Low-cost BSD early warning method and system based on yov10

Abstract

The invention discloses a low-cost BSD early warning method based on yov10, and the method comprises the steps: obtaining an environment image of a preset region of a vehicle, carrying out the fusion of the environment image, and obtaining a fused image; inputting the fused image into a lightweight backbone network MobilenetV4 detection network, and carrying out feature extraction on the image to obtain a feature extraction image; inputting the feature extraction images into a Neck feature fusion layer, and fusing the feature extraction images; and inputting the image after feature fusion into a head layer distributed by double labels, carrying out image identification on the image after feature fusion, and carrying out early warning on the vehicle according to an image identification result. The method does not need a traditional sensor, does not need an independent terminal to complete the function, and greatly reduces the cost of a product. According to the method, data are spliced and fused according to scenes, lightweight optimization and attention mechanism optimization are performed based on the latest yov10 algorithm, and the accuracy of the algorithm is improved while the computing power required by the algorithm is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle-mounted intelligent devices, and in particular to a low-cost BSD early warning method and system based on yolov10. Background Art

[0002] The BSD (Blind Spot Detection) blind spot warning system is a vehicle safety assistance technology designed to help drivers identify and warn of potential dangers in the vehicle's blind spots. The system usually monitors the blind spots around the vehicle through devices such as radar, cameras or ultrasonic sensors, and issues warnings when other vehicles, pedestrians or obstacles are detected to help the driver avoid collision accidents. The current working principle of the BSD system is to monitor the blind spots through sensors installed around the vehicle. These sensors can be radars, cameras or ultrasonic sensors, which can monitor the situation around the vehicle in real time. When the system determines that there is a potential risk of collision, it will warn the driver, usually through sound prompts, visual prompts or vibration seats to remind the driver to pay attention. However, the existing BSD expected system is either implemented by relying on sensor solutions or by independent BSD terminals. Both solutions are costly. Since each car company is currently pursuing extreme costs, a low-cost BSD warning method and system based on yolov10 is urgently needed to solve this problem. Summary of the invention

[0003] In view of the above problems, the present invention is proposed to provide a low-cost BSD early warning method and system based on yolov10 that overcomes the above problems or at least partially solves the above problems.

[0004] In order to solve the above technical problems, the embodiments of the present application disclose the following technical solutions:

[0005] In a first aspect, an embodiment of the present invention discloses a low-cost BSD early warning method based on yolov10, comprising:

[0006] S100. Obtaining an environmental image of a preset area of ​​the vehicle, and fusing the environmental image to obtain a fused image;

[0007] S200. Input the fused image into a lightweight backbone network MobilenetV4 detection network, perform feature extraction on the image, and obtain a feature-extracted image;

[0008] S300. Inputting the feature extraction image into the Neck feature fusion layer, and fusing the feature extraction image;

[0009] S400. Input the feature-fused image into the head layer of dual label allocation, perform image recognition on the feature-fused image, and issue an early warning to the vehicle according to the image recognition result.

[0010] Furthermore, in S100, an environmental image of a preset area of ​​the vehicle is obtained, and the environmental image is fused to obtain a fused image. The specific method includes: obtaining the vehicle environmental image through cameras installed in the front, left front and right front of the vehicle, and cropping, splicing and Mosaic data enhancement of the environmental image to obtain a fused image.

[0011] Furthermore, the environment image is cropped, spliced ​​and Mosaic data enhanced, and the specific method includes: obtaining the front, left front and right front images of the vehicle, and applying a cropping operation using a cropping tool according to the determined cropping area, retaining only the image portion within the cropping area and discarding the rest, to obtain a cropped image;

[0012] Determine if the cropped images have some overlap in content, stitch the cropped images together through the automatic alignment and fusion functions of the image stitching software, optimize the stitched images using color correction, brightness and contrast adjustment, edge fusion, noise removal and sharpening methods to obtain a Mosaic image, and input the generated Mosaic image as training data into a machine learning or deep learning model to enhance the Mosaic data.

