Optical splitter port detection method and device, and program product

Abstract

The invention discloses an optical splitter port detection method and device and a program product, and the method comprises the steps: carrying out the feature extraction of an optical splitter port image through a feature extraction module in a port detection model, and obtaining a feature graph; and determining a feature optimization weight of the feature map, and performing multi-scale feature optimization to obtain an optimized feature map under multiple scales. And performing feature fusion on the plurality of optimized feature maps through a feature fusion module to obtain a fused feature map. Through a port detection module, port state detection and port position detection are carried out, and port state information and port position information of each optical splitter port are determined. According to the technical scheme provided by the invention, high-precision and high-efficiency automatic detection can be carried out on the occupancy state and position of the optical splitter port through the port detection model, the detection speed and accuracy are remarkably improved, the application range and flexibility are remarkably enhanced, and the practicability is high. And the resource utilization efficiency and the port allocation effectiveness in the optical splitter port allocation process are improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of image processing, and particularly relates to a splitter port detection method and device and program product. BACKGROUND

[0002] In the modern field of optical fiber communication, a splitter is a key device for realizing efficient distribution of optical signals, and can significantly improve signal transmission rate and optical fiber resource utilization efficiency. Precise detection of the occupancy state of each port on the splitter is of great help to efficient operation and maintenance of communication networks, resource optimization, and fault troubleshooting.

[0003] Currently, the traditional splitter port detection method has obvious defects and limitations. In addition to manual detection by labor, which is time-consuming and labor-intensive, there is also a port image detection through a multi-step image processing process. This detection process is tedious and time-consuming, which seriously reduces the detection efficiency of the splitter port occupancy, cannot meet the real-time needs of the monitoring personnel, and has limited detection types, such as being able to detect only the splitter with single-row ports. There is strong limitation. There are other ways to detect the ports by detecting the splitter tail fiber. This processing process is heavy and cannot accurately identify when the tail fiber blocks the port. It is inefficient and has strong limitations, which seriously affects the detection efficiency and accuracy of the splitter port occupancy.

[0004] Therefore, how to efficiently and accurately detect the port occupancy of the splitter is an important problem to be solved at present. SUMMARY

[0005] The embodiments of the present application provide a splitter interface detection method, device, equipment and program product, which can efficiently and accurately detect the port occupancy of the splitter, accurately control the port state, and improve the rationality of resource distribution and resource utilization efficiency.

[0006] In a first aspect, the embodiments of the present application provide a splitter interface detection method, comprising: Collecting a splitter port image of a target splitter, the splitter port image containing a plurality of splitter ports; extracting features from the splitter port image through a feature extraction module in a pre-trained port detection model to obtain a feature map and determine a feature optimization weight of the feature map, and performing multi-scale feature optimization on the feature map based on the feature optimization weight to obtain an optimized feature map of the splitter port image in multiple scales; performing feature fusion on the plurality of optimized feature maps through a feature fusion module in the port detection model to obtain a fused feature map corresponding to the splitter port image; The port detection module in the port detection model detects the state of the ports of the optical splitter and the positions of the ports of the optical splitter based on the fused feature map, to determine the port state information and the port position information corresponding to each port of the optical splitter in the port image of the optical splitter, the port state information indicating that the port of the optical splitter is in an occupied state or an unoccupied state.

[0007] In a second aspect, an embodiment of the present application provides an optical splitter port detection device, comprising: The acquisition module is configured to acquire a port image of an optical splitter, the port image comprising a plurality of ports of the optical splitter. The extraction module is configured to extract features from the port image of the optical splitter by a feature extraction module in a pre-trained port detection model, to obtain a feature map and determine a feature optimization weight of the feature map, and perform multi-scale feature optimization on the feature map based on the feature optimization weight, to obtain an optimized feature map of the port image of the optical splitter at a plurality of scales. The fusion module is configured to fuse features of the plurality of optimized feature maps by a feature fusion module in the port detection model, to obtain a fused feature map of the port image of the optical splitter. The detection module is configured to detect the state of the ports of the optical splitter and the positions of the ports of the optical splitter based on the fused feature map by a port detection module in the port detection model, to determine the port state information and the port position information corresponding to each port of the optical splitter in the port image of the optical splitter, the port state information indicating that the port of the optical splitter is in an occupied state or an unoccupied state.

[0008] In a third aspect, an embodiment of the present application provides a terminal device, comprising a processor and a memory storing computer program instructions. The processor executes the computer program instructions to implement the optical splitter port detection of the first aspect.

[0009] In a fourth aspect, an embodiment of the present application provides a computer storage medium, the computer readable storage medium storing computer program instructions, the computer program instructions being executed by a processor to implement the optical splitter port detection of the first aspect.

[0010] In a fifth aspect, an embodiment of the present application provides a computer program product, the instructions in the computer program product being executed by a processor of an electronic device to cause the electronic device to perform the optical splitter port detection method of the first aspect.

[0011] The technical solutions provided by the embodiments of the present application at least bring the following beneficial effects: This application provides a method for detecting the port of a beam splitter, comprising: extracting features from a pre-trained port detection model using a feature extraction module to obtain a feature map from a captured beam splitter port image; determining feature optimization weights for the feature map and performing multi-scale feature optimization to obtain optimized feature maps at multiple scales; fusing the multiple optimized feature maps using a feature fusion module in the port detection model to obtain a fused feature map; and then, using a port detection module in the port detection model, performing port state detection and port position detection based on the fused feature map to determine the port state information and port position information of each beam splitter port in the beam splitter port image.

[0012] The technical solution provided in this application achieves high-precision and high-efficiency automated detection of the occupancy status of optical splitter ports through an end-to-end port detection model. This overcomes the drawbacks of traditional methods, such as cumbersome processes, long processing times, and poor fault tolerance, significantly improving detection speed and accuracy. Furthermore, the application scenarios are no longer limited by port arrangement or partial obstruction by pigtails, significantly enhancing the scope of application and flexibility. This further improves resource utilization efficiency and port allocation effectiveness in the optical splitter port allocation process, providing reliable technical support for network resource management and maintenance quality.

[0013] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0014] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 A schematic flowchart illustrating a method for detecting a beam splitter interface provided in one embodiment of this application; Figure 2 A schematic diagram illustrating the process of determining feature optimization weights according to an embodiment of this application; Figure 3(a) is one of the structural schematic diagrams of a feature extraction module provided in an embodiment of this application; Figure 3(b) is a second schematic diagram of the structure of a feature extraction module provided in one embodiment of this application; Figure 4 A schematic diagram illustrating the determination and fusion process of a third fusion feature map according to an embodiment of this application; Figure 5 This is a schematic diagram of the structure of a port detection module provided in one embodiment of this application; Figure 6 This is a schematic diagram of the structure of a port detection model provided in one embodiment of this application; Figure 7 A schematic diagram of a beam splitter interface detection device provided in another embodiment of this application; Figure 8 This is a schematic diagram of the structure of a terminal device provided in another embodiment of this application. Detailed Implementation

[0016] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0017] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0018] In modern fiber optic communication, optical splitters are key devices for efficient optical signal distribution, significantly improving signal transmission rates and fiber optic resource utilization efficiency. Accurate detection of splitter port occupancy is crucial for ensuring efficient operation and maintenance of communication networks, optimized resource allocation, and rapid fault location.

