A target identification method for a communication machine room, an electronic device, and a medium

Abstract

Embodiments of the present application disclose a target identification method for a communication machine room, an electronic device and a medium. The method comprises: collecting a target image inside the communication machine room, and inputting the target image into a backbone network in a target detection network for feature extraction to obtain multi-scale features; wherein the backbone network comprises a strip gating network and a feature extraction and fusion module; inputting the multi-scale features into a neck network for feature updating to obtain updated features; wherein the neck network comprises a feature extraction and fusion module to enhance the recovery ability of high-frequency details; inputting the updated features into a detection head for target identification to obtain an identification result of the target in the communication machine room. The above scheme can introduce a strip convolution and a directional spatial gating mechanism in the feature extraction stage, enhance the sensitivity of the model to the longitudinal structural features, and thus improve the accuracy and stability of the equipment boundary detection. The feature extraction and fusion module is used to enhance the recovery ability of high-frequency details.

Description

Technical Field

[0001] This application relates to the field of target recognition technology, and in particular to a target recognition method, electronic device and medium for a communication equipment room. Background Technology

[0002] Currently, asset management in telecommunications equipment rooms generally employs two technical solutions: traditional manual inspection and data entry, and automatic identification based on image recognition models. In the manual inspection approach, maintenance personnel periodically enter the equipment room to photograph, record, and verify each device, manually entering the device model, serial number, and location information into the asset management system. This method relies on manual completion of identification, data entry, and verification processes. While it offers some intuitiveness and operability, its overall efficiency is low, labor-intensive, and data updates are delayed. With the development of artificial intelligence technology, some research and engineering practices have begun to introduce deep learning-based image recognition methods for the automatic detection and classification of equipment in equipment rooms. These methods are typically based on object detection frameworks such as YOLOv5, YOLOv6, and YOLOv7, identifying device categories and locations by detecting device images. Some improved solutions employ data augmentation techniques (such as Mosaic and Cutout) and lightweight structure optimization to enhance recognition accuracy and inference speed.

[0003] However, in the complex application environment of communication equipment rooms (with uneven lighting, dense equipment, glare, and partial obstruction), the above technologies still have significant limitations. In particular, in areas such as small target detection, cross-scale feature fusion, multimodal information expansion, and mobile deployment, the performance of traditional deep learning detection models still cannot meet the needs of practical asset identification and management. Summary of the Invention

[0004] This application provides a target recognition method, electronic device, and medium for communication equipment rooms, which improves the network structure based on the characteristics of target images in communication equipment rooms during the target detection process, thereby improving the accuracy of target detection.

[0005] According to one aspect of this application, a target identification method for a communication equipment room is provided, the method comprising:

[0006] The target image inside the communication equipment room is acquired, and the target image is input into the backbone network of the target detection network for feature extraction to obtain multi-scale features; wherein, the backbone network includes a strip gating network and a feature extraction and fusion module;

[0007] The multi-scale features are input into the neck network for feature updating to obtain updated features; wherein, the neck network includes a cross-scale residual fusion module to capture multi-scale texture features;

[0008] The updated features are input into the detection head for target recognition, resulting in the recognition results of targets in the communication room.

[0009] According to one aspect of this application, a target identification device for a communication equipment room is provided, the device comprising:

[0010] The feature extraction module is used to acquire target images inside the communication equipment room and input the target images into the backbone network of the target detection network for feature extraction to obtain multi-scale features; wherein, the backbone network includes a strip gating network and a feature extraction and fusion module;

[0011] The feature update module is used to input the multi-scale features into the neck network for feature update to obtain updated features; wherein, the neck network includes a cross-scale residual fusion module to capture multi-scale texture features;

[0012] The target recognition module is used to input the updated features into the detection head for target recognition, and obtain the recognition result of the target in the communication room.

[0013] According to another aspect of this application, an electronic device is provided, the electronic device comprising:

[0014] At least one processor; and,

[0015] A memory that is communicatively connected to at least one processor; wherein,

[0016] The memory stores a computer program that can be executed by at least one processor, which enables the at least one processor to perform the target identification method for the communication room according to any embodiment of this application.

[0017] According to another aspect of this application, a computer-readable storage medium is provided, which stores computer instructions for causing a processor to execute and implement the target identification method for a communication room according to any embodiment of this application.

[0018] According to another aspect of this application, a computer program product is provided, which includes a computer program that, when executed by a processor, implements the target identification method for a communication room according to any embodiment of this application.

[0019] The technical solution of this application embodiment acquires target images inside a communication equipment room and inputs the target images into the backbone network of a target detection network for feature extraction to obtain multi-scale features. The backbone network includes a strip-gated network and a feature extraction and fusion module. The multi-scale features are then input into a neck network for feature updating to obtain updated features. The neck network includes a cross-scale residual fusion module to capture multi-scale texture features. The updated features are then input into a detection head for target recognition to obtain the recognition result of the target in the communication equipment room. The above solution introduces strip convolution and directional spatial gating mechanisms in the feature extraction stage to enhance the model's sensitivity to vertical structural features, thereby improving the accuracy and stability of equipment boundary detection. This structure has good adaptability to complex scenarios where equipment is densely arranged and structurally similar within a communication equipment room. The cross-scale residual fusion structure is designed to effectively superimpose and transfer features of different scales in the feature fusion stage, enabling the model to simultaneously detect both large and small targets, ensuring recognition stability in complex equipment room scenarios such as dense equipment, varying lighting, and partial occlusion.