[0013] Furthermore, in S200, the fused image is input into a lightweight backbone network MobilenetV4 detection network to perform feature extraction on the image. The specific method includes: loading a pre-trained MobileNetV4 model, modifying the MobileNetV4 model so that the MobileNetV4 model returns a feature map; scaling the fused image to the input size required by the MobileNetV4 model, and performing standardization; inputting the preprocessed image into the modified MobileNetV4 model to perform feature extraction on the image.

[0014] Further, in S300, the feature extraction image is input into the Neck feature fusion layer, and the Neck layer includes an FPN module and a PAN module. The FPN module is used to construct a network with a pyramid-shaped feature hierarchy with high-level semantic information, and enhance the semantic information of the feature map through a top-down path and a lateral connection. The FPN module starts from the highest layer of the backbone network, and combines the high-level feature map with the lower-level feature map through upsampling and lateral connection. The PAN layer adds a bottom-up path on the basis of the FPN layer to enhance the expression ability of the feature pyramid. Through the bottom-up path, the PAN layer can effectively transfer the detailed information of the low-level feature map to the high-level feature map, thereby compensating for the detail information lost in the high-level feature map of FPN due to multiple upsampling.

[0015] Furthermore, when the Neck layer contains both FPN and PAN modules, a feature pyramid with rich semantic information is first constructed through the FPN module, and then the PAN module further enhances the expressiveness of the feature map based on the FPN module, and effectively integrates the detailed information of the low-level feature map into the high-level feature map through a bottom-up path.

[0016] Further, in S400, the feature-fused image is input into the head layer of the dual label assignment, and image recognition is performed on the feature-fused image. The specific method includes: designing a detection head, wherein the detection head includes multiple convolutional layers for further processing feature maps and outputting prediction information about the target object, including at least category, position and confidence; assigning dual labels, and implementing an efficient label assignment strategy in the training phase to ensure that each anchor point is properly assigned to one or more target true value boxes, wherein the strategy considers multiple factors between the anchor point and the true value box, including at least intersection ratio, center point distance, and feature quality; calculating the loss function, and calculating the loss function according to the output of the detection head and the result of the label assignment; using the calculated loss to perform back propagation and update the weights of the model, and inputting the feature map after feature fusion into the trained detection head to obtain the prediction result of the target in the image.

[0017] In a second aspect, an embodiment of the present invention discloses a low-cost BSD early warning system based on yolov10, including an image fusion unit, an image feature extraction unit, an image fusion unit and an image recognition unit; wherein:

[0018] An image fusion unit is used to obtain an environmental image of a preset area of ​​the vehicle, and fuse the environmental image to obtain a fused image;

[0019] An image feature extraction unit, used for inputting the fused image into a lightweight backbone network MobilenetV4 detection network, performing feature extraction on the image, and obtaining a feature-extracted image;

[0020] An image fusion unit, used for inputting the feature extraction image into a Neck feature fusion layer to fuse the feature extraction image;

[0021] The image recognition unit is used to input the feature-fused image into the head layer of the dual label allocation, perform image recognition on the feature-fused image, and issue an early warning to the vehicle according to the image recognition result.

[0022] Furthermore, an image fusion unit is used to input the feature extraction image into the Neck feature fusion layer to fuse the feature extraction image; the Neck layer includes an FPN module and a PAN module, the FPN module is used to construct a network with a pyramid-shaped feature hierarchy with high-level semantic information, and enhance the semantic information of the feature map through a top-down path and a lateral connection. The FPN module starts from the highest layer of the backbone network and combines the high-level feature map with the lower-level feature map through upsampling and lateral connections; the PAN layer adds a bottom-up path on the basis of the FPN layer to enhance the expression ability of the feature pyramid. Through the bottom-up path, the PAN layer can effectively transfer the detailed information of the low-level feature map to the high-level feature map, thereby making up for the detail information lost in the high-level feature map of FPN due to multiple upsampling.

[0023] In a third aspect, an embodiment of the present invention discloses an electronic device, including:

[0024] one or more processors;

[0025] A memory for storing one or more programs;

[0026] When the one or more programs are executed by the one or more processors, the one or more processors implement the early warning method.