[0019] However, current traditional methods for detecting optical splitter ports have significant limitations. Besides relying on inefficient manual on-site inspections, existing image recognition methods also have many shortcomings. The overall detection method requires complex multi-step image processing, resulting in a cumbersome and time-consuming process. This inefficiency makes it difficult to meet the real-time requirements of actual operation and maintenance, and limits its application scenarios. Other methods indirectly determine port status by identifying pigtails, but this method involves cumbersome steps, cannot be applied when pigtails obstruct the port, and has low accuracy in identifying port occupancy. These defects and limitations of existing methods severely restrict the efficiency and accuracy of optical splitter port detection, further impacting network operation and maintenance quality and user experience.

[0020] Based on the aforementioned technical problems, this application provides a method, apparatus, and program product for detecting beam splitter ports. The method includes: extracting features from a pre-trained port detection model's feature extraction module on a captured beam splitter port image to obtain a feature map. Then, determining the feature optimization weights of the feature map and performing multi-scale feature optimization to obtain optimized feature maps at multiple scales. Next, using a feature fusion module in the port detection model, fusing the multiple optimized feature maps to obtain a fused feature map. Following this, using a port detection module in the port detection model, performing port state detection and port position detection based on the fused feature map to determine the port state information and port position information of each beam splitter port in the beam splitter port image.

[0021] In the technical solution provided in this application embodiment, a pre-trained end-to-end port detection model can be used to perform high-precision and high-efficiency automated detection of the occupancy status and location of optical splitter ports, significantly improving detection speed and accuracy. Furthermore, the application scenarios are no longer limited by port arrangement or partial obstruction by pigtails, significantly enhancing the scope of application and flexibility. This further improves resource utilization efficiency and port allocation effectiveness in the optical splitter port allocation process, providing reliable technical support for network resource management and operation and maintenance quality.

[0022] Regarding the execution subject used in the technical solutions provided in the embodiments of this application, it can specifically be a terminal device used to acquire images of the splitter port, such as a mobile terminal, desktop computer, or laptop computer, or a remote device, such as a server capable of remote data connection. In addition, the execution subject used in the embodiments of this application can also be a software execution subject, such as a client or software program installed on a terminal device. The specific type of execution subject corresponding to the splitter port detection method, device, and program product provided in the embodiments of this application is not strictly limited here; it can be flexibly selected and set according to the application scenario and actual needs.

[0023] It should be noted that the embodiments provided in this application do not limit the specific application scenarios corresponding to the above-mentioned splitter port detection method, device and program products. The technical solutions provided in the embodiments of this application can be flexibly applied to any application scenario that requires analysis and determination of the splitter port occupancy according to actual needs.

[0024] For example, in real-world scenarios where the supervisor of the optical splitter needs to actively allocate network resources by adjusting the optical splitter interface, the technical solution provided in this application can use a pre-trained port detection model to perform sufficient and effective feature extraction, optimization, and multi-scale feature fusion on the images acquired by the optical splitter interface.

[0025] Furthermore, by employing a port detection model, precise port location and status detection can be performed on the fused feature map of the interface image. This determines the usage status and location of each port in the splitter port image, enabling regulatory personnel to make reasonable port allocation and resource scheduling based on the detection results. The technical solution provided by the embodiments of this application can effectively improve the detection accuracy and efficiency of splitter interface occupancy, providing a real and effective reference for the reasonable allocation of ports on the splitter and full utilization of resources, reducing resource waste and improving resource utilization efficiency.

[0026] It should be noted that the application scenarios described in the above embodiments of this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. Those skilled in the art will understand that with the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems. The splitter port detection method, apparatus, and program products provided by the embodiments of this application can be applied to any application scenario that requires analysis and determination of the occupancy status of splitter ports.

[0027] Figure 1 This is a flowchart illustrating a method for detecting a splitter port according to an embodiment of this application.

[0028] S101: Acquire images of the target beam splitter's ports. The beam splitter port images contain multiple beam splitter ports.

[0029] In step S101, the technical solution provided in this application embodiment can acquire a splitter port image containing multiple splitter ports on the target splitter through a preset image acquisition device.

[0030] In this embodiment, the specific type of image acquisition device performing the image collection process is not strictly limited. In some embodiments, an image can be actively acquired by a user using a mobile terminal with image acquisition capabilities (such as a mobile phone, tablet, camera, etc.) to capture images of the port portion of the beam splitter, thereby obtaining the beam splitter port image.

[0031] In other embodiments, high-definition surveillance cameras deployed at the location of the beam splitter can be used to capture images of the beam splitter's port portion in real time or at regular intervals. The specific image acquisition device can be flexibly selected according to the application scenario and actual needs.

[0032] In subsequent steps, the acquired beam splitter port images can be input into a pre-trained port detection model, which will then perform a series of processes such as feature extraction, optimization, and fusion to detect the state and position information of each beam splitter port in the beam splitter port image.

[0033] S102: Through the feature extraction module in the pre-trained port detection model, feature extraction is performed on the beam splitter port image to obtain a feature map, and the feature optimization weights of the feature map are determined. Based on the feature optimization weights, multi-scale feature optimization is performed on the feature map to obtain optimized feature maps at multiple scales corresponding to the beam splitter port image.

[0034] In step S102, the port detection model provided in this application embodiment can be used to extract features and detect ports in the beam splitter port image. The specific model structure can be composed of the feature extraction module in this step, the feature fusion module and the port detection module in subsequent steps.

[0035] Specifically, in the embodiments provided in this application, feature extraction can be performed on the beam splitter port image using a partial convolutional neural network in the feature extraction module to obtain a feature map. Then, global average pooling can be performed on the feature map to determine multiple channel feature data, which are then combined to obtain the channel feature map. Based on the channel feature map, global and local feature information of the feature map can be determined. Furthermore, feature optimization weights of the feature map are determined based on the global and local feature information. For the specific process of determining the feature optimization weights, please refer to [reference needed]. Figure 2 As shown in the image.

[0036] Figure 2 This is a schematic diagram illustrating the process of determining feature optimization weights according to an embodiment of this application.