[0020] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent from the following description. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 A flowchart illustrating a target identification method for a communication equipment room provided in this application embodiment;

[0023] Figure 2 A flowchart illustrating a target identification method for a communication equipment room, as provided in another embodiment of this application;

[0024] Figure 3 An overall block diagram provided for another embodiment of this application;

[0025] Figure 4 A schematic diagram of the SCN module and Strip CGLU module provided in another embodiment of this application;

[0026] Figure 5 A schematic diagram of the feature extraction and fusion module C2f_SFA provided in another embodiment of this application;

[0027] Figure 6 A flowchart illustrating a target identification method for a communication equipment room, as provided in another embodiment of this application;

[0028] Figure 7 A schematic diagram of a CSRF module provided in yet another embodiment of this application;

[0029] Figure 8 A schematic diagram of the structure of a target identification device for a communication equipment room provided in an embodiment of this application;

[0030] Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0033] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0034] It should be noted that the terms "first," "second," "third," "fourth," "actual," "preset," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0035] The acquisition, storage, use, and processing of data in this application comply with relevant national laws and regulations. The acquired data is obtained with authorization and will not be disclosed without permission, used for illegal purposes, purposes detrimental to the interests of others, or for personalized analysis or product promotion. It should be noted that certain software, components, models, and other existing industry solutions may be mentioned in the embodiments of this application. These should be considered exemplary and intended only to illustrate the feasibility of implementing the technical solution of this application, but do not imply that the applicant has already used or necessarily used the relevant content of such solutions.

[0036] Figure 1 This flowchart illustrates a target identification method for a communication equipment room, provided as an embodiment of this application. This embodiment is applicable to detecting targets within a communication equipment room. The method can be executed by a target identification device for the communication equipment room, which can be implemented in hardware and / or software and can be configured within an electronic device. Figure 1 As shown, the method includes:

[0037] S110. Acquire target images inside the communication equipment room and input the target images into the backbone network of the target detection network for feature extraction to obtain multi-scale features; wherein, the backbone network includes a strip gating network and a feature extraction and fusion module.

[0038] A communication equipment room is a dedicated space for installing communication equipment and supporting facilities, ensuring stable 24-hour operation. It is a core infrastructure of communication / power systems and is often referred to as the "nerve center" of a communication network. This application's embodiment focuses on target identification within the communication equipment room. The target can be an asset within the room, such as equipment. The target image is an image obtained through communication data acquisition within the communication equipment room. The target detection network can be a network with target detection capabilities, such as the YOLO series of detection algorithms. The target detection algorithm includes a backbone network, a neck network, and a detection head.

[0039] For example, target images inside a communication equipment room are acquired and input into the backbone network of a target detection network for feature extraction, yielding multi-scale features. In this embodiment, the backbone network is improved to be suitable for feature extraction of target images inside a communication equipment room. A strip gating network and a feature extraction and fusion module are designed in the backbone network. The strip gating network enhances the sensitivity to vertical structural features, improving the accuracy and stability of target edge detection. The feature extraction and fusion module is used to integrate features to obtain multi-scale features.

[0040] S120. The multi-scale features are input into the neck network for feature update to obtain updated features; wherein, the neck network includes a cross-scale residual fusion module to capture multi-scale texture features.

[0041] After extracting multi-scale features, these features are input into the neck network for feature updating, resulting in updated features. In this embodiment, the neck network is improved by including a cross-scale residual fusion module to capture multi-scale texture features. The cross-scale residual fusion module includes multi-scale feature extraction, parallel depthwise separable convolution, and residual feature fusion. For given input features from three different scales, fusion and updating are performed.

[0042] S130. Input the updated features into the detection head for target recognition to obtain the recognition result of the target in the communication room.

[0043] For example, the updated features output by the neck network are input into the detection head for target recognition, thereby determining the recognition result of the target in the communication room through the recognition by the detection head. The detection head is unchanged from the detection heads in the current target detection network and still uses the network structure of the detection heads in the current target detection network.

[0044] The technical solution of this application embodiment acquires target images inside a communication equipment room and inputs the target images into the backbone network of a target detection network for feature extraction to obtain multi-scale features. The backbone network includes a strip-gated network and a feature extraction and fusion module. The multi-scale features are then input into a neck network for feature updating to obtain updated features. The neck network includes a cross-scale residual fusion module to capture multi-scale texture features. The updated features are then input into a detection head for target recognition to obtain the recognition result of the target in the communication equipment room. The above solution introduces strip convolution and directional spatial gating mechanisms in the feature extraction stage to enhance the model's sensitivity to vertical structural features, thereby improving the accuracy and stability of equipment boundary detection. This structure has good adaptability to complex scenarios where equipment is densely arranged and structurally similar within a communication equipment room. The cross-scale residual fusion structure is designed to effectively superimpose and transfer features of different scales in the feature fusion stage, enabling the model to simultaneously detect both large and small targets, ensuring recognition stability in complex equipment room scenarios such as dense equipment, varying lighting, and partial occlusion.

[0045] Figure 2 This is a flowchart illustrating a target identification method for a communication equipment room, provided as another embodiment of this application. This embodiment is an optimization based on the above embodiment; schemes not described in detail in this embodiment are found in the above embodiment. Figure 2 As shown, the method in this embodiment of the application specifically includes the following steps:

[0046] S210, Acquire target images inside the communication equipment room.

[0047] S220. The target image is input into the composite convolution module for processing, and then processed through two layers of strip gating network to obtain the first scale feature.