[0027] The beneficial effects of the above technical solution provided by the embodiment of the present invention include at least:

[0028] The present invention discloses a low-cost BSD early warning method and system based on yolov10, the method comprising: obtaining an environmental image of a preset area of ​​a vehicle, fusing the environmental image to obtain a fused image; inputting the fused image into a lightweight backbone network MobilenetV4 detection network, extracting features from the image, and obtaining a feature-extracted image; inputting the feature-extracted image into a Neck feature fusion layer, fusing the feature-extracted image; inputting the feature-fused image into a head layer of dual label allocation, performing image recognition on the feature-fused image, and early warning the vehicle according to the image recognition result. The present invention does not require traditional sensors or independent terminals to complete the function, which greatly reduces the cost of the product. The present invention splices and fuses data according to the scene, and at the same time performs lightweight optimization and attention mechanism optimization based on the latest yolov10 algorithm, which reduces the computing power required for the algorithm while improving the accuracy of the algorithm.

[0029] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0031] Figure 1 This is a flow chart of a low-cost BSD early warning method based on yolov10 in Example 1 of the present invention;

[0032] Figure 2 This is a schematic diagram of the structure of an electronic device in Example 3 of the present invention. DETAILED DESCRIPTION

[0033] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0034] In order to solve the problems existing in the prior art, the embodiment of the present invention provides a low-cost BSD early warning method and system based on yolov10.

[0035] Example 1

[0036] The present invention discloses a low-cost BSD early warning method based on yolov10, such as Figure 1 ,include:

[0037] S100. Obtaining the environment image of the preset area of ​​the vehicle, fusing the environment image to obtain a fused image; cropping, splicing and Mosaic data enhancement of the environment image are common operations in image processing, which are widely used in the fields of image preprocessing, data enhancement and image analysis. In S100 of this embodiment, obtaining the environment image of the preset area of ​​the vehicle, fusing the environment image to obtain a fused image, the specific method includes: obtaining the vehicle environment image through the cameras installed in front, left front and right front of the vehicle, cropping, splicing and Mosaic data enhancement of the environment image to obtain a fused image.

[0038] Specifically, the environment image is cropped, spliced, and Mosaic data enhanced, and the specific method includes:

[0039] Acquire the front, left and right front images of the vehicle, and apply a cropping operation using a cropping tool according to the determined cropping area, retain only the image portion within the cropping area and discard the rest, to obtain a cropped image;

[0040] Determine that the cropped images have some overlap in content, stitch the cropped images together through the automatic alignment and fusion functions of the image stitching software, optimize the stitched images by color correction, brightness and contrast adjustment, edge fusion, noise removal and sharpening, and obtain a Mosaic image. The generated Mosaic image is input as training data into the machine learning or deep learning model to enhance the Mosaic data. In short, image cropping and stitching are basic operations in image processing, which can be achieved through a variety of software and tools. When choosing tools, you can choose the most suitable method according to your needs and proficiency. At the same time, when cropping and stitching, you need to pay attention to maintaining the clarity and proportion of the image to obtain the best visual effect.

[0041] In summary, the optimization of the stitched image needs to start from multiple aspects, including color correction, brightness and contrast adjustment, edge fusion, noise removal and sharpening, transformation model optimization, automatic processing and algorithm optimization, evaluation and iteration, etc. By combining these methods, the quality and visual effect of the stitched image can be significantly improved.

[0042] S200. Input the fused image into the lightweight backbone network MobilenetV4 detection network, perform feature extraction on the image, and obtain a feature-extracted image; wherein MobileNetV4 (MNv4) is a general and efficient neural network architecture designed for mobile devices, which combines a variety of advanced network design elements and technologies to achieve efficient target detection performance. The MobileNetV4 detection network is suitable for target detection tasks on various resource-constrained mobile or embedded devices. For example, it can be used for real-time target detection on devices such as smartphones, drones, and smart cameras, such as face recognition, vehicle detection, pedestrian detection, etc.