[0037] like Figure 2 As shown, we can first target feature map 201 (i.e., the feature map containing global information in the beam splitter port image) Global average pooling is performed to determine the number of channel feature data corresponding to each channel. The specific determination process can be referred to the following formula (1): Formula (1) in, Representation of feature map In the The corresponding channel feature data on the channel, and Representing feature maps respectively Length and width, Represents the first in the feature graph The coordinates on the channel are The value, This represents the global average pooling function, and the feature map... , Represents the space of real numbers. This indicates the number of channels. Formula (1) can be used to accurately determine the channel feature data corresponding to each channel of the feature map. After combination, the channel feature map 202 (i.e.) can be obtained. ), channel feature map .

[0038] Then, in order to optimize the weights of local feature information in the feature map and minimize the number of model parameters, local information can be extracted from the channel feature map 202 by using the local parameter matrix of the strip matrix. For details, please refer to the following formula (2): Formula (2) in, Represents the channel feature map Extracted local feature information, For the strip matrix used to extract local information, the first... Each matrix element is specifically... , The number of local feature channels, Formula (2) can be used to accurately extract local feature information 203 from the channel feature map 202 (i.e., ), can be combined with global feature information 204 (i.e. Together, they are used to determine the feature optimization weights.

[0039] Similarly, to optimize the weights of local features in the feature map, global information can be extracted from the channel feature map 202 by using the global parameter matrix of the diagonal matrix. For details, please refer to formula (3): Formula (3) in, Represents the channel feature map Extracted global feature information, The first element in the diagonal matrix used to extract global information Each matrix element is specifically... , The number of local feature channels, Formula (3) can accurately extract global feature information 204 from channel feature map 202 (i.e., ).

[0040] Then, as Figure 2 As shown, global and local information can be combined based on local feature information 203 and global feature information 204 to obtain a comprehensive matrix 205 of the feature map and a transpose matrix 206 of the comprehensive matrix. The specific process can be found in formula (4): Formula (4) in, Indicates the combination of local feature information and global feature information The resulting composite matrix, in Figure 2 The middle part is the composite matrix 205, and the transpose matrix 206 is the transpose of the composite matrix 205.

[0041] Based on the comprehensive matrix, the global and local optimization weights of the feature map can be further determined, such as... Figure 2 In this process, single-channel features are extracted from the comprehensive matrix 205 and the transpose matrix 206 respectively, resulting in global optimization weights 207 and local optimization weights 208. For details, please refer to formulas (5) and (6): Formula (5) Formula (6) in, For the global optimization weights of the feature map, For local optimization weights of the feature map, This represents the number of channels corresponding to the feature map. The exact value can be determined using the above formulas (5) and (6). Figure 2 Global optimization weights 207 and local optimization weights 208 are applied to the feature map. Further optimization is then performed based on the weight parameters from the feature map extraction module. By combining the global optimization weight 207 and the local optimization weight 208, the feature optimization weight 209 of the feature map can be obtained (i.e., The specific determination process can be referenced in the following formula (7): Formula (7) in, The weight parameters in the feature map extraction module can be learned and optimized during model training. A predefined weight determination function is used to optimize weights globally. Local optimization weights and weight parameters Perform feature optimization weights The determination.

[0042] like Figure 2 As shown, the determined feature optimization weights It can be used for feature maps Feature optimization is performed, and the specific processing steps can be found in formula (8): Formula (8) in, To determine the feature optimization weights obtained from formula (7), This is a feature map extracted from the image of the beam splitter port. To optimize weights based on features For feature maps The optimized feature map obtained by feature optimization, i.e. Figure 2 As shown in Figure 210.

[0043] In the above embodiments, the process of determining the corresponding feature optimization weights for the feature map can be implemented through the Adaptive Fine-grained Channel Attention Mechanism (AFCAM). In other embodiments, other processing methods with the same function can also be used, and can be flexibly selected and used according to actual needs and application scenarios.

[0044] Based on the above embodiments, global and local feature information in the feature map can be used synergistically to adaptively generate feature optimization weights, enabling the port detection model to accurately focus on global and local details in the image. This effectively enhances the global and local feature representation capabilities in the optimized feature map, accurately strengthens practical features related to the port, and suppresses useless features unrelated to the port. This provides real and effective data for the subsequent port detection process, thereby significantly improving the detection accuracy and robustness of the beam splitter port status and position in different complex scenarios.

[0045] Regarding the specific optimization process of feature maps based on feature optimization weights, in the embodiments provided in this application, feature information in the beam splitter port image can be extracted from multiple scales through the feature extraction module in the port detection model, and the feature optimization weights and feature optimization process described in the above embodiments are determined for the feature map at each scale.

[0046] To facilitate understanding of the specific optimization process of multi-scale feature maps, the following explanation will use a specific example structure of the feature extraction module in the port detection model as an example. In some embodiments, the structural composition of the feature extraction module can be seen in Figures 3(a) and 3(b).

[0047] Figures 3(a) and 3(b) are schematic diagrams of the structure of a feature extraction module provided in one embodiment of this application.

[0048] Figure 3(a) is an example structure diagram of the feature extraction module, which includes an input module 301 for inputting the image of the beam splitter port, a first convolutional layer 302, a first batch of normalization layers 303, a max pooling layer 304, and three interconnected basic residual layers 305, 306, and 307 that apply the adaptive fine-grained channel attention mechanism in the above embodiment.

[0049] Figure 3(b) shows an example structure of a single basic residual layer, which includes a second convolutional layer 308 that receives the feature map input from the previous layer, a second batch normalization layer 309, a first rectified linear unit (ReLU) 310, a third convolutional layer 311, a third batch normalization layer 312, a second rectified linear unit 313, and an adaptive fine-grained channel attention mechanism layer 314 that applies the feature optimization weight determination process described above. Finally, it is combined with the input feature map and processed by the third rectified linear unit 315 before output.

[0050] After the beam splitter port image is input into the input module 301, it undergoes processing by the input module 301, the first convolutional layer 302, the first batch normalization layer 303, and the max pooling layer 304 to obtain the feature map of the beam splitter port image. Then, a first feature map with high resolution and low semantic features can be extracted through the first basic residual layer 305. The first feature map is an optimized feature map obtained after processing by the modules in Figure 3(b), especially after weight optimization by the adaptive fine-grained channel attention mechanism layer 313 in this embodiment.

[0051] Then, the first basic residual layer 305 outputs the first feature map from the feature extraction module, and can also input it into the next layer, that is, the second basic residual layer 306 takes the first feature map as the input feature map. Based on the processing of each module in Figure 3(b), a second feature map with semantic features at medium resolution is obtained, which is also an optimized feature map of the first feature map. The third basic residual layer 307 does the same, obtaining and outputting a third feature map with high semantic features at low resolution.

