect targets of different scales. In addition, ELAN-W performs deep feature extraction at the neck; UPSample is an upsampling operation; Cat represents the fusion of multiple feature maps. Head is the detection head in the YOLOv7 model, which is responsible for predicting the bounding box and category information of the target. Among them, REP is a branched convolution layer, which is used to extract category and position information; CBM is a traditional convolution integration module. This technical solution is based on the original YOLOv7 model, adds an attention mechanism to adaptively select black boxes, improves the ELAN (layered convolution) module, and improves the SPPCSPC (pyramid pooling convolution) structure. The overall framework of the improved YOLOv7 is as follows: Figure 2 .

[0064] In some feasible embodiments, the attention mechanism adaptively selects a black box, which specifically includes:

[0065] Considering that the feature extraction process needs to pay attention to the channel information and spatial information of the feature map, a total of four types of attention mechanisms are designed, namely independent channel attention mechanism, independent spatial attention mechanism, serial channel and spatial attention mechanism, and parallel channel and spatial attention mechanism. Previously, a parameter α was set for model training, and four parameter thresholds were set for it, representing four attention mechanism modes. When the parameter α value learned by the model is within the threshold set for a certain mode, the model will move the input feature map to the attention mechanism of that mode for processing and output. The independent channel attention mechanism focuses on the channel information of the feature map; the independent spatial attention mechanism focuses on the spatial information of the feature map; the serial mode of the channel and spatial attention mechanism mainly focuses on obtaining better performance; the parallel mode of the channel and spatial attention mechanism mainly focuses on the comprehensive feature information;

[0066] The working principle of the attention mechanism adaptive selection black box is as follows Figure 3 shown.

[0067] This embodiment designs an attention mechanism adaptive selection black box, which is characterized by selecting multiple attention mechanism schemes in parallel, with a learnable parameter α for control, adjusting the α parameter through model training, and autonomously selecting the best attention mechanism route under the original set threshold to obtain the best detection effect. This design effectively improves the self-learning ability of the model, enables the model to adaptively adjust in different detection tasks, and select the most appropriate attention mechanism scheme, which improves the flexibility of the model and reduces manual intervention in parameter adjustment.

[0068] In some feasible embodiments, the channel attention mechanism specifically includes:

[0069] The channel attention mechanism introduces a Squeeze-and-Excitation module to learn a channel weight on each channel to adaptively adjust the importance of channel features. The feature map is compressed through a global average pooling operation to obtain the importance representation of each channel, and then these representations are nonlinearly mapped through a small fully connected neural network to obtain the channel attention weight. Finally, the channel attention weight is multiplied by the original channel feature map so that more important features receive more attention, thereby improving the performance and expressiveness of the model.

[0070] The channel attention mechanism works as follows Figure 4 shown.

[0071] In some feasible embodiments, the spatial attention mechanism specifically includes:

[0072] The STN spatial attention mechanism introduces a learnable spatial transformation module to enable the network to adaptively transform the input spatially, thereby improving the geometric deformation invariance and robustness of the model. STN can perform deformation operations such as translation, rotation, and scaling on the input by learning a spatial transformation parameter matrix. This spatial transformation module can improve the performance of the task by learning adaptive transformation parameters, so that the network can better model and locate the target at different positions and scales of the input.

[0073] The working principle of spatial attention mechanism is as follows Figure 5 shown.

[0074] In some feasible embodiments, the SPPCSPC-A structure specifically includes:

[0075] like Figure 6 As shown in the figure, the SPPCSPC-A (pyramid pooling convolution with adaptive attention mechanism) structure mainly introduces the adaptive selection mechanism of attention mechanism on the original SPPCSPC structure. The feature map input first passes through three multi-layer convolution modules, whose main function is to transform the channels of the feature map and extract basic feature information, and then enters the multi-branch structure composed of maximum pooling operations. Each branch extracts the feature information of small, medium and large scale targets by setting pooling kernels of different sizes of 5, 9, and 13. Considering the task of having a tendency to select the target scale, a learnable parameter β is set to choose whether to use the attention mechanism. Finally, all feature maps are fused and output under the channel transformation of the multi-layer convolution structure.