[0043] In S200 of this embodiment, the fused image is input into a lightweight backbone network MobilenetV4 detection network, and features are extracted from the image. The specific method includes: loading a pre-trained MobileNetV4 model, modifying the MobileNetV4 model so that the MobileNetV4 model returns a feature map; scaling the fused image to the input size required by the MobileNetV4 model, and performing standardization; inputting the preprocessed image into the modified MobileNetV4 model, and performing feature extraction on the image.

[0044] S300. Input the feature extraction image into the Neck feature fusion layer to fuse the feature extraction image; the Neck feature fusion layer is located between the backbone network (such as MobileNetV4) and the detection head. The main purpose of the Neck feature fusion layer is to further process and fuse the feature map output by the backbone network to enhance the expression ability of the features and provide better input for subsequent detection tasks.

[0045] Feature fusion is an important function of the Neck layer, which can be implemented in many ways, such as Feature Pyramid Network (FPN), Path Aggregation Network (PANet), Dual Feature Pyramid Network (BiFPN), etc. These techniques aim to combine feature maps from different layers in an effective way to exploit the spatial details of low-level features and the semantic information of high-level features.

[0046] In S300 of this embodiment, the feature extraction image is input into the Neck feature fusion layer, and the Neck layer includes an FPN module and a PAN module. The FPN module is used to construct a network with a pyramid-shaped feature hierarchy with high-level semantic information, and enhance the semantic information of the feature map through a top-down path and a lateral connection. The FPN module starts from the highest layer of the backbone network, and combines the high-level feature map with the lower-level feature map through upsampling and lateral connection. The PAN layer adds a bottom-up path on the basis of the FPN layer to enhance the expression ability of the feature pyramid. Through the bottom-up path, the PAN layer can effectively transfer the detailed information of the low-level feature map to the high-level feature map, thereby compensating for the detail information lost in the high-level feature map of FPN due to multiple upsampling.

[0047] In this embodiment, when the Neck layer includes both FPN and PAN modules, a feature pyramid with rich semantic information is first constructed through the FPN module, and then the PAN module further enhances the expression ability of the feature map based on the FPN module, and effectively integrates the detail information of the low-level feature map into the high-level feature map through a bottom-up path.

[0048] S400. Input the feature-fused image into the head layer of dual label assignment, perform image recognition on the feature-fused image, and warn the vehicle based on the image recognition result. The head layer is responsible for converting feature maps into final prediction results, such as category probabilities for classification tasks or bounding boxes and category labels for detection tasks. For a network such as MobileNetV4, although it is mainly designed as a lightweight feature extractor, it can also include head layers for different tasks.

[0049] In S400 of this embodiment, the feature-fused image is input into the head layer of the dual label assignment, and image recognition is performed on the feature-fused image. The specific method includes: designing a detection head, wherein the detection head includes multiple convolutional layers for further processing feature maps and outputting prediction information about the target object, including at least category, position and confidence; assigning dual labels, and implementing an efficient label assignment strategy in the training phase to ensure that each anchor point is properly assigned to one or more target true value boxes, wherein the strategy considers multiple factors between the anchor point and the true value box, including at least intersection-over-union ratio, center point distance, and feature quality; calculating the loss function, and calculating the loss function according to the output of the detection head and the result of the label assignment; using the calculated loss to perform back propagation and update the weights of the model, and inputting the feature map after feature fusion into the trained detection head to obtain the prediction result of the target in the image.

[0050] The present embodiment discloses a low-cost BSD warning method based on yolov10, including: obtaining an environmental image of a preset area of ​​a vehicle, fusing the environmental image to obtain a fused image; inputting the fused image into a lightweight backbone network MobilenetV4 detection network, extracting features from the image, and obtaining a feature-extracted image; inputting the feature-extracted image into a Neck feature fusion layer, fusing the feature-extracted image; inputting the feature-fused image into a head layer of dual label allocation, performing image recognition on the feature-fused image, and warning the vehicle according to the image recognition result. The present invention does not require traditional sensors or independent terminals to complete the function, which greatly reduces the cost of the product. The present invention splices and fuses data according to the scene, and at the same time performs lightweight optimization and attention mechanism optimization based on the latest yolov10 algorithm, which reduces the computing power required for the algorithm while improving the accuracy of the algorithm.