[0052] The above process enables multi-scale feature extraction from the feature map of the beam splitter port image. For each scale, the feature optimization weights described in the above embodiments are determined and the features are optimized, resulting in optimized feature maps at multiple scales corresponding to the beam splitter port image. The above embodiments and examples are for illustrative purposes only; the specific model structure and scale classification can be flexibly set according to actual needs and application scenarios.

[0053] The above embodiments establish an iterative optimization mechanism among multi-scale features of the feature maps of the beam splitter port image. Feature analysis is performed at different scales, effectively fusing and enhancing the detailed information of high-resolution features with the semantic information of low-resolution features. Global and local feature optimization can be performed on feature maps at different scales, providing practical data for subsequent port detection. This significantly improves the model's accuracy in capturing targets of different sizes and complex textures at the beam splitter port, thereby greatly enhancing the accuracy and robustness of the subsequent port detection process.

[0054] In addition, this application considers that the method of image acquisition at the beam splitter port can be flexibly selected, which may lead to differences in size or pixel value between different acquired images. Therefore, to further improve the accuracy and uniformity of the beam splitter port images, in the embodiments provided in this application, the beam splitter port images can be preprocessed before being input into the port detection model to unify the image size and pixel value representation, resulting in preprocessed port images.

[0055] Specifically, the image preprocessing process may include, but is not limited to: image size unification processing of the beam splitter port image, and image pixel value normalization processing of the beam splitter port image. Image size unification processing can be based on a preset fixed image size, adjusting the size of the beam splitter port image. To prevent image distortion, the original aspect ratio of the beam splitter port image must remain unchanged during size adjustment. If empty pixel value areas appear due to size adjustment, they can be automatically filled based on preset pixel values.

[0056] Image pixel value normalization can be performed by normalizing the pixel value of each pixel in the beam splitter port image. Specifically, the pixel value can be divided by 255 to adjust each pixel value to the range [0,1]. Then, mean and standard deviation normalization can be performed based on multiple beam splitter port images or a preset global pixel value mean and standard deviation. That is, the mean value is subtracted from the pixel value of each color channel of each pixel in the beam splitter port image and divided by the standard deviation to obtain the beam splitter port image with normalized pixel values.

[0057] The above preprocessing process can also be used in the pre-training process of the port detection model. Furthermore, the image preprocessing process can also include image enhancement processing during model training, such as horizontal rotation, vertical flipping, or random rotation of the beam splitter port image, thereby enriching the image sample data during training and improving the model's detection capability for different acquired images.

[0058] The above embodiments enable standardized processing of the input images for the port detection model, effectively unifying the pixel quality and size format of different acquired images. This effectively avoids the impact of image quality fluctuations caused by differences in shooting environment and equipment on the detection accuracy of subsequent port detection processes. It provides a stable and reliable image foundation for subsequent feature extraction and port detection, thereby improving the overall generalization ability and robustness of the detection model in complex real-world application scenarios.

[0059] S103: The feature fusion module in the port detection model performs feature fusion on multiple optimized feature maps to obtain the fused feature map corresponding to the beam splitter port image.

[0060] In step S103, the technical solution provided in this application embodiment can receive optimized feature maps at multiple scales output by the feature extraction model through the feature fusion model in the port detection model, fully fuse the interface features in the multiple optimized feature maps, and determine the fused feature map of the beam splitter port image.

[0061] Specifically, in the embodiments provided in this application, multiple optimized feature maps can be sequentially fused from top to bottom based on a scale from low to high to obtain a first fused feature map for each optimized feature map. Then, multiple first fused feature maps can be sequentially fused from bottom to top based on a scale from high to low to obtain a second fused feature map for each optimized feature map.

[0062] Each second fused feature map is then fused with its corresponding optimized feature map to obtain a third fused feature map for each optimized feature map. After fusing multiple third fused feature maps, a fused feature map of the beam splitter port image can be obtained.

[0063] Taking the extraction of the first feature map, the second feature map, and the third feature map at three scales based on the beam splitter port image in the above embodiment as an example, we will illustrate the point.

[0064] In this example, the third feature map is the optimized feature map of the highest semantic feature at the smallest scale. In the fusion process of this embodiment, the third feature map is first fused with the second feature map at a slightly smaller scale. The high semantic features in the third feature map are upsampled into the second feature map, and the resulting fused feature map is used as the first fused feature map of the second feature map.

[0065] Then, the first fused feature map of the second feature map can be fused with the first feature map with the smallest scale. The high semantic features of the first fused feature map of the second feature map are upsampled into the first feature map, and the resulting fused feature map is used as the first fused feature map of the first feature map. The third feature map can be used as the first fused feature map itself.

[0066] Next, based on the scale from high to low, multiple first fusion feature maps are sequentially fused from bottom to top. That is, the first fusion feature map of the first feature map and the first fusion feature map of the second feature map are fused by feature downsampling to obtain the second fusion feature map of the second feature map.

[0067] Similarly, the second fused feature map of the second feature map and the first fused feature map of the third feature map (i.e., the third feature map itself) are fused through feature downsampling to obtain the second fused feature map of the third feature map. The first feature map can use the first fused feature map as the second fused feature map.

[0068] The multiple second fusion feature maps determined through the above process can be further fused with the corresponding optimized feature maps to determine the third fusion feature map for each optimized feature map. In this example, the third fusion feature maps corresponding to the three feature maps at different scales can be further fused to obtain the fusion feature map of the beam splitter port image.

[0069] It should also be noted that before feature fusion, the number of channels in the feature images at different scales needs to be unified. For example, in the example above, the number of feature channels in the first and second feature images can be adjusted to match that of the third feature image. Figure 1 This ensures the smooth execution of the feature fusion process. Furthermore, the process of obtaining the second fused feature map can be implemented using a CNN-based Cross-scale Feature-fusion Module (CCFM). Other embodiments may also employ techniques or methods with the same functionality, and the choice can be flexible and tailored to specific needs and application scenarios.

[0070] The above embodiments enable dual-path feature fusion of multiple optimized feature maps, employing both top-down and bottom-up approaches. This approach fully preserves image contextual information while enhancing detail representation, reducing false positives and false negatives caused by feature scale variations in subsequent detection processes. The fused feature map obtained from these embodiments achieves comprehensive fusion of feature details at different scales, preserving high-level semantic information at low scales while also considering low-level details at high scales. This provides accurate and practical feature data for subsequent port detection processes, significantly improving detection accuracy.

[0071] Regarding the specific process of determining the third fusion feature map for each optimized feature map, in the embodiments provided in this application, the optimized feature map and the corresponding second fusion feature map can be processed by feature summation and feature difference in two ways to determine the feature sum data and feature difference data respectively.

[0072] Then, feature fusion is performed on the feature sum data and feature difference data to obtain the third fused feature map for each optimized feature image. For the specific fusion process, please refer to [reference needed]. Figure 4 As shown.