[0076] The value of the β parameter determines whether to use the channel attention mechanism SEnet (Squeeze-and-Excitation Network). The choice of introducing the channel attention mechanism in the SPPCSPC structure is mainly to consider that the attention of large, medium and small target scales will be emphasized due to different tasks. The introduced learnable parameter β enables the model to have autonomous learning and adaptability, and can better adapt to different target detection tasks.

[0077] In some feasible embodiments, the ELAN-A structure specifically includes:

[0078] like Figure 7As shown in the figure, the ELAN-A (multi-layer convolution with adaptive attention mechanism) structure introduces the adaptive spatial attention mechanism into the original ELAN module. The ELAN-A structure is composed of multiple CBS modules with different functions. The residual network idea is used. One branch retains the input feature information while coordinating the number of channels and transmits it to the output for fusion with the main branch output. The main branch is composed of multiple different types of CBS multi-layer convolution structures connected in series, which respectively realize the functions of feature map size transformation, channel number conversion, and feature extraction. In the process of deep convolution, the feature maps of several specific layers will be taken out and directly output to the adaptive spatial attention mechanism layer for fusion with the residual branch. Considering that the feature maps output by different layers of convolution have rich spatial feature information, it is necessary to increase the attention to spatial information during the target detection process. Therefore, the spatial attention mechanism STN (Spatial Transformer Network) is introduced. In order to improve the model's ability to autonomously select the spatial attention mechanism, this technology introduces a learnable parameter γ for the model to obtain the best detection effect for different target detection tasks.

[0079] The γ parameter is learned through repeated iterative training of the model. In order to allow the model to independently decide whether to use the spatial attention mechanism STN (Spatial Transformer Network), this technology designs a γ determination formula, that is, when the γ training value is between 0 and 0.5, the ELAN structure in the model will directly access the concat fusion operation to fuse the feature map, and when the γ value is between 0.5-1, the model will enable the spatial attention mechanism STN to pay attention to the spatial information of each layer of feature map before output. The introduced γ parameter can enable the model to better detect different target detection tasks, improve the overall fault tolerance of the model, increase the learnability of the model itself, and reduce human intervention.

[0080] Based on the above scheme, this embodiment takes into account the functional characteristics of the ELAN structure and the SPPCSPC structure in the YOLOv7 model, and makes certain adaptive improvements thereto. An adaptive spatial attention mechanism is introduced in the ELAN structure, and an adaptive channel attention mechanism is introduced in the SPPCSPC structure, so that the model can decide whether to use the attention mechanism according to the requirements of different tasks, thereby serving the final detection effect.

[0081] In some feasible embodiments, the step S3 specifically includes:

[0082] Collect training samples from different detection tasks, build a training data set, divide the training set and the data set in a certain ratio, and put the training samples into the model for training. This training step is no different from conventional model training.

[0083] like Figure 8As shown, a target detection system for adaptively selecting an attention mechanism includes:

[0084] An image acquisition unit, used for acquiring an image to be tested;

[0085] The model building unit, based on the basic framework of the YOLOv7 model, introduces the attention mechanism in the backbone network to adaptively select black boxes and build a target detection model;

[0086] A model training unit, used to train the detection model to obtain a trained detection model;

[0087] The detection unit is used to input the image to be detected into the trained detection model to obtain a detection result.

[0088] The contents of the above method embodiments are all applicable to the present system embodiments. The functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0089] An object detection device with adaptive selection of attention mechanism:

[0090] at least one processor;

[0091] at least one memory for storing at least one program;

[0092] When the at least one program is executed by the at least one processor, the at least one processor implements the target detection method of adaptively selecting the attention mechanism as described above.

[0093] The contents of the above method embodiments are all applicable to the present device embodiments. The functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0094] A storage medium stores processor-executable instructions, which, when executed by a processor, are used to implement a target detection method with an adaptive selection attention mechanism as described above.