[0051] Example 2

[0052] Based on the same inventive concept, the disclosed embodiment also provides a low-cost BSD early warning system based on yolov10, including an image fusion unit, an image feature extraction unit, an image fusion unit and an image recognition unit; wherein:

[0053] An image fusion unit is used to obtain an environmental image of a preset area of ​​the vehicle, and fuse the environmental image to obtain a fused image;

[0054] An image feature extraction unit, used for inputting the fused image into a lightweight backbone network MobilenetV4 detection network, performing feature extraction on the image, and obtaining a feature-extracted image;

[0055] An image fusion unit, used for inputting the feature extraction image into a Neck feature fusion layer to fuse the feature extraction image;

[0056] The image recognition unit is used to input the feature-fused image into the head layer of the dual label allocation, perform image recognition on the feature-fused image, and issue an early warning to the vehicle according to the image recognition result.

[0057] In this embodiment, the image fusion unit is used to input the feature extraction image into the Neck feature fusion layer to fuse the feature extraction image; the Neck layer includes an FPN module and a PAN module, the FPN module is used to construct a network with a pyramid-shaped feature hierarchy with high-level semantic information, and enhance the semantic information of the feature map through a top-down path and a lateral connection. The FPN module starts from the highest layer of the backbone network and combines the high-level feature map with the lower-level feature map through upsampling and lateral connections; the PAN layer adds a bottom-up path on the basis of the FPN layer to enhance the expression ability of the feature pyramid. Through the bottom-up path, the PAN layer can effectively transfer the detailed information of the low-level feature map to the high-level feature map, thereby making up for the detail information lost in the high-level feature map of FPN due to multiple upsampling.

[0058] Among them, other specific working methods of the image fusion unit, the image feature extraction unit, the image fusion unit and the image recognition unit have been described in detail in Example 1, and will not be repeated in this embodiment.

[0059] Example 3

[0060] Based on the same inventive concept, an embodiment of the present disclosure also provides an electronic device. Figure 2 FIG. 1 is a schematic diagram of the structure of an electronic device according to an embodiment of the present disclosure. Figure 2 As shown, an embodiment of the present disclosure provides an electronic device including: one or more processors 101, a memory 102, and one or more I / O interfaces 103. The memory 102 stores one or more programs, and when the one or more programs are executed by the one or more processors, the one or more processors implement any optimization method in the above embodiments; the one or more I / O interfaces 103 are connected between the processor and the memory, and are configured to implement information interaction between the processor and the memory.

[0061] Among them, the processor 101 is a device with data processing capabilities, including but not limited to a central processing unit (CPU), etc.; the memory 102 is a device with data storage capabilities, including but not limited to random access memory (RAM, more specifically SDRAM, DDR, etc.), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory (FLASH); the I / O interface (read-write interface) 103 is connected between the processor 101 and the memory 102, and can realize information interaction between the processor 101 and the memory 102, including but not limited to a data bus (Bus), etc.

[0062] In some embodiments, the processor 101 , the memory 102 , and the I / O interface 103 are connected to each other via a bus 104 , and further connected to other components of the computing device.

[0063] In some embodiments, the one or more processors 101 include a field programmable gate array.

[0064] According to an embodiment of the present disclosure, a computer-readable medium is further provided, wherein a computer program is stored on the computer-readable medium, wherein when the program is executed by a processor, the steps in any optimization method in the above-mentioned embodiment are implemented.

[0065] It should be understood that the specific order or hierarchy of steps in the disclosed process is an example of an exemplary method. Based on design preferences, it should be understood that the specific order or hierarchy of steps in the process can be rearranged without departing from the scope of protection of the present disclosure. The attached method claims present the elements of the various steps in an exemplary order and are not intended to be limited to the specific order or hierarchy described.

[0066] In the above detailed description, various features are grouped together in a single embodiment to simplify the disclosure. This method of disclosure should not be interpreted as reflecting an intention that the embodiments of the claimed subject matter require more features than are clearly stated in each claim. On the contrary, as reflected in the appended claims, the invention is in a state of having less than all the features of the disclosed individual embodiments. Therefore, the appended claims are hereby expressly incorporated into the detailed description, with each claim standing on its own as a separate preferred embodiment of the invention.

