Methods, equipment and products for identifying and detecting floating objects on the water surface

Abstract

The present invention discloses a method, device and product for identifying and detecting floating objects on water surface, comprising: establishing a neural network model based on an RT-DETR neural network model; training the neural network model using a labeled water surface floating object image dataset; and identifying and detecting images that may contain water surface floating objects using the trained neural network model. The present invention discloses a method, device and product for identifying and detecting floating objects on water surface, which innovatively uses a multi-branch backbone network to ensure the extraction of shallow spatial features of the input image and to extract the global context fusion semantic features of the input image, and can adaptively and efficiently fuse the two types of features, thereby effectively improving the accuracy of identifying and detecting floating objects on water surface and improving the applicability to various application scenarios.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence, and in particular relates to a method, device and product for identifying and detecting floating objects on water surfaces. Background Art

[0002] Under existing technologies, the identification and detection of floating objects on the water surface still mainly relies on manual inspection methods, including on-site inspections, or the use of drones, ships and other vehicles to capture water surface images or videos, and then operators perform subsequent image analysis and floating object identification. Therefore, in the monitoring of floating objects on the water surface in large areas of water, manual inspections often cannot fully cover the entire water area, which can easily lead to missed detections and false detections, and the response to environmental changes is relatively slow; on the other hand, the existing artificial intelligence-based floating object detection methods are also very unstable and have poor adaptability due to reasons such as water surface fluctuations, floating object movement, weather changes, floating object types and changes in size. Summary of the Invention

[0003] In view of this, the present invention aims to overcome the defects in the prior art and proposes a method, equipment and product for identifying and detecting floating objects on the water surface.

[0004] To achieve the above object, the technical solution of the present invention is achieved as follows:

[0005] In a first aspect, the present invention discloses a method for identifying and detecting floating objects on a water surface, comprising:

[0006] Establishing a neural network model and using a labeled image dataset of floating objects on the water surface to train the neural network model to obtain a neural network model that can identify floating objects on the water surface;

[0007] Use the trained neural network model to identify and detect images that may contain floating objects on the water surface;

[0008] Establishing the neural network model includes: establishing the neural network model based on the RT-DETR neural network model, replacing the backbone network of the RT-DETR neural network model with a multi-branch backbone network as the neural network model;

[0009] The multi-branch backbone network includes: using a shallow spatial feature module to extract shallow spatial features of the input image; using a semantic feature module to extract context-fused semantic features of the input image; and using a fusion module to fuse the shallow spatial features and the context-fused semantic features and output them.

[0010] In one embodiment of the present invention, a fusion module is used to fuse shallow spatial features and contextual fusion semantic features and output the fusion results, including:

[0011] The shallow spatial features and context-fused semantic features are convolved separately and then concatenated to obtain the first concatenated features;

[0012] Perform Sigmoid normalization on the first concatenated feature to generate dynamic alignment weights;

[0013] Perform split weight division on the dynamic alignment weight to obtain weight parameters corresponding to shallow spatial features and context-fused semantic features respectively;

[0014] The fused output of shallow spatial features and context-fused semantic features is the weighted sum of the two according to the corresponding weight parameters.

[0015] In one embodiment of the present invention, after replacing the backbone network of the RT-DETR neural network model with a multi-branch backbone network as the neural network model, it also includes: replacing the attention image feature fusion AIFI module of the neural network model with a dynamic range histogram self-attention DHSA module or a cascade group attention CGA module.

[0016] In one embodiment of the present invention, the shallow spatial feature module is composed of a cascade stack of a Conv module and a C2f module.

[0017] In one embodiment of the present invention, the semantic feature module is composed of a cascade stack of a DWConv module and an HGBlock module.

[0018] In one embodiment of the present invention, the fusion module is a multi-scale adaptive spatial attention gate MASAG module.

[0019] In one embodiment of the present invention, the HGStem module is used to obtain basic features of the input image, and the shallow spatial feature module and the semantic feature module both extract features of the input image based on the basic features.