[0048] For example, such as Figure 3 As shown, in the backbone network of the object detection network, the target image undergoes feature extraction via the backbone network. First, it is processed by the Composite Convolutional Branch (CBS) module, and then by a two-layer strip-gated network (SCN) to obtain the first-scale features. The SCN module consists of strip attention, directional spatial gating, and convolutional gating linear units. The strip attention module aims to model directional spatial dependencies and suppress irrelevant noise, inputting the feature map of the target image. Through two convolution Nonlinear transformation with the GELU activation function enhances expressive power; directional spatial gating is employed. Depthwise convolution aggregates local contextual information, and then sequentially applies kernels with a kernel size of [size missing]. and The strip convolutions capture the horizontal and vertical structural features of the device, respectively, and then... Convolution refines the attention weights, resulting in the final attention map. Element-wise multiplication with the input enhances the response of the target region: Residual connections preserve the original features and avoid information loss; convolutional gated linear units dynamically select effective features through gating mechanisms, including channel expansion, deep feature interaction, channel compression, and residual learning.

[0049] In this embodiment of the application, the processing is carried out through two layers of strip gating networks, including:

[0050] The target image is input into the output features of the composite convolution module and used as the input features of the strip gating network. The input features are then passed through two deep convolutional layers and an activation function for nonlinear transformation.

[0051] The input features after nonlinear transformation are processed by one layer of strip convolution gated linear module, two layers of strip convolution gated linear module, and three layers of strip convolution gated linear module, and then fused with the input features after depth convolution.

[0052] For example, such as Figure 4As shown in the SCN module, in each layer of the strip-gated network, the target image is input to the output features processed by the composite convolution module, which serve as the input features of the strip-gated network. These input features are then processed through a deep convolutional layer (Conv). The convolutionally processed input features are then processed by one, two, and three strip-gated linear convolutional modules, respectively, before being fused with the input features processed by the deep convolution module. This fusion, achieved through a composite convolution module, yields the output features.

[0053] In this embodiment of the application, the processing procedure after passing through the strip convolution gated linear module includes:

[0054] The input features that need to be input into the strip convolutional gated linear module are nonlinearly transformed by two convolutional layers and activation functions to obtain the transformed features;

[0055] For the aforementioned transformation features, deep convolution is used to aggregate local contextual information, and then two strip convolutions with different kernel sizes are used in sequence to capture horizontal and vertical structural features respectively. The attention weights are then refined through convolution to generate an attention map.

[0056] The attention map is multiplied element-wise with the transformed features, and then fused with the input features to determine the output features.

[0057] For example, such as Figure 4 As shown in the Strip CGLU module, the input features from the strip convolutional gated linear module are nonlinearly transformed through two convolutional layers and activation functions to obtain transformed features. For these transformed features, depthwise convolution is used to aggregate local contextual information. Then, two strip convolutions with different kernel sizes are used sequentially to capture horizontal and vertical structural features respectively. Attention weights are then refined through convolution to generate an attention map. The attention map is multiplied element-wise with the transformed features and then fused with the input features to determine the output features.

[0058] S230. The first-scale features are further processed through two layers of composite convolution modules and feature extraction and fusion modules to obtain the second-scale features.

[0059] For example, such as Figure 3 As shown, the first-scale features are further processed through two layers of composite convolution modules and feature extraction and fusion modules, which means they are processed through two layers of CBS modules + C2f_SFA modules to obtain the second-scale features.

[0060] In this embodiment, the first-scale features are further processed through two layers of a composite convolution module and a feature extraction and fusion module to obtain second-scale features, including:

[0061] During the process of the first-scale feature being processed by each layer of the composite convolution module and the feature extraction and fusion module, the first-scale feature, after being processed by the composite convolution module, can be used as the input feature of the feature extraction and fusion module.

[0062] The input features are decomposed into high-frequency and low-frequency components by discrete cosine transform, and then the inverse transform is performed to obtain the reconstructed features.

[0063] Attention values ​​are determined by windowed multi-head attention and frequency-guided analysis of the reconstructed features.

[0064] The first weight is generated based on the high-frequency component, and the second weight is generated based on the high-frequency component and the low-frequency component.

[0065] The second-scale feature is obtained by gating and weighting the spatial frequency attention branch and the frequency projection branch according to the first and second weights.

[0066] For example, the features at the first scale are processed by the compound convolution module (CBS) and used as input features for the feature extraction and fusion module (C2f_SFA). The feature extraction and fusion module C2f_SFA is as follows: Figure 5 As shown, the input features are decomposed into high-frequency and low-frequency components by discrete cosine transform, and then inversely transformed to obtain reconstructed features. Attention values ​​are determined by windowed multi-head attention and frequency guidance in the reconstructed features. A first weight is generated based on the high-frequency components, and a second weight is generated based on the high-frequency and low-frequency components. The spatial frequency attention branch and the frequency projection branch are gated and weighted and fused according to the first and second weights to obtain the second scale features.