[0067] Those skilled in the art will also appreciate that the various illustrative logic blocks, modules, circuits, and algorithmic steps described in conjunction with the embodiments herein can all be implemented as electronic hardware, computer software, or a combination thereof. In order to clearly illustrate the interchangeability between hardware and software, various illustrative components, blocks, modules, circuits, and steps are generally described above around their functions. Whether such functions are implemented as hardware or software depends on specific applications and the design constraints imposed on the entire system. A skilled person can implement the described functions in an alternative manner for each specific application, but such implementation decisions should not be interpreted as departing from the scope of protection of the present disclosure.

[0068] The steps of the method or algorithm described in conjunction with the embodiments herein may be directly embodied as hardware, a software module executed by a processor, or a combination thereof. The software module may be located in a RAM memory, a flash memory, a ROM memory, an EPROM memory, an EEPROM memory, a register, a hard disk, a mobile disk, a CD-ROM, or any other form of storage medium well known in the art. An exemplary storage medium is connected to the processor so that the processor can read information from the storage medium and can write information to the storage medium. Of course, the storage medium may also be an integral part of the processor. The processor and the storage medium may be located in an ASIC. The ASIC may be located in a user terminal. Of course, the processor and the storage medium may also be present in a user terminal as discrete components.

[0069] For software implementation, the techniques described in this application can be implemented with modules (e.g., procedures, functions, etc.) that perform the functions described in this application. These software codes can be stored in a memory unit and executed by a processor. The memory unit can be implemented within the processor or outside the processor. In the latter case, it is coupled to the processor in a communication manner via various means, which are well known in the art.

[0070] The above description includes examples of one or more embodiments. Of course, it is impossible to describe all possible combinations of components or methods for the purpose of describing the above embodiments, but it should be recognized by those skilled in the art that the various embodiments may be further combined and arranged. Therefore, the embodiments described herein are intended to cover all such changes, modifications and variations that fall within the scope of protection of the appended claims. In addition, with respect to the term "comprising" used in the specification or claims, the word is covered in a manner similar to the term "including", just as "including," is explained as a transitional word in the claims. In addition, any term "or" used in the specification of the claims is intended to mean "non-exclusive or".

Claims

1. A low-cost BSD early warning method based on yolov10, characterized in that: include: S100. Obtaining an environmental image of a preset area of ​​the vehicle, and fusing the environmental image to obtain a fused image; S200. Input the fused image into a lightweight backbone network MobilenetV4 detection network, perform feature extraction on the image, and obtain a feature-extracted image; S300. Inputting the feature extraction image into the Neck feature fusion layer to fuse the feature extraction image; S400. Input the feature-fused image into the head layer of dual label allocation, perform image recognition on the feature-fused image, and issue an early warning to the vehicle according to the image recognition result.

2. A low-cost BSD early warning method based on yolov10 as claimed in claim 1, characterized in that: In S100, an environmental image of a preset area of ​​the vehicle is obtained, and the environmental image is fused to obtain a fused image. The specific method includes: obtaining the vehicle environmental image through cameras installed in the front, left front and right front of the vehicle, and cropping, splicing and Mosaic data enhancement of the environmental image to obtain a fused image.

3. A low-cost BSD early warning method based on yolov10 as claimed in claim 2, characterized in that: The environment image is cropped, spliced ​​and Mosaic data enhanced, and the specific method includes: Acquire the front, left and right front images of the vehicle, and apply a cropping operation using a cropping tool according to the determined cropping area, retain only the image portion within the cropping area and discard the rest, to obtain a cropped image; Determine if the cropped images have some overlap in content, stitch the cropped images together through the automatic alignment and fusion functions of the image stitching software, optimize the stitched images using color correction, brightness and contrast adjustment, edge fusion, noise removal and sharpening methods to obtain a Mosaic image, and input the generated Mosaic image as training data into a machine learning or deep learning model to enhance the Mosaic data.