[0020] In a second aspect, the present invention discloses a device for identifying and detecting floating objects on a water surface, the device comprising:

[0021] A model building module is used to build a neural network model and train the neural network model using a labeled water surface floating object image dataset to obtain a neural network model capable of identifying water surface floating objects; building the neural network model includes: building a neural network model based on the RT-DETR neural network model, replacing the backbone network of the RT-DETR neural network model with a multi-branch backbone network; the multi-branch backbone network includes: using a shallow spatial feature module to extract shallow spatial features of the input image; using a semantic feature module to extract context-fused semantic features of the input image; using a fusion module to fuse the shallow spatial features and the context-fused semantic features and output them;

[0022] The recognition and detection module is used to use the trained neural network model to identify and detect images that may contain floating objects on the water surface.

[0023] In a third aspect, the present invention discloses an electronic device comprising: one or more processors; a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors execute the above method.

[0024] In a fourth aspect, the present invention discloses a computer program product, comprising a computer program, which implements the above method when executed by a processor.

[0025] Compared with the prior art, the present invention has the following advantages:

[0026] The present invention discloses a method, device and product for identifying and detecting floating objects on water surface, comprising: establishing a neural network model based on an RT-DETR neural network model; training the neural network model using a labeled water surface floating object image data set; and identifying and detecting images that may contain water surface floating objects using the trained neural network model. The present invention discloses a method, device and product for identifying and detecting floating objects on water surface, which innovatively uses a multi-branch backbone network to ensure the extraction of shallow spatial features of an input image and to extract global context fusion semantic features of the input image, and can adaptively and efficiently fuse the two types of features, thereby effectively improving the accuracy of identifying and detecting floating objects on water surface and improving the applicability to various application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0028] In the attached figure:

[0029] Figure 1 This is a schematic diagram of an application scenario of a method for identifying and detecting floating objects on a water surface according to an embodiment of the present invention;

[0030] Figure 2 This is a schematic diagram of a method for identifying and detecting floating objects on a water surface according to an embodiment of the present invention;

[0031] Figure 3 This is a schematic diagram of a neural network model for a method for identifying and detecting floating objects on a water surface according to an embodiment of the present invention;

[0032] Figure 4 This is a schematic diagram of a shallow space feature module of a method for identifying and detecting floating objects on a water surface according to an embodiment of the present invention;

[0033] Figure 5 This is a schematic diagram of a semantic feature module of a method for identifying and detecting floating objects on a water surface according to an embodiment of the present invention;

[0034] Figure 6 This is a schematic diagram of the SCFB module of a method for identifying and detecting floating objects on a water surface according to an embodiment of the present invention;

[0035] Figure 7 This is a schematic diagram of a fusion module of a method for identifying and detecting floating objects on a water surface according to an embodiment of the present invention;

[0036] Figure 8 This is a schematic diagram of an electronic device for a method for identifying and detecting floating objects on a water surface according to an embodiment of the present invention;

[0037] Figure 9 The figure is a schematic diagram of a device for identifying and detecting floating objects on the water surface according to an embodiment of the present invention. DETAILED DESCRIPTION

[0038] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0039] In the description of the present invention, it should be further clarified that the terms "first," "second," etc. are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature specified as "first," "second," etc. may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "plurality" means two or more.

[0040] The present invention discloses a method for identifying and detecting floating objects on the water surface, a device and an application scenario of the product. Figure 1 As shown, under existing technical conditions, the identification and detection of floating objects on the water surface, on the one hand, still mainly relies on manual inspection methods, which are prone to missed detection and false detection; on the other hand, the existing artificial intelligence-based floating object detection methods are also very unstable and have poor adaptability due to reasons such as water surface fluctuations, floating object movement, weather changes, floating object types and changes in size. The present invention discloses a method, device and product for identifying and detecting floating objects on the water surface. The innovative use of a multi-branch backbone network not only ensures the extraction of shallow spatial features of the input image, but also can extract the global context fusion semantic features of the input image, and can adaptively and efficiently fuse the two types of features, effectively improving the accuracy of the identification and detection of floating objects on the water surface, and improving the applicability to various application scenarios.

[0041] The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0042] In one embodiment of the present invention, Figure 2 As shown, the present invention discloses a method for identifying and detecting floating objects on a water surface, comprising:

[0043] Step S201: Establishing a neural network model and using a labeled water surface floating object image dataset to train the neural network model to obtain a neural network model capable of identifying water surface floating objects;

[0044] In this embodiment, a labeling tool such as LabelImg or VGG Image Annotator is used to manually label the captured images. The locations of floating objects on the water surface are selected and category labels are assigned to them, such as "plastic bottle" or "wood block". Finally, a labeled image dataset of floating objects on the water surface is obtained.