[0067] Specifically, input features The components are decomposed into high-frequency and low-frequency components using Discrete Cosine Transform (DCT):

[0068] ;

[0069] In the formula, and As a frequency band mask, high and low frequencies are enhanced separately, and features are reconstructed through inverse transform:

[0070] ;

[0071] Reconstructed features As input to the windowed multi-head self-attention module, frequency information is used to analyze the query. ,key Sum The feature representation is constrained and enhanced to calculate the attention weights. Then, the calculated attention weights are applied to the spatial frequency attention branch to complete the projection and enhancement of spatial features;

[0072] The spatial-frequency attention mechanism achieves dual-domain collaboration through windowed multi-head attention and frequency guidance:

[0073] ;

[0074] in These represent the query, key, and value, respectively. It is the dimension of the key. It is the scaling factor;

[0075] The design incorporates a dual-gated channel-space system with adaptive fusion features; the channel gating is based on high-frequency energy generation weights. The expression is:

[0076] ;

[0077] in This represents the Sigmoid activation function, MLP stands for Multilayer Perceptron, which is used to learn the nonlinear relationship between channels, and GAP stands for Global Average Pooling.

[0078] Spatial gating combined with high and low frequency features to generate spatial masks The expression is:

[0079] ;

[0080] use and For containing The spatial frequency attention branch is adaptively gated and fused with the frequency projection branch to obtain the final output features:

[0081] .

[0082] S240. The second-scale features are processed by a fast spatial pyramid pooling layer to obtain the third-scale features.

[0083] For example, such as Figure 3 As shown, the second-scale features are processed by the Fast Spatial Pyramid Pooling (SPPF) layer to obtain the third-scale features.

[0084] S250. The multi-scale features are input into the neck network for feature update to obtain updated features; wherein, the neck network includes a cross-scale residual fusion module to capture multi-scale texture features.

[0085] S260. Input the updated features into the detection head for target recognition to obtain the recognition result of the target in the communication room.

[0086] The scheme in this embodiment inputs the target image into a composite convolutional module for processing, and then processes it through two layers of striped gating networks to obtain first-scale features. The first-scale features are further processed through two layers of composite convolutional modules and feature extraction and fusion modules to obtain second-scale features. The second-scale features are then processed through a fast spatial pyramid pooling layer to obtain third-scale features. Striped convolution and directional spatial gating mechanisms are introduced in the feature extraction stage to enhance the model's sensitivity to vertical structural features (such as cabinet edges, vertical cables, etc.), thereby improving the accuracy and stability of equipment boundary detection. This structure has good adaptability to complex scenarios where communication equipment is densely arranged and structurally similar within a communication equipment room. A C2f_SFA dual-domain attention mechanism combining spatial and frequency domain feature extraction is proposed to focus on enhancing high-frequency details (such as nameplates, labels, and port text), solving the problem of missed detections and false detections that traditional convolutional networks easily encounter in the recognition of small targets and text regions, and significantly improving the accuracy and robustness of nameplate recognition.

[0087] Figure 6 This is a flowchart illustrating a target identification method for a communication equipment room, provided as another embodiment of this application. This embodiment is an optimization based on the above embodiments; schemes not described in detail in this embodiment are found in the above embodiments. Figure 6 As shown, the method in this embodiment of the application specifically includes the following steps:

[0088] S310. Acquire target images inside the communication equipment room and input the target images into the backbone network of the target detection network for feature extraction to obtain multi-scale features; wherein, the backbone network includes a strip gating network and a feature extraction and fusion module.

[0089] S320. The multi-scale features are processed by the cross-scale residual fusion module to obtain intermediate features.

[0090] For example, during feature updating, multi-scale features are processed by a cross-scale residual fusion module to obtain intermediate features, such as... Figure 3 As shown, the multi-scale features are processed by the cross-scale residual fusion module CSRF to obtain intermediate features.

[0091] In this embodiment of the application, the multi-scale features are processed by a cross-scale residual fusion module to obtain intermediate features, including:

[0092] The first scale features are processed by bilinear upsampling and convolution to obtain the first processed features;

[0093] The second-scale feature is obtained by performing an identity mapping on the second-scale feature and adjusting the number of channels;

[0094] The third-scale features are adaptively downsampled and then subjected to average pooling, max pooling, and convolution to obtain the third-processed features.

[0095] The first, second, and third processed features are concatenated along the channel dimension and then processed using parallel depthwise separable convolutions with different kernel sizes to obtain multi-scale contextual features.

[0096] The intermediate feature is obtained by fusing the multi-scale contextual features with the concatenated first, second, and third processing features.

[0097] For example, for the first-scale features, bilinear upsampling and convolution are applied to obtain the first processed features. The second-scale features are then subjected to identity mapping and channel adjustment to obtain the second processed features. The third-scale features are then adaptively downsampled and subjected to average pooling, max pooling, and convolution to obtain the third processed features. The first, second, and third processed features are concatenated by channel dimension and processed using parallel depthwise separable convolutions with different kernel sizes to obtain the scale context features. Finally, the multi-scale context features are fused with the concatenated first, second, and third processed features to obtain intermediate features.

[0098] Specifically, such as Figure 7 As shown, given input features from three different scales... x1 represents the first-scale feature, x2 represents the second-scale feature, and x3 represents the third-scale feature. The cross-scale residual fusion module CSRF first processes each branch as follows:

[0099] Upsampling branch recovers device details through bilinear upsampling and Convolution aligns channel dimensions to obtain the first processed feature. :

[0100] ;

[0101] The identity mapping branch preserves spatial resolution and adjusts the number of channels to obtain the second processed feature. :

[0102] ;

[0103] In the formula, Indicates the expansion factor;

[0104] The downsampling branch captures the global contour of large ships, and uses the ADown module to combine average pooling, max pooling, and convolution to achieve hierarchical downsampling and obtain third-scale features. :

[0105] ;

[0106] The processed multi-scale features are then concatenated along the channel dimension:

[0107] ;

[0108] Subsequently, different core sizes were applied. Parallel depthwise separation can smooth out speckle noise from convolutions to capture multi-scale contextual information:

[0109] ;

[0110] By summing and aggregating the outputs of parallel convolutions, and utilizing... Point convolution for channel fusion:

[0111] ;

[0112] Finally, the integrity of the original features is preserved through residual connections:

[0113] .