4. A low-cost BSD early warning method based on yolov10 as claimed in claim 1, characterized in that: In S200, the fused image is input into a lightweight backbone network MobilenetV4 detection network, and features of the image are extracted. The specific method includes: loading a pre-trained MobileNetV4 model, modifying the MobileNetV4 model so that the MobileNetV4 model returns a feature map; scaling the fused image to the input size required by the MobileNetV4 model, and performing standardization; inputting the preprocessed image into the modified MobileNetV4 model, and extracting features of the image.

5. A low-cost BSD early warning method based on yolov10 as claimed in claim 1, characterized in that: In S300, the feature extraction image is input into the Neck feature fusion layer, and the Neck layer includes an FPN module and a PAN module. The FPN module is used to construct a network with a pyramid-shaped feature hierarchy with high-level semantic information, and enhance the semantic information of the feature map through a top-down path and a lateral connection. The FPN module starts from the highest layer of the backbone network, and combines the high-level feature map with the lower-level feature map through upsampling and lateral connection. The PAN layer adds a bottom-up path on the basis of the FPN layer to enhance the expression ability of the feature pyramid. Through the bottom-up path, the PAN layer can effectively transfer the detailed information of the low-level feature map to the high-level feature map, thereby compensating for the detail information lost in the high-level feature map of FPN due to multiple upsampling.

6. A low-cost BSD early warning method based on yolov10 as claimed in claim 5, characterized in that: When the Neck layer contains both FPN and PAN modules, a feature pyramid with rich semantic information is first constructed through the FPN module. Then, the PAN module further enhances the expressiveness of the feature map based on the FPN module, and effectively integrates the detailed information of the low-level feature map into the high-level feature map through a bottom-up path.

7. A low-cost BSD early warning method based on yolov10 as claimed in claim 1, characterized in that: In S400, the image after feature fusion is input into the head layer of dual label assignment, and image recognition is performed on the image after feature fusion. The specific method includes: designing a detection head, wherein the detection head includes multiple convolutional layers for further processing feature maps and outputting prediction information about the target object, including at least category, position and confidence; assigning dual labels, and implementing an efficient label assignment strategy in the training stage to ensure that each anchor point is properly assigned to one or more target true value boxes, and the strategy considers multiple factors between the anchor point and the true value box, including at least intersection ratio, center point distance, and feature quality; loss function calculation, calculating the loss function according to the output of the detection head and the result of label assignment; using the calculated loss to perform back propagation and update the weight of the model, and inputting the feature map after feature fusion into the trained detection head to obtain the prediction result of the target in the image.

8. A low-cost BSD early warning system based on yolov10, characterized in that: It includes an image fusion unit, an image feature extraction unit, an image fusion unit and an image recognition unit; wherein: An image fusion unit, used to obtain an environmental image of a preset area of ​​the vehicle, and fuse the environmental image to obtain a fused image; An image feature extraction unit, used for inputting the fused image into a lightweight backbone network MobilenetV4 detection network, performing feature extraction on the image, and obtaining a feature-extracted image; An image fusion unit, used for inputting the feature extraction image into a Neck feature fusion layer to fuse the feature extraction image; The image recognition unit is used to input the feature-fused image into the head layer of the dual label allocation, perform image recognition on the feature-fused image, and issue an early warning to the vehicle according to the image recognition result.

9. A low-cost BSD early warning system based on yolov10 as claimed in claim 8, characterized in that: The image fusion unit is used to input the feature extraction image into the Neck feature fusion layer to fuse the feature extraction image; the Neck layer includes an FPN module and a PAN module. The FPN module is used to construct a network with a pyramid-shaped feature hierarchy with high-level semantic information, and enhance the semantic information of the feature map through a top-down path and a lateral connection. The FPN module starts from the highest layer of the backbone network and combines the high-level feature map with the lower-level feature map through upsampling and lateral connection. The PAN layer adds a bottom-up path on the basis of the FPN layer to enhance the expression ability of the feature pyramid. Through the bottom-up path, the PAN layer can effectively transfer the detailed information of the low-level feature map to the high-level feature map, thereby making up for the detailed information lost in the high-level feature map of FPN due to multiple upsampling.

10. An electronic device, comprising: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the early warning method.

Images