[0095] The contents of the above method embodiments are all applicable to the present storage medium embodiments. The functions specifically implemented by the present storage medium embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0096] The above is a specific description of the preferred implementation of the present invention, but the invention is not limited to the embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.

Claims

1. A target detection method with adaptive selection of attention mechanism, characterized in that: The following steps are involved: Acquire the image to be tested; Based on the basic framework of the YOLOv7 model, the attention mechanism is introduced into the backbone network to adaptively select black boxes and build a target detection model; Training the detection model to obtain a trained detection model; The image to be tested is input into the trained detection model to obtain a detection result.

2. According to claim 1, a target detection method for adaptively selecting an attention mechanism is characterized in that: The target detection model includes a backbone network, a neck network and a head network, wherein: The backbone network includes an attention mechanism adaptive multi-layer convolution structure, a convolution pooling structure, an attention mechanism adaptive selection black box and a composite convolution layer; The neck network ELAN-A structure, SPPCSPC-A structure and composite convolutional layer.

3. According to claim 2, a target detection method for adaptively selecting an attention mechanism is characterized in that: In the attention mechanism adaptive selection black box: Set parameter threshold interval; Combining the current image parameters to be tested and the parameter threshold interval, the input feature map is moved to the corresponding attention mechanism for processing; The attention mechanism includes an independent channel attention mechanism, an independent spatial attention mechanism, a serial channel and spatial attention mechanism, and a parallel channel and spatial attention mechanism.

4. According to claim 3, a target detection method for adaptively selecting an attention mechanism is characterized in that: A compression and excitation module is introduced into the channel attention mechanism, through which a channel weight is learned on each channel to adaptively adjust the importance of channel features.

5. According to claim 4, a target detection method for adaptively selecting an attention mechanism is characterized in that: A spatial transformation module is introduced into the spatial attention mechanism, and the target detection model is enabled to adaptively perform spatial transformation on the input through the spatial transformation module.

6. According to claim 5, a method for target detection using an adaptive selection attention mechanism is characterized in that: In the SPPCSPC-A structure: The feature map undergoes channel transformation and feature extraction through a multi-layer convolution module, and then enters a multi-branch structure consisting of a maximum pooling operation; According to the first preset parameter, determining whether the output of the multi-branch structure uses channel attention, and obtaining a determination result; The data processing path is executed based on the decision results and the output is fused.

7. According to claim 6, a target detection method for adaptively selecting an attention mechanism is characterized in that: In the ELAN-A structure, a first branch and a second branch are provided, wherein: The first branch adopts a CBS convolution structure; The second branch is provided with at least two CBS convolution structures; When the second preset parameter is in the first interval, fusing the outputs of the first branch and the second branch; When the second preset parameter is in the second interval, the spatial attention mechanism is enabled to pay attention to the spatial information of the outputs of the first branch and the second branch, and then the outputs are fused.

8. An object detection system with adaptive selection of attention mechanism, characterized in that: include: An image acquisition unit, used for acquiring an image to be tested; The model building unit, based on the basic framework of the YOLOv7 model, introduces the attention mechanism in the backbone network to adaptively select black boxes and build a target detection model; A model training unit, used to train the detection model to obtain a trained detection model; The detection unit is used to input the image to be detected into the trained detection model to obtain a detection result.

9. An object detection device with an adaptive selection attention mechanism, characterized in that: include: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements a target detection method with an adaptive selection attention mechanism as described in any one of claims 1-7.

Citation Information

Patent Citations

  • Traffic flow statistical method based on improved YOLO V7 and Deep-Sort

    CN116434159A

  • Improved industrial PCB defect detection method and device based on YOLOv7-Tiny model

    CN116934696A

  • Single-frame infrared weak and small target detection method based on improved YOLOv7

    CN117611911A

  • Aviation fastener fuse winding direction automatic identification method

    CN117710740A

  • Grain storage pest detection method based on YOLOv7 model

    CN118154846A