[0114] S330. The intermediate features are fused with the first-scale features and the third-scale features respectively to obtain the updated features.

[0115] S340. Input the updated features into the detection head for target recognition to obtain the recognition result of the target in the communication room.

[0116] In this embodiment of the application, the process of determining the loss function during training includes:

[0117] The high-quality detection box loss is determined based on the real-time detection box position, the target box position, and the height and width of the minimum bounding box.

[0118] The first loss function is determined based on the high-quality detection box loss and the cross-union ratio loss; or...

[0119] The second loss function is determined by multiplying the first loss function by the monotonic focusing coefficient; or,

[0120] A third loss function is determined based on the first loss function and the outlier; the outlier is used to describe the quality of the detection box.

[0121] The DIoU and CIoU loss functions integrated in YOLOv8 both have specific limitations: the DIoU loss function introduces the normalized distance between center points as a penalty term and adds a negative gradient to the minimum bounding box size, which may inversely cause the predicted box size to expand, weakening the overlap between boxes; although CIoU integrates aspect ratio consistency constraints on the basis of DIoU, its adaptability to low-quality labeled data and small-scale targets is still insufficient. To address the above problems, this invention introduces the Wise-IoU loss function (WIoU) based on dynamic gradient gain, which optimizes the processing effect of multi-scale targets in the target detection task, especially improving the detection performance of small targets. There are three versions of the WIoU loss function. WIoU-v1 constructs distance attention based on distance metric, applying a two-layer attention mechanism to itself, so that the model has better generalization ability. Its formula is shown below:

[0122] ;

[0123] ;

[0124] In the formula, This represents the intersection-union loss, which measures the degree of overlap between the predicted bounding box and the ground truth bounding box. The smaller the value, the higher the degree of overlap between the predicted and ground truth bounding boxes, and the higher the localization accuracy. Indicates the prediction box. Represents the true bounding box. This represents the intersection area between the predicted bounding box and the ground truth bounding box; This represents the union region of the predicted bounding box and the ground truth bounding box; Indicates the area of ​​the corresponding region;

[0125] ;

[0126] In the formula, Indicates high-quality detection frame loss. Represents an exponential function. This indicates the position of the real-time detection box corresponds to the position of the target box. , , These are the height and width of the minimum bounding box, respectively. Represented as a separation operation, this effectively eliminates factors hindering convergence; constructed in WIoU-v2 monotonic focusing coefficient This is used to adjust the gradient gain corresponding to different quality bounding boxes, so that high-quality bounding boxes receive greater optimization weights. Under the condition that the monotonic focusing coefficient is satisfied, the monotonic focusing coefficient is combined with Multiplying them together yields the WIoU-v2 loss, as shown in the following formula:

[0127] ;

[0128] However, during model training, as the IoU decreases, the gradient gain... This also reduces the convergence speed in the later stages of training; therefore, the mean of IoU is introduced. This problem can be solved by using a normalization factor, as shown in the following equation:

[0129] ;

[0130] In WIoU-v3, the concept of outlier is defined to describe the quality of the detection box. The higher the outlier, the more significantly the predicted box deviates from the average quality. The definition is as follows:

[0131] ;

[0132] Therefore, the WIoU-v3 loss function can be defined as follows:

[0133] ;

[0134] in, Represents the dynamic gradient gain factor. The adjustment parameter is used to control the rate of gradient change. The hyperparameters controlling the location of gradient peaks are used to determine the gradient assignment strategy corresponding to different outlier degrees. hour, Make When the outlier level of the detection box meets the following conditions ( When the value is constant, the detection box will obtain the highest gradient gain, because It is dynamic, and the quality division criteria of the detection box are also dynamically adjusted, which enables WIoU-v3 to adopt the gradient gain allocation strategy that is most suitable for the current situation at each moment.

[0135] The solution in this embodiment processes the multi-scale features through a cross-scale residual fusion module to obtain intermediate features; the intermediate features are then fused with the first-scale features and the third-scale features to obtain the updated features. A cross-scale residual fusion structure is designed to effectively superimpose and transfer features of different scales during the feature fusion stage, enabling the model to simultaneously detect both large and small targets, ensuring recognition stability in complex data center scenarios such as dense equipment, varying lighting, and partial occlusion.

[0136] This application provides a specific implementation method, including:

[0137] Collect images of communication equipment in the equipment room and establish an image dataset of equipment in the equipment room containing multiple types of equipment;

[0138] The collected images were labeled using the labeling tool LabelImg to generate device type bounding boxes and model labels, and then divided into training set, validation set and prediction set.

[0139] Perform data enhancement operations on the acquired device images, such as random cropping, rotation, noise addition, and brightness adjustment.

[0140] In the model building phase, a strip gated network (SCN), a feature extraction and fusion module (C2f_SFA module), and a CSRF module were designed, and the WIoU loss function was introduced to improve the YOLOv8 network, forming the SGCF-YOLO network architecture.

[0141] By combining the Qianwen VL multimodal large model with the SGCF-YOLO model, collaborative recognition of communication equipment appearance inspection and nameplate information extraction can be achieved.