[0073] Figure 4 This is a schematic diagram illustrating the determination and fusion process of a third fusion feature map, provided as an embodiment of this application.

[0074] like Figure 4 As shown, for each optimized feature map, and These can be the optimized feature map 401 and the second fused feature map 402 of the optimized feature map, respectively.

[0075] The optimized feature map 401 and the second fused feature map 402 can be processed by multiple convolutional layers and modified linear units respectively. The two processed feature maps can be added element-by-element to determine the feature sum data 403, and element-by-element subtraction to determine the feature difference data 404. Then, the feature sum data 403 and the feature difference data 404 are processed and fused through convolutional layers and modified linear units to obtain the third fused feature map 405 of the optimized feature map 401.

[0076] In the above example, the summation branch can enhance the edge information in the optimized feature map and the second fused feature map, while the difference branch can enhance and generate different feature regions in the optimized feature map and the second fused feature map. Both branches can be constructed based on a densely connected layer with a weight-sharing mechanism. In some embodiments, this can be implemented using a densely connected feature fusion (DCFF) module; in other embodiments, other feasible technologies or methods with the same functionality can be used. The specific application can be flexibly selected according to actual needs and application scenarios.

[0077] In the above embodiments, by fusing the feature sum and feature difference between the optimized feature map and the second fused feature map, the common and differential information in the feature map is fully explored and integrated, which effectively enhances the model's representation effect on the subtle features and contextual relationships of the beam splitter port. In the subsequent port detection process, it can significantly improve the recognition accuracy and robustness of port status and position in complex scenarios.

[0078] In addition to the above, this application embodiment considers that further feature enhancement can be performed on the optimized feature map with low-scale high semantic features, thereby further improving the feature performance related to the beam splitter port in the optimized feature map, providing an optimized feature map with more prominent port semantic features for the subsequent fusion process, improving the feature fusion effect, and further improving the subsequent port detection accuracy.

[0079] Based on this, before performing multi-feature fusion on multiple optimized feature maps through the above embodiments, the technical solution provided in this application can also perform feature enhancement on the optimized feature map with the lowest scale and high semantic features.

[0080] Specifically, the port detection model may also include a feature enhancement module, which can receive the optimized feature map with low-scale high semantic features output by the feature extraction module, that is, the optimized feature map with the lowest resolution corresponding to the scale, and perform feature enhancement to obtain the enhanced feature map.

[0081] When performing feature fusion in the subsequent feature fusion module, feature fusion of the top-down and bottom-up paths in the above embodiments can be performed based on the enhanced feature map. High semantic features are fully preserved on the basis of the enhanced feature map, resulting in an integrated feature map with more obvious and prominent interface features.

[0082] This embodiment can be specifically implemented through Attention-based Intra-scale Feature Interaction (AIFI), where the optimized feature map with low-scale high semantic features is processed by an attention-based intra-scale feature interaction through a Transformer encoder.

[0083] Specifically, for example, optimized feature maps with high semantic features at low scales need to first adjust the number of channels through convolution, and then flatten them into one-dimensional vector data to adapt to the processing requirements of the Transformer encoder. Then, the feature vector is added to a preset positional encoding and input into the Transformer encoder for feature encoding. The feature data processed by the Transformer encoder needs to be restored to its original size to provide an appropriately sized enhanced feature map for subsequent feature fusion processes. Other feasible feature enhancement methods can also be used in other embodiments, and can be flexibly selected and applied according to actual needs and application scenarios.

[0084] The above embodiments enhance the optimized feature map, which has the lowest scale but the highest semantic features, effectively strengthening the deepest semantic information and global contextual features of the image. This allows the subsequent feature fusion process to fully utilize richer and more advanced abstract port features. Consequently, the port detection model's ability to understand and detect the overall structure of the beam splitter port and complex occlusion situations can be significantly improved during the detection process, thereby enhancing the accuracy of port status and position detection.

[0085] S104: Using the port detection module in the port detection model, the splitter port status and port position are detected based on the fused feature map, and the port status information and port position information corresponding to each splitter port in the splitter port image are determined.

[0086] In step S104, the technical solution provided in this application embodiment can use the port detection module in the port detection model to perform port status detection and port position detection on each port in the splitter port image based on the fusion feature map obtained in the above steps, thereby determining the port status information and port position information corresponding to each port.

[0087] The port status information indicates whether the port is occupied or idle. The port location information indicates the specific location of the port in the splitter port image. In some embodiments, the port location can be represented by adding a bounding box marker to each port in the splitter port image.

[0088] Specifically, in the embodiments provided in this application, the specific model structure of the port detection module can be referred to Figure 5 As shown in the image.

[0089] Figure 5 This is a schematic diagram of the structure of a port detection module provided in one embodiment of this application.

[0090] like Figure 5 As shown, in some embodiments, a Transformer decoder architecture can be used as the port detection module. The specific detection process includes: a multi-head self-attention mechanism 501 can be used to capture the relationships between elements in the input target query (the number of targets is K, which is the same as the number of ports in the beam splitter port image).

[0091] Then, a multi-head cross-attention mechanism 502 can be used to enable the decoder to dynamically focus on key regions in the cross-scale fused features, i.e., the fused feature map obtained in the above steps. The attention output is then further processed by a feedforward neural network 503, which generates optimized features through fully connected layers and nonlinear activation.

[0092] Each decoding layer is followed by a state prediction head 504 and a position prediction head 505 to accurately predict the usage status and location of each port in the beam splitter port image. Specifically, bounding box markers can be added to each port on the original beam splitter port image, allowing for a clear understanding of the usage status and location of each port based on the detection results. In addition, other feasible detection methods or technologies can be selected in other embodiments, allowing for flexible selection and application based on actual needs and applications.

[0093] Regarding the model training process corresponding to the port detection model in the above embodiments, in the embodiments provided in this application, the port detection model to be trained can be trained in advance based on sample data to improve the model's ability to identify and locate the state of the beam splitter port, so that in the practical application of the above embodiments, each port in the beam splitter port image can be accurately detected.

[0094] Specifically, multiple sample port images can be collected in advance for model training. State and position annotations are then performed on each beam splitter port in the sample port images to determine the standard port state and position information for each sample port.

[0095] In addition, in some embodiments, ports that are occupied can be used as positive samples, ports that are vacant can be used as negative samples, ports that have occlusion effects or other effects that increase the difficulty of detection can be used as difficult samples, and ports that are clear and unobstructed can be used as easy samples.

[0096] Then, a sample image preprocessing process similar to that in the above embodiments can be performed on the sample port image. Unlike the actual process image preprocessing, the sample image preprocessing, in addition to image size adjustment and pixel value normalization of the sample port image, can also perform image enhancement processing to further enrich the sample image.