[0045] In this embodiment, the annotated images are transformed, such as rotated, cropped, scaled, and color-changed, to increase the diversity of the data set, thereby helping the model to perform more robustly in different environments. This enables the neural network model to stably detect floating objects under different lighting, angles, and backgrounds, thereby avoiding overfitting of the neural network model and improving the generalization ability of the neural network model.

[0046] The neural network model can be trained using any one of the DioU loss function, the CioU loss function, or the EioU loss function. Preferably, this embodiment uses the DioU loss function for training.

[0047] Step S202: using the trained neural network model to identify and detect images that may contain floating objects on the water surface;

[0048] In this embodiment, the trained neural network model is deployed on edge devices for floating object detection, significantly improving the system's real-time performance, reliability, cost-effectiveness, privacy protection, and network independence. Edge computing offers a powerful solution, particularly in scenarios requiring efficient processing of large amounts of data and ensuring continuous system operation. By optimizing hardware and utilizing dedicated AI accelerators, edge devices can efficiently handle floating object detection tasks with lower power consumption, minimal latency, and reduced operating costs.

[0049] In this embodiment, the detected floating objects on the water surface are visualized in real time on the input image, and the bounding boxes and category labels are drawn on the image to display the detection results. OpenCV or other graphics rendering libraries can be used to achieve real-time display of the image. Furthermore, the results detected by the edge device can be uploaded to the cloud for storage or further analysis through efficient data transmission protocols, such as HTTP or WebSocket. The cloud server can perform in-depth analysis, historical data storage or cross-regional collaboration based on this data, providing more powerful support for the task of detecting floating objects on the water surface.

[0050] Step S203, establishing a neural network model includes: establishing a neural network model based on the RT-DETR neural network model, replacing the backbone network of the RT-DETR neural network model with a multi-branch backbone network as the neural network model;

[0051] Step S204, the multi-branch backbone network includes: using the shallow spatial feature module to extract the shallow spatial features of the input image; using the semantic feature module to extract the context fusion semantic features of the input image; using the fusion module to fuse the shallow spatial features and the context fusion semantic features and output them.

[0052] In this embodiment, the HGStem module is used to obtain basic features of the input image, and the shallow spatial feature module and the semantic feature module both extract features of the input image based on the basic features.

[0053] The shallow spatial feature module and the semantic feature module share an HGStem module, which can effectively reduce the amount of computation.

[0054] In this embodiment, the shallow spatial feature module is primarily responsible for extracting shallow spatial features, preserving high-resolution information, and achieving efficient feature expression while maintaining a lightweight design. The semantic feature module captures semantic features through global context modeling and integrates multi-scale information to achieve context-sensitive expression.

[0055] In another embodiment of the present invention, Figure 7 As shown in the figure, the fusion module is used to fuse the shallow spatial features and the contextual semantic features and output them, including:

[0056] The shallow spatial features and context-fused semantic features are convolved separately and then concatenated to obtain the first concatenated features;

[0057] Perform Sigmoid normalization on the first concatenated feature to generate dynamic alignment weights;

[0058] Perform split weight division on the dynamic alignment weight to obtain weight parameters corresponding to shallow spatial features and context-fused semantic features respectively;

[0059] The fused output of shallow spatial features and context-fused semantic features is the weighted sum of the two according to the corresponding weight parameters.

[0060] In this embodiment, the fusion module uses adaptively adjustable weight parameters to dynamically adjust the feature weights of the two paths of shallow spatial features and context-fused semantic features, thereby enhancing the model's adaptability to diverse features, avoiding feature loss, ensuring efficient combination of global features, and providing higher detection performance and reasoning efficiency for target detection tasks.

[0061] In another embodiment of the present invention, the fusion module may also directly use the multi-scale adaptive spatial attention gate MASAG module.

[0062] The multi-scale adaptive spatial attention gate (MASAG) module can effectively highlight key features while suppressing redundant information, thereby improving the model's target recognition ability in complex scenes.