[0142] The model is trained using the training set and its performance is validated using the validation set. After training, the model is used to identify data center equipment in the prediction set.

[0143] The model is deployed on mobile and PC platforms to achieve cross-platform real-time identification and intelligent management of communication equipment room assets.

[0144] Overall block diagram as follows Figure 3 As shown.

[0145] The SCN module consists of strip attention, directional spatial gating, and convolutional gating linear units, such as... Figure 4 As shown, the strip attention module aims to model directional spatial dependencies and suppress irrelevant noise, with the input feature map... Through two convolution Nonlinear transformation with the GELU activation function enhances expressive power; directional spatial gating is employed. Depthwise convolution aggregates local contextual information, and then sequentially applies kernels with a kernel size of [size missing]. and The strip convolutions capture the horizontal and vertical structural features of the device, respectively, and then... Convolution refines the attention weights, resulting in the final attention map. Element-wise multiplication with the input enhances the response of the target region: Residual connections preserve the original features and avoid information loss; convolutional gated linear units dynamically select effective features through gating mechanisms, including channel expansion, deep feature interaction, channel compression, and residual learning.

[0146] To enhance the focus on high-frequency components, the C2f-SFA structure proposed in this application combines a Spatial Frequency Attention (SFA) module with a C2f module to improve the recovery of high-frequency details, such as... Figure 5 As shown, its core design combines a spatial-frequency dual-domain attention mechanism with a dynamic gating fusion strategy, and its mathematical expression is as follows:

[0147] ;

[0148] In the formula, For spatial frequency attention branch, For frequency projection branch, This indicates a feature fusion operation. For adaptive gating weights;

[0149] Input features The components are decomposed into high-frequency and low-frequency components using Discrete Cosine Transform (DCT):

[0150] ;

[0151] In the formula, and As a frequency band mask, high and low frequencies are enhanced separately, and features are reconstructed through inverse transform:

[0152] ;

[0153] The spatial-frequency attention mechanism achieves dual-domain collaboration through windowed multi-head attention and frequency guidance:

[0154] ;

[0155] in These represent the query, key, and value, respectively. It is the dimension of the key. It is the scaling factor;

[0156] The design incorporates a dual-gated channel-space system with adaptive fusion features; the channel gating is based on high-frequency energy generation weights. The expression is:

[0157] ;

[0158] Spatial gating combined with high and low frequency features to generate spatial masks The expression is:

[0159] ;

[0160] Finally, gated weighted fusion is performed:

[0161] ;

[0162] To address the challenge of large target scale variations, embodiments of this application propose a CSRF module to capture multi-scale texture features, such as... Figure 7 As shown, this module comprises three core components: multi-scale feature extraction, parallel depthwise separable convolution, and residual feature fusion, given input features from three different scales. CSRF first processes each branch as follows:

[0163] Upsampling branch recovers device details through bilinear upsampling and Convolutional alignment channel dimensions are obtained :

[0164] ;

[0165] The identity mapping branch preserves spatial resolution and adjusts the number of channels to obtain :

[0166] ;

[0167] In the formula, Indicates the expansion factor;

[0168] The downsampling branch captures the global contours of large ships, using the ADown module to combine average pooling, max pooling, and convolution to achieve layered downsampling:

[0169] ;

[0170] The processed multi-scale features are then concatenated along the channel dimension:

[0171] ;

[0172] Subsequently, different core sizes were applied. Parallel depthwise separation can smooth out speckle noise from convolutions to capture multi-scale contextual information:

[0173] ;

[0174] By summing and aggregating the outputs of parallel convolutions, and utilizing... Point convolution for channel fusion:

[0175] ;

[0176] Finally, the integrity of the original features is preserved through residual connections:

[0177] ;

[0178] The DIoU and CIoU loss functions integrated in YOLOv8 both have specific limitations: the DIoU loss function introduces the normalized distance between center points as a penalty term and adds a negative gradient to the minimum bounding box size, which may inversely cause the predicted box size to expand, weakening the overlap between boxes; although CIoU integrates aspect ratio consistency constraints on the basis of DIoU, its adaptability to low-quality labeled data and small-scale targets is still insufficient. To address the above problems, this invention introduces the Wise-IoU loss function (WIoU) based on dynamic gradient gain, which optimizes the processing effect of multi-scale targets in the target detection task, especially improving the detection performance of small targets. There are three versions of the WIoU loss function. WIoU-v1 constructs distance attention based on distance metric, applying a two-layer attention mechanism to itself, so that the model has better generalization ability. Its formula is shown below:

[0179] ;

[0180] ;

[0181] In the formula, Indicates high-quality detection frame loss. This indicates the position of the real-time detection box corresponds to the position of the target box. , , These are the height and width of the minimum bounding box, respectively. Represented as a separation operation, this effectively eliminates factors hindering convergence; constructed in WIoU-v2 monotonic focusing coefficient ,exist Under the condition that the monotonic focusing coefficient is satisfied, the monotonic focusing coefficient is combined with Multiplying them together yields the WIoU-v2 loss, as shown in the following formula:

[0182] ;

[0183] However, during model training, as the IoU decreases, the gradient gain... This also reduces the convergence speed in the later stages of training; therefore, the mean of IoU is introduced. This problem can be solved by using a normalization factor, as shown in the following equation:

[0184] ;

[0185] In WIoU-v3, the concept of outlier is defined to describe the quality of the detection box, as shown in the following equation:

[0186] ;

[0187] Therefore, the WIoU-v3 loss function can be defined as follows:

[0188] ;

[0189] Among them, when hour, Make When the outlier level of the detection box meets the following conditions ( When the value is constant, the detection box will obtain the highest gradient gain, because It is dynamic, and the quality division criteria of the detection box are also dynamically adjusted, which enables WIoU-v3 to adopt the gradient gain allocation strategy that is most suitable for the current situation at each moment.