[0097] Image enhancement processing may include, but is not limited to, horizontally or vertically flipping or randomly rotating sample port images to enhance the port detection model's adaptability and detection capability for beam splitter port images acquired from different directions during training. Adjusting the brightness, contrast, and saturation of sample port images can also enhance the port detection model's detection capability for beam splitter port images under different lighting conditions during training. Randomly cropping sample port images can further enhance the port detection model's ability to detect the same port in different positions during training.

[0098] Next, the preprocessed sample image can be input into the port detection model to be trained. The port detection model, through the processing steps of each module and the above embodiments, performs port state detection and port position detection on each sample port in the preprocessed sample image, determining the sample port state information and sample port position information. Specific processing procedures can be found in the above embodiments and will not be elaborated further here.

[0099] Based on the sample port state information and the standard port state information, the port classification loss of the port detection model during training can be determined. Simultaneously, based on the sample port position information and the standard port position information, the port localization loss, consisting of the port position loss and the port boundary localization loss, can be determined.

[0100] The specific process for determining the port classification loss can be referenced in the following formula (9): Formula (9) in, This represents the port classification loss of the port detection model. This represents the preset weights for positive and negative samples in the balanced sample port image. This refers to adjusting the preset focusing parameters for easy and difficult samples in the sample port image. The port state prediction probability is calculated based on the sample port state information and the standard port state information. Based on the processing calculation of formula (8), the port classification loss of the port detection model during training can be accurately determined, and the model's classification accuracy for port usage status can be optimized.

[0101] The specific process for determining the port location loss can be referenced in the following formula (10): Formula (10) in, This represents the port location loss within the port location loss. , , as well as These represent the x-axis coordinates, y-axis coordinates, bounding box width, and bounding box length of the standard bounding box corresponding to the port in the standard port location information, respectively. , , as well as These represent the x-axis coordinates, y-axis coordinates, bounding box width, and bounding box length of the sample bounding box corresponding to the port in the sample port location information, respectively. The number of samples represents the number of sample ports in the sample port image. Based on the processing calculation of formula (10), the port position information loss of the port detection model during training can be accurately determined, and the model's positioning accuracy for port positions and the accuracy of bounding box labeling can be optimized.

[0102] The specific process for determining the port boundary positioning loss can be referenced in the following formula (11): Formula (11) in, This represents the port boundary localization loss in the port localization loss, used to measure the quality of bounding box segmentation during the port detection model's training process. This represents the area of ​​the bounding box in the sample port location information. This represents the area of ​​the standard bounding box in the standard port location information. This represents the intersection-over-union ratio (IoU) between the standard bounding box and the bounding box applied at the port detection location during training. Based on the processing calculation of formula (11), the port boundary localization loss of the port detection model during training can be accurately determined, thereby optimizing the model's bounding box segmentation capability and accuracy.

[0103] Based on formulas (9)-(11), loss integration can be performed to determine the comprehensive model loss of the port detection model. For details, please refer to formula (12): Formula (12) in, The overall model loss of the port detection model is... , as well as These are the preset loss weights corresponding to the loss. Based on the comprehensive model loss determined by formula (12), the port detection model is trained in multiple rounds to optimize and improve the model parameters of each module in the above embodiments, and a port detection model that can be used to execute the above embodiments in actual applications is obtained.

[0104] The above training process effectively combines port classification loss and localization loss, which significantly improves the model's learning ability and generalization performance for the features of the beam splitter ports. This allows the model to maintain port detection accuracy while adapting to complex real-world scenarios, thereby significantly enhancing the accuracy and robustness of port state classification and position detection, and improving the precision and efficiency of the actual port detection process.

[0105] Furthermore, this application does not strictly limit the specific model type and model structure of the port detection model in the above embodiments. In some embodiments, it may be an improvement on the above embodiments based on the Transformer-based target detection model (DEtectionTransformer, DETR), or it may be a real-time Transformer detection model (Real-Time DEtection Transformer, RT-DETR), etc., which can be flexibly selected according to actual needs and application scenarios.

[0106] The RT-DETR model during training can solve the matching problem between predicted and ground truth bounding boxes using the Hungarian algorithm (Kuhn–Munkres Algorithm, KM). Specifically, the loss during model training can be modeled as a minimum-weighted matching bipartite graph problem, with the goal of finding a set of matches that minimizes the sum of the losses for all matching pairs. In this embodiment, the matching cost between each predicted and ground truth bounding box consists of classification error and bounding box error. After matching is completed, the loss function can be calculated for the matched predicted boxes using the above-described embodiment.

[0107] Based on the above embodiments, the splitter port detection method provided in this application is comprehensively explained using RT-DETR as the port detection model. For the specific port detection model structure, please refer to... Figure 6 As shown: Figure 6 This is a schematic diagram of the structure of a port detection model provided in one embodiment of this application.

[0108] like Figure 6 As shown, the port detection model includes a feature extraction module 601, a feature enhancement module 602, a feature fusion module 603, and a port detection module 604. The feature extraction module 601 may include the backbone network of the aforementioned AFCAM, extracting multi-scale optimized feature maps from the beam splitter port image 605, i.e., low-scale, high-semantic-feature features in the image. semantic features at the mesoscale and high-scale low-semantic features .

[0109] Then, low-scale high semantic features The input can be sent to the feature enhancement module 602 for feature enhancement processing as described in the above embodiments. Specifically, it can be selected... Figure 6 The AIFI implementation shown above yields an enhanced feature map. The enhanced feature map is then output to the feature fusion module 603. The feature fusion module 603 processes the enhanced feature map. semantic features at the mesoscale and high-scale low-semantic features The multi-feature fusion process described in the above embodiments can be performed by selecting, for example... Figure 6 The combination of CCFM and DCFF is shown to determine the fusion feature map 606 corresponding to the splitter port image 605.

[0110] Finally, the fused feature map 606 and the preset position code can be input together into the port detection module 604, and then... Figure 5 A similar model structure is used to detect the port status and port position of each port in the beam splitter port image 605, and outputs the image after the bounding box and port status are marked for each port in the beam splitter port image 605.

[0111] The above describes the specific implementation of the splitter port detection method provided in this application embodiment. In the technical solution provided in this application embodiment, a pre-trained end-to-end port detection model can be used to perform high-precision and high-efficiency automated detection of the splitter port occupancy status and location. This includes global and local feature weight optimization, and multiple fusion of feature maps at different scales, significantly improving detection speed and accuracy. Furthermore, the application scenarios of the technical solution provided in this application embodiment are no longer limited by port arrangement or partial occlusion by pigtails, significantly enhancing its applicability and flexibility. This further improves resource utilization efficiency and port allocation effectiveness in the splitter port allocation process, providing reliable technical support for network resource management and operation and maintenance quality.

[0112] Based on the same inventive concept, this application also provides a splitter port detection device, which can be found in the following embodiments: Figure 7 As shown.