[0063] In another embodiment of the present invention, Figure 3 As shown, after replacing the backbone network of the RT-DETR neural network model with a multi-branch backbone network as the neural network model, it also includes: replacing the attention image feature fusion AIFI module of the neural network model with a dynamic range histogram self-attention DHSA module or a cascade group attention CGA module.

[0064] Dynamic Range Histogram Self-Attention (DHSA) dynamically adjusts the self-attention weight based on the pixel value distribution in the image, helping the model adapt to different pixel ranges and visual features. The accuracy of the self-attention mechanism is improved through dynamic adjustment of the histogram. Especially in scenes with large brightness changes, dynamic range histogram self-attention (DHSA) can better capture long-term dependencies in the image away from the region of interest, effectively suppress background noise, and improve the detection performance of the model.

[0065] In another embodiment of the present invention, Figure 4 As shown in Figure 1, the shallow spatial feature module is composed of a cascade stack of a Conv module and a C2f module, including a Conv module and a C2f module.

[0066] In another embodiment of the present invention, Figure 5 As shown in Figure 1, the semantic feature module is composed of a cascade stack of a DWConv module and an HGBlock module, including a DWConv module and an HGBlock module.

[0067] In another embodiment of the present invention, Figure 6As shown in Figure 1, the semantic feature module can also be composed of a DWConv module and three HGBlock modules cascaded and stacked.

[0068] In one embodiment of the present invention, Figure 3 As shown in the figure, the multi-branch backbone network is represented as DuoAlignNet, the shallow spatial feature module is represented as LEB module, and the fusion module is represented as AFA module; when the semantic feature module is composed of a DWConv module and a HGBlock module in cascade, it is represented as CFB module; when the semantic feature module is composed of a DWConv module and three HGBlock modules in cascade, it is represented as SCFB module;

[0069] like Figure 3 As shown, in this embodiment, the multi-branch backbone network includes three cascaded feature extraction and fusion modules:

[0070] The first feature extraction and fusion module includes: after the input image is processed by the HGStem module to extract basic features, it is processed by the C2f module and the HGBlock module respectively, and then input into the corresponding LEB module and CFB module; after the AFA module fuses the two features output by the LEB module and the CFB module, it inputs them into the Neck module of the neural network model and the second feature extraction and fusion module of the next layer respectively;

[0071] The second feature extraction and fusion module includes: the fusion features output by the first feature extraction and fusion module are input into the LEB module and the SCFB module respectively; the AFA module fuses the two features output by the LEB module and the SCFB module, and then inputs them into the Neck module of the neural network model and the third feature extraction and fusion module of the next layer respectively;

[0072] The third feature extraction and fusion module includes: the fusion features output by the second feature extraction and fusion module are input into the LEB module and the CFB module respectively, and the MASAG module fuses the two features output by the LEB module and the CFB module, and then inputs them into the Neck module of the neural network model;

[0073] In this embodiment, the MASAG module may also be replaced by an AFA module.

[0074] The present invention also discloses an electronic device, such as Figure 8 As shown, an embodiment is disclosed, which is a block diagram of an electronic device for identifying and detecting the above-mentioned floating objects on the water surface.

[0075] The electronic device 80 of this embodiment includes a processor 801, which can perform various appropriate actions and processes according to the program stored in the ROM 802 or the program loaded from the storage part 808 into the RAM 803. The processor 801 may include, for example, a general-purpose microprocessor, an instruction set processor and / or a related chipset and / or a dedicated microprocessor, etc. The processor 801 may also include onboard memory for caching purposes. The processor 801 may include a single processing unit or multiple processing units for performing different actions of the method flow according to the embodiment of the present invention.

[0076] The RAM 803 stores various programs and data required for the operation of the electronic device 80. The processor 801, ROM 802, and RAM 803 are connected to each other via a bus 804. The processor 801 executes the programs in the ROM 802 and / or RAM 803 to perform various operations according to the method flow of the embodiment of the present invention. It should be noted that the programs may also be stored in one or more memories other than the ROM 802 and RAM 803, and the processor 801 may also execute the programs stored in one or more memories to perform various operations according to the method flow of the embodiment of the present invention.