[0190] Figure 8 This is a schematic diagram of a target identification device for a communication equipment room provided in an embodiment of this application. This device can execute the target identification method for a communication equipment room provided in any embodiment of this application, and has the corresponding functional modules and beneficial effects for executing the method. Figure 8 As shown, the device includes:

[0191] The feature extraction module 410 is used to acquire target images inside the communication equipment room and input the target images into the backbone network of the target detection network for feature extraction to obtain multi-scale features; wherein, the backbone network includes a strip gating network and a feature extraction and fusion module;

[0192] The feature update module 420 is used to input the multi-scale features into the neck network for feature update to obtain updated features; wherein, the neck network includes a cross-scale residual fusion module to capture multi-scale texture features;

[0193] The target recognition module 430 is used to input the updated features into the detection head for target recognition, and obtain the recognition result of the target in the communication room.

[0194] In this embodiment, the feature extraction module 410 inputs the target image into the backbone network of the target detection network for feature extraction to obtain multi-scale features, including:

[0195] The target image is input into a composite convolution module for processing, and then processed through two layers of strip gating network to obtain the first scale features;

[0196] The first-scale features are further processed through two layers of composite convolution modules and feature extraction and fusion modules to obtain the second-scale features;

[0197] The second-scale features are processed through a fast spatial pyramid pooling layer to obtain the third-scale features.

[0198] In this embodiment, the feature extraction module 410 processes data through a two-layer striped gating network, including:

[0199] The target image is input into the output features of the composite convolution module and then used as the input features of the strip gating network. The input features are then processed through a deep convolutional layer.

[0200] The input features after convolution are processed by one layer of strip convolution gated linear module, two layers of strip convolution gated linear module, and three layers of strip convolution gated linear module, and then fused with the input features after depth convolution.

[0201] In this embodiment of the application, the feature extraction module 410 undergoes processing by a strip convolution gated linear module, including:

[0202] The input features that need to be input into the strip convolutional gated linear module are nonlinearly transformed by two convolutional layers and activation functions to obtain the transformed features;

[0203] For the aforementioned transformation features, deep convolution is used to aggregate local contextual information, and then two strip convolutions with different kernel sizes are used in sequence to capture horizontal and vertical structural features respectively. The attention weights are then refined through convolution to generate an attention map.

[0204] The attention map is multiplied element-wise with the transformed features, and then fused with the input features to determine the output features.

[0205] In this embodiment, the feature extraction module 410 further processes the first-scale features through two layers of processing by a composite convolution module and a feature extraction and fusion module to obtain second-scale features, including:

[0206] During the process of the first-scale feature being processed by each layer of the composite convolution module and the feature extraction and fusion module, the first-scale feature, after being processed by the composite convolution module, can be used as the input feature of the feature extraction and fusion module.

[0207] The input features are decomposed into high-frequency and low-frequency components by discrete cosine transform, and then the inverse transform is performed to obtain the reconstructed features.

[0208] Attention values ​​are determined by windowed multi-head attention and frequency-guided analysis of the reconstructed features.

[0209] The first weight is generated based on the high-frequency component, and the second weight is generated based on the high-frequency component and the low-frequency component.

[0210] The second-scale feature is obtained by gating and weighting the spatial frequency attention branch and the frequency projection branch according to the first and second weights.

[0211] In this embodiment, the feature update module 420 inputs the multi-scale features into the neck network for feature update to obtain updated features, including:

[0212] The multi-scale features are processed by a cross-scale residual fusion module to obtain intermediate features;

[0213] The intermediate features are fused with the first-scale features and the third-scale features respectively to obtain the updated features.

[0214] In this embodiment, the feature update module 420 processes the multi-scale features through a cross-scale residual fusion module to obtain intermediate features, including:

[0215] The first scale features are processed by bilinear upsampling and convolution to obtain the first processed features;

[0216] The second-scale feature is obtained by performing an identity mapping on the second-scale feature and adjusting the number of channels;

[0217] The third-scale features are adaptively downsampled and then subjected to average pooling, max pooling, and convolution to obtain the third-processed features.

[0218] The first, second, and third processed features are concatenated along the channel dimension and then processed using parallel depthwise separable convolutions with different kernel sizes to obtain multi-scale contextual features.

[0219] The intermediate feature is obtained by fusing the multi-scale contextual features with the concatenated first, second, and third processing features.

[0220] In this embodiment of the application, the method further includes a loss function determination module, used for:

[0221] The high-quality detection box loss is determined based on the real-time detection box position, the target box position, and the height and width of the minimum bounding box.

[0222] The first loss function is determined based on the high-quality detection box loss and the cross-union ratio loss; or...

[0223] The second loss function is determined by multiplying the first loss function by the monotonic focusing coefficient; or,

[0224] A third loss function is determined based on the first loss function and the outlier; the outlier is used to describe the quality of the detection box.

[0225] The target identification device for a communication equipment room provided in this application embodiment can execute the target identification method for a communication equipment room provided in any embodiment of this application, and has the corresponding functional modules and beneficial effects of executing the method.