[0113] Figure 7 This is a schematic diagram of a beam splitter interface detection device provided in another embodiment of this application.

[0114] like Figure 7 As shown, this application embodiment also provides a splitter port detection device 700, applied to electronic devices. The splitter port detection device 700 includes: Acquisition module 701 is used to acquire images of the splitter ports of the target splitter, and the images of the splitter ports contain multiple splitter ports. The extraction module 702 is used to extract features from the beam splitter port image through the feature extraction module in the pre-trained port detection model to obtain a feature map, and to determine the feature optimization weights of the feature map. Based on the feature optimization weights, the feature map is optimized at multiple scales to obtain optimized feature maps at multiple scales corresponding to the beam splitter port image. The fusion module 703 is used to perform feature fusion on multiple optimized feature maps through the feature fusion module in the port detection model to obtain the fused feature map corresponding to the beam splitter port image; The detection module 704 is used to perform splitter port status detection and splitter port position detection based on the fused feature map through the port detection module in the port detection model, and to determine the port status information and port position information corresponding to each splitter port in the splitter port image. The port status information indicates whether the splitter port is in an occupied state or an idle state.

[0115] In some embodiments, the acquisition module 701 described above is specifically used for: Image preprocessing is performed on the beam splitter port image to obtain the preprocessed port image. Image preprocessing includes at least image size unification and image pixel value normalization. Extraction module 702 is specifically used for: The feature extraction module extracts features from the preprocessed port image to obtain a feature map and determines the feature optimization weights of the feature map. Based on the feature optimization weights, multi-scale feature optimization is performed on the feature map to obtain optimized feature maps at multiple scales corresponding to the preprocessed port image.

[0116] In some embodiments, the extraction module 702 described above is specifically used for: Global average pooling is performed on the feature map to obtain multiple channel feature data corresponding to the number of channels in the feature map. The multiple channel feature data are combined to obtain the channel feature map of the feature map. Based on a preset global parameter matrix and channel feature map, the global feature information of the feature map is determined, and based on a preset local parameter matrix and channel feature map, the local feature information of the feature map is determined. Based on global feature information, the global optimization weights of the feature map are determined, and based on local feature information, the local optimization weights of the feature map are determined. The feature optimization weights are determined based on the global optimization weights and the local optimization weights.

[0117] In some embodiments, the extraction module 702 described above is specifically used for: For each scale, based on the feature optimization weights at that scale, feature optimization is performed on the feature map at that scale to determine the optimized feature map at that scale. The optimized feature map at this scale is used as the feature map at the adjacent scale, and the feature optimization weight of the optimized feature map at this scale is determined and used as the feature optimization weight of the adjacent scale. The feature map resolution of the adjacent scale is lower than that of this scale.

[0118] In some embodiments, the fusion module 703 is specifically used for: Multiple optimized feature maps are sequentially fused based on a scale from low to high to obtain the first fused feature map of each optimized feature map. Multiple first fused feature maps are sequentially fused based on a scale from high to low to obtain a second fused feature map for each optimized feature map. For each optimized feature map, feature fusion is performed between the feature map of the optimized feature map and the second fused feature map of the optimized feature map to obtain the third fused feature map of the optimized feature map; Multiple third-level fusion feature maps are fused to obtain a fusion feature map.

[0119] In some embodiments, the fusion module 703 is specifically used for: For each optimized feature map, determine the feature and value data of the feature map of the optimized feature map and the second fused feature map of the optimized feature map, and determine the feature difference data of the feature map of the optimized feature map and the second fused feature map of the optimized feature map; Feature fusion is performed on the feature sum data and feature difference data to determine the third fused feature map of the optimized feature map.

[0120] In some embodiments, the fusion module 703 is specifically used for: By using the feature enhancement module in the port detection model, feature enhancement is performed on the lowest resolution optimized feature map corresponding to the scale in the optimized feature map at multiple scales to obtain the enhanced feature map; The aforementioned fusion module 703 is specifically used for: The feature fusion module fuses the enhanced feature map and the optimized feature maps at other scales, except for the optimized feature map with the lowest scale resolution, to obtain a fused feature map.

[0121] In some embodiments, the above-described beam splitter port detection device 700 further includes a training module 705, which is used for: Acquire sample port images, and determine the standard port status information and standard port location information for each sample port in the sample port images; The sample port image is preprocessed to obtain a preprocessed sample image. The sample image preprocessing includes at least image size unification, image pixel value normalization, and image enhancement. Using the port detection model to be trained, the state and position of the beam splitter ports are detected in the preprocessed sample images to determine the corresponding sample port state and position information of each sample port in the preprocessed sample images. Based on the sample port status information and the standard port status information, the port classification loss is determined, and based on the sample port location information and the standard port location information, the port location loss is determined. The port location loss includes the port location loss and the port boundary location loss. The port detection model to be trained is trained based on port classification loss and port localization loss.

[0122] Figure 8 This is a schematic diagram of the structure of a terminal device provided in another embodiment of this application.

[0123] The terminal device may include a processor 801 and a memory 802 storing computer program instructions.

[0124] Specifically, the processor 801 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0125] Memory 802 may include mass storage for data or instructions. For example, and not limitingly, memory 802 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 802 may include removable or non-removable (or fixed) media. Where appropriate, memory 802 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 802 is non-volatile solid-state memory.

[0126] In a particular embodiment, memory 802 may include read-only memory (ROM), random access memory (RAM), disk storage media device, optical storage media device, flash memory device, electrical, optical, or other physical / tangible memory storage device. Thus, typically, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the splitter port detection method according to this disclosure.

[0127] The processor 801 reads and executes computer program instructions stored in the memory 802 to implement any of the optical splitter port detection methods in the above embodiments.

[0128] In one example, the terminal device may also include a communication interface 803 and a bus 810. Wherein, for example... Figure 8 As shown, the processor 801, memory 802, and communication interface 803 are connected through bus 810 and complete communication with each other.

[0129] The communication interface 803 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0130] Bus 810 includes hardware, software, or both, that couples components of an online data traffic metering device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 810 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, any suitable bus or interconnect is contemplated herein.

[0131] Furthermore, in conjunction with the splitter port detection method in the above embodiments, this application embodiment can provide a computer storage medium for implementation. The computer storage medium stores computer program instructions; when these computer program instructions are executed by a processor, they implement any of the splitter port detection methods in the above embodiments.

[0132] This application also provides a computer program product, including a computer program, which, when executed, implements any of the optical splitter port detection methods described in the above embodiments.

[0133] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0134] The functional blocks shown in the above block diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0135] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0136] The aspects of this disclosure have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by special-purpose hardware performing the specified functions or actions, or can be implemented by a combination of special-purpose hardware and computer instructions.