[0077] According to an embodiment of the present invention, the electronic device 80 may further include an I / O interface 805, which is also connected to the bus 804. The electronic device 80 may further include one or more of the following components connected to the I / O interface 805: an input unit 806 including a keyboard, a mouse, etc.; an output unit 807 including a cathode ray tube, a liquid crystal display, and a speaker; a storage unit 808 including a hard disk; and a communication unit 809 including a network interface card such as a LAN card or a modem. The communication unit 809 performs communication processing via a network such as the Internet. A drive 8010 is also connected to the I / O interface 805 as needed. A removable medium 8011, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed in the drive 8010 as needed, so that a computer program read therefrom can be installed into the storage unit 808 as needed.

[0078] like Figure 9 As shown, the present invention also discloses a device for identifying and detecting floating objects on the water surface, the device comprising:

[0079] The model building module 901 is used to establish a neural network model and train the neural network model using a labeled water surface floating object image dataset to obtain a neural network model capable of identifying water surface floating objects. Establishing the neural network model includes: establishing a neural network model based on the RT-DETR neural network model, replacing the backbone network of the RT-DETR neural network model with a multi-branch backbone network as the neural network model; the multi-branch backbone network includes: extracting shallow spatial features of the input image using a shallow spatial feature module; extracting context-fused semantic features of the input image using a semantic feature module; and fusing the shallow spatial features and the context-fused semantic features using a fusion module and outputting the result.

[0080] The recognition and detection module 902 is used to use the trained neural network model to recognize and detect images that may contain floating objects on the water surface.

[0081] The present invention also provides a computer-readable storage medium.

[0082] The computer-readable storage medium may be included in the electronic device / device system described in the above embodiments, or may exist independently and not be incorporated into the electronic device / device. The computer-readable storage medium carries one or more programs that, when executed, implement the method according to the embodiments of the present invention.

[0083] According to embodiments of the present invention, a computer-readable storage medium may be a non-volatile computer-readable storage medium. Examples include, but are not limited to, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM) or flash memory, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0084] Embodiments of the present invention also include a computer program product.

[0085] The computer program product includes a computer program, which contains program code for executing the method provided by the embodiment of the present invention. When the computer program product runs on an electronic device, the program code is used to enable the electronic device to implement the method provided by the embodiment of the present invention.

[0086] In one embodiment, the computer program may be stored on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may be transmitted and distributed in the form of a signal over a network medium. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to wireless, wired, or any suitable combination thereof.

[0087] According to an embodiment of the present invention, the program code for executing the computer program provided by the embodiment of the present invention can be written by any combination of one or more programming languages. Specifically, these computing programs can be implemented using high-level procedural and / or object-oriented programming languages. Programming languages include, but are not limited to, Java, C++, Python, C language or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network or a wide area network, or can be connected to an external computing device.

[0088] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of the systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each box in the flowchart or block diagram may represent a module, program segment, or portion of code, which contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the boxes may occur in an order different from that marked in the accompanying drawings. For example, two boxes shown in succession may actually be executed substantially in parallel, or they may sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, as well as the combination of boxes in the block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or may be implemented using a combination of dedicated hardware and computer instructions. It will be understood by those skilled in the art that the features described in the various embodiments and / or claims of the present invention may be combined and / or coupled in various ways, even if such combinations or couplings are not explicitly described in the present invention. In particular, without departing from the spirit and teachings of the present invention, the features described in the various embodiments and / or claims of the present invention may be combined and / or coupled in various ways, and all such combinations and / or couplings fall within the scope of the present invention.

[0089] The embodiments of the present invention have been described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of the present invention. Although each embodiment has been described above separately, this does not mean that the measures in each embodiment cannot be advantageously used in combination. The scope of the present invention is defined by the appended claims and their equivalents. Without departing from the scope of the present invention, those skilled in the art may make various substitutions and modifications, which should all fall within the scope of the present invention.