[0226] Figure 9 A schematic diagram of the structure of an electronic device 10 that can be used to implement embodiments of this application is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workbenches, personal digital processors, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of this application as described herein and / or the first preset requirements.

[0227] like Figure 9 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0228] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0229] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as target recognition methods in a communication room.

[0230] In some embodiments, the target identification method for a communication equipment room may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded into and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the target identification method for a communication equipment room described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform the target identification method for a communication equipment room by any other suitable means (e.g., by means of firmware).

[0231] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0232] Computer programs used to implement the methods of this application may be written in any combination of one or more programming languages. These computer programs may be provided to the processor of a general-purpose computer, a special-purpose computer, or other programmable communication equipment, such that when executed by the processor, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0233] In the context of this application, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium can be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0234] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0235] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0236] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through a communication network. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0237] This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the target identification method for communication equipment rooms as provided in any embodiment of this application.

[0238] In implementing the computer program product, computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0239] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this application can be executed in parallel, sequentially, or in different orders, as long as the desired information of the technical solution of this application can be achieved, and this is not limited herein.

[0240] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made based on the first design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A target identification method for a communication equipment room, characterized in that, The method includes: The target image inside the communication equipment room is acquired, and the target image is input into the backbone network of the target detection network for feature extraction to obtain multi-scale features; wherein, the backbone network includes a strip gating network and a feature extraction and fusion module; The multi-scale features are input into the neck network for feature update to obtain updated features; wherein, the neck network includes a cross-scale residual fusion module to capture multi-scale texture features; The updated features are input into the detection head for target recognition, resulting in the recognition results of targets in the communication equipment room.

2. The method according to claim 1, characterized in that, The target image is input into the backbone network of the target detection network for feature extraction to obtain multi-scale features, including: The target image is input into a composite convolution module for processing, and then processed through two layers of strip gating network to obtain the first scale features; The first-scale features are further processed through two layers of composite convolution modules and feature extraction and fusion modules to obtain the second-scale features; The second-scale features are processed through a fast spatial pyramid pooling layer to obtain the third-scale features.

3. The method according to claim 2, characterized in that, Processed through two layers of strip gating network, including: The target image is input into the output features of the composite convolution module and then used as the input features of the strip gating network. The input features are then processed through a deep convolutional layer. The input features after convolution are processed by one layer of strip convolution gated linear module, two layers of strip convolution gated linear module, and three layers of strip convolution gated linear module, and then fused with the input features after depth convolution.

4. The method according to claim 3, characterized in that, The processing steps after passing through the strip convolution gated linear module include: The input features that need to be input into the strip convolutional gated linear module are nonlinearly transformed by two convolutional layers and activation functions to obtain the transformed features; For the aforementioned transformation features, deep convolution is used to aggregate local contextual information, and then two strip convolutions with different kernel sizes are used in sequence to capture horizontal and vertical structural features respectively. The attention weights are then refined through convolution to generate an attention map. The attention map is multiplied element-wise with the transformed features, and then fused with the input features to determine the output features.

5. The method according to claim 2, characterized in that, The first-scale features are further processed through two layers of composite convolutional modules and feature extraction and fusion modules to obtain the second-scale features, including: During the process of the first-scale feature being processed by each layer of the composite convolution module and the feature extraction and fusion module, the first-scale feature, after being processed by the composite convolution module, can be used as the input feature of the feature extraction and fusion module. The input features are decomposed into high-frequency and low-frequency components by discrete cosine transform, and then the inverse transform is performed to obtain the reconstructed features. Attention values ​​are determined by windowed multi-head attention and frequency-guided analysis of the reconstructed features. The first weight is generated based on the high-frequency component, and the second weight is generated based on the high-frequency component and the low-frequency component. The second-scale feature is obtained by gating and weighting the spatial frequency attention branch and the frequency projection branch according to the first and second weights.

6. The method according to claim 1, characterized in that, The multi-scale features are input into the neck network for feature updating to obtain updated features, including: The multi-scale features are processed by a cross-scale residual fusion module to obtain intermediate features; The intermediate features are fused with the first-scale features and the third-scale features respectively to obtain the updated features.

7. The method according to claim 6, characterized in that, The multi-scale features are processed by a cross-scale residual fusion module to obtain intermediate features, including: The first scale features are processed by bilinear upsampling and convolution to obtain the first processed features; The second-scale feature is obtained by performing an identity mapping on the second-scale feature and adjusting the number of channels; The third-scale features are adaptively downsampled and then subjected to average pooling, max pooling, and convolution to obtain the third-processed features. The first, second, and third processed features are concatenated along the channel dimension and then processed using parallel depthwise separable convolutions with different kernel sizes to obtain multi-scale contextual features. The intermediate feature is obtained by fusing the multi-scale contextual features with the concatenated first, second, and third processing features.

8. The method according to claim 1, characterized in that, The process of determining the loss function during training includes: The high-quality detection box loss is determined based on the real-time detection box position, the target box position, and the height and width of the minimum bounding box. The first loss function is determined based on the high-quality detection box loss and the cross-union ratio loss; or... The second loss function is determined by multiplying the first loss function by the monotonic focusing coefficient; or, A third loss function is determined based on the first loss function and the outlier; the outlier is used to describe the quality of the detection box.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the target identification method for the communication room as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a processor, implement the target identification method for a communication room as described in any one of claims 1-8.

Images