[0137] The above are merely specific embodiments of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. A method for detecting the port of a beam splitter, characterized in that, include: Acquire images of the beam splitter ports of the target beam splitter, wherein the beam splitter port images contain multiple beam splitter ports; The feature extraction module in the pre-trained port detection model extracts features from the beam splitter port image to obtain a feature map, and determines the feature optimization weights of the feature map. Based on the feature optimization weights, the feature map is optimized at multiple scales to obtain optimized feature maps at multiple scales corresponding to the beam splitter port image. The feature fusion module in the port detection model performs feature fusion on multiple optimized feature maps to obtain the fused feature map corresponding to the beam splitter port image. The port detection module in the port detection model performs splitter port status detection and splitter port position detection based on the fused feature map, and determines the port status information and port position information corresponding to each splitter port in the splitter port image. The port status information indicates whether the splitter port is in an occupied state or an idle state.

2. The method according to claim 1, characterized in that, Before extracting features from the beam splitter port image using the feature extraction module in the pre-trained port detection model to obtain a feature map, and determining the feature optimization weights of the feature map, and performing multi-scale feature optimization on the feature map based on the feature optimization weights to obtain optimized feature maps at multiple scales corresponding to the beam splitter port image, the method further includes: The image of the beam splitter port is preprocessed to obtain a preprocessed port image. The image preprocessing includes at least image size unification processing and image pixel value normalization processing. The feature extraction module in the pre-trained port detection model extracts features from the beam splitter port image to obtain a feature map, and determines the feature optimization weights of the feature map. Based on the feature optimization weights, multi-scale feature optimization is performed on the feature map to obtain optimized feature maps at multiple scales corresponding to the beam splitter port image, including: The feature extraction module extracts features from the preprocessed port image to obtain the feature map, and determines the feature optimization weights of the feature map. Based on the feature optimization weights, the feature map is optimized at multiple scales to obtain optimized feature maps at multiple scales corresponding to the preprocessed port image.

3. The method according to claim 1, characterized in that, Determining the feature optimization weights of the feature map includes: The feature map is subjected to global average pooling to obtain multiple channel feature data corresponding to the number of channels of the feature map. The multiple channel feature data are combined to obtain the channel feature map of the feature map. Based on a preset global parameter matrix and the channel feature map, the global feature information of the feature map is determined, and based on a preset local parameter matrix and the channel feature map, the local feature information of the feature map is determined. Based on the global feature information, the global optimization weights of the feature map are determined, and based on the local feature information, the local optimization weights of the feature map are determined. The feature optimization weights are determined based on the global optimization weights and the local optimization weights.

4. The method according to claim 1, characterized in that, Based on the aforementioned feature optimization weights, multi-scale feature optimization is performed on the feature map to obtain optimized feature maps at multiple scales corresponding to the beam splitter port image, including: For each scale, feature optimization is performed on the feature map at that scale based on the feature optimization weights at that scale, and the optimized feature map at that scale is determined. The optimized feature map at this scale is used as the feature map at the adjacent scale, and the feature optimization weight of the optimized feature map at this scale is determined as the feature optimization weight of the adjacent scale, wherein the resolution of the feature map at the adjacent scale is lower than that of the scale.

5. The method according to claim 4, characterized in that, The optimized feature maps are fused to obtain a fused feature map corresponding to the beam splitter port image, including: The multiple optimized feature maps are sequentially fused based on a scale from low to high to obtain a first fused feature map for each optimized feature map; The multiple first fused feature maps are sequentially fused based on a scale from high to low to obtain a second fused feature map for each of the optimized feature maps; For each optimized feature map, feature fusion is performed between the feature map of the optimized feature map and the second fused feature map of the optimized feature map to obtain the third fused feature map of the optimized feature map; The third fusion feature maps are fused to obtain the fusion feature map.

6. The method according to claim 5, characterized in that, For each optimized feature map, feature fusion is performed on the feature map of the optimized feature map and the second fused feature map of the optimized feature map to obtain the third fused feature map of the optimized feature map, including: For each optimized feature map, determine the feature and value data of the feature map of the optimized feature map and the second fused feature map of the optimized feature map, and determine the feature difference data of the feature map of the optimized feature map and the second fused feature map of the optimized feature map; The feature sum data and the feature difference data are fused to determine the third fused feature map of the optimized feature map.

7. The method according to claim 1, characterized in that, Before fusing multiple optimized feature maps through the feature fusion module in the port detection model to obtain the fused feature map corresponding to the beam splitter port image, the method further includes: The feature enhancement module in the port detection model enhances the lowest resolution optimized feature map at each of the multiple scales to obtain an enhanced feature map. The feature fusion module in the port detection model fuses multiple optimized feature maps to obtain a fused feature map corresponding to the beam splitter port image, including: The feature fusion module performs feature fusion on the enhanced feature map and optimized feature maps at other scales, except for the optimized feature map with the lowest resolution corresponding to the scale, to obtain the fused feature map.

8. The method according to claim 1, characterized in that, The training process of the port detection model includes: Acquire sample port images, and determine the standard port status information and standard port location information for each sample port in the sample port images; The sample port image is preprocessed to obtain a preprocessed sample image. The sample image preprocessing includes at least image size unification processing, image pixel value normalization processing, and image enhancement processing. Using the port detection model to be trained, the beam splitter port status and position are detected in the preprocessed sample image to determine the corresponding sample port status information and sample port position information for each sample port in the preprocessed sample image. Based on the sample port status information and the standard port status information, a port classification loss is determined, and based on the sample port location information and the standard port location information, a port positioning loss is determined, wherein the port positioning loss includes port location loss and port boundary positioning loss. The port detection model to be trained is trained based on the port classification loss and the port localization loss.

9. A beam splitter port detection device, characterized in that, include: The acquisition module is used to acquire images of the beam splitter ports of the target beam splitter, wherein the beam splitter port images contain multiple beam splitter ports; The extraction module is used to extract features from the beam splitter port image through the feature extraction module in the pre-trained port detection model to obtain a feature map, and to determine the feature optimization weights of the feature map. Based on the feature optimization weights, the feature map is optimized at multiple scales to obtain optimized feature maps at multiple scales corresponding to the beam splitter port image. The fusion module is used to perform feature fusion on multiple optimized feature maps through the feature fusion module in the port detection model to obtain the fused feature map corresponding to the beam splitter port image; The detection module is used to perform splitter port status detection and splitter port position detection based on the fused feature map through the port detection module in the port detection model, and to determine the port status information and port position information corresponding to each splitter port in the splitter port image. The port status information indicates whether the splitter port is in an occupied state or an idle state.

10. A computer program product, characterized in that, When the instructions in the computer program product are executed by the processor of the electronic device, the electronic device performs the splitter port detection method as described in any one of claims 1-8.

Images