Claims

1. A method for identifying and detecting floating objects on a water surface, characterized in that: include: Establishing a neural network model and using a labeled water surface floating object image dataset to train the neural network model to obtain the neural network model capable of identifying water surface floating objects; Using the trained neural network model to identify and detect images that may contain floating objects on the water surface; The establishing of the neural network model comprises: establishing the neural network model based on the RT-DETR neural network model, replacing the backbone network of the RT-DETR neural network model with a multi-branch backbone network as the neural network model; The multi-branch backbone network comprises: extracting shallow spatial features of an input image using a shallow spatial feature module; extracting context-fused semantic features of the input image using a semantic feature module; and fusing the shallow spatial features and the context-fused semantic features using a fusion module and outputting the result. The multi-branch backbone network includes three cascaded feature extraction and fusion modules: The first feature extraction and fusion module includes: after the input image is processed by the HGStem module to extract basic features, it is processed by the C2f module and the HGBlock module respectively, and then input into the corresponding LEB module and CFB module; after the AFA module fuses the two features output by the LEB module and the CFB module, it inputs them into the Neck module of the neural network model and the second feature extraction and fusion module of the next layer respectively; The second feature extraction and fusion module includes: the fusion features output by the first feature extraction and fusion module are input into the LEB module and the SCFB module respectively; the AFA module fuses the two features output by the LEB module and the SCFB module, and then inputs them into the Neck module of the neural network model and the third feature extraction and fusion module of the next layer respectively; The third feature extraction and fusion module includes: the fusion features output by the second feature extraction and fusion module are input into the LEB module and the CFB module respectively, and the multi-scale adaptive spatial attention gate MASAG module fuses the two features output by the LEB module and the CFB module, and then inputs them into the Neck module of the neural network model; The LEB module is composed of a cascade stack of a Conv module and a C2f module, including a Conv module and a C2f module; The CFB module is composed of a DWConv module and an HGBlock module cascaded and stacked; The SCFB module consists of a DWConv module and three HGBlock modules cascaded and stacked.

2. The method for identifying and detecting floating objects on a water surface according to claim 1, characterized in that: After replacing the backbone network of the RT-DETR neural network model with a multi-branch backbone network as the neural network model, it also includes: replacing the attention image feature fusion AIFI module of the neural network model with a dynamic range histogram self-attention DHSA module or a cascade group attention CGA module.

3. A method for identifying and detecting floating objects on a water surface according to claim 2, characterized in that: Also includes: The HGStem module is used to obtain basic features of the input image, and the shallow spatial feature module and the semantic feature module both extract features of the input image based on the basic features.

4. A device for identifying and detecting floating objects on a water surface, characterized in that: The device comprises: A model building module is used to build a neural network model, and use a labeled water surface floating object image data set to train the neural network model to obtain the neural network model that can identify water surface floating objects; the building of the neural network model includes: building the neural network model based on the RT-DETR neural network model, replacing the backbone network of the RT-DETR neural network model with a multi-branch backbone network as the neural network model; the multi-branch backbone network includes: using a shallow spatial feature module to extract shallow spatial features of the input image; using a semantic feature module to extract context-fused semantic features of the input image; using a fusion module to fuse the shallow spatial features and the context-fused semantic features and output them; the multi-branch backbone network includes three cascaded feature extraction and fusion modules: the first feature extraction and fusion module includes: after the basic features of the input image are extracted by the HGStem module, they are processed by the C2f module and the HGBlock module respectively, and then input into the corresponding LEB module and CFB module, and the AFA module fuses the two outputs of the LEB module and the CFB module. After the features are extracted, they are respectively input into the Neck module of the neural network model and the second feature extraction and fusion module of the next layer; the second feature extraction and fusion module includes: the fusion features output by the first feature extraction and fusion module are respectively input into the LEB module and the SCFB module, the AFA module fuses the two features output by the LEB module and the SCFB module, and then respectively input into the Neck module of the neural network model and the third feature extraction and fusion module of the next layer; the third feature extraction and fusion module includes: the fusion features output by the second feature extraction and fusion module are respectively input into the LEB module and the CFB module, the multi-scale adaptive spatial attention gate MASAG module fuses the two features output by the LEB module and the CFB module, and then input into the Neck module of the neural network model; the LEB module is composed of a cascade stack of a Conv module and a C2f module, including a Conv module and a C2f module; the CFB module is composed of a cascade stack of a DWConv module and an HGBlock module; the SCFB module is composed of a cascade stack of a DWConv module and three HGBlock modules; The recognition and detection module is used to use the trained neural network model to recognize and detect images that may contain floating objects on the water surface.

5. An electronic device, characterized in that: include: one or more processors; A storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors are caused to perform the method according to any one of claims 1 to 3.

6. 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.

Images