Antenna quantity statistics method of target base station and training method of base station antenna target detection model

Abstract

The application provides an antenna quantity statistical method of a target base station. The method comprises the following steps: performing data enhancement on each frame image in a video file of the target base station to obtain an enhanced frame image set; performing feature extraction on the enhanced frame image set by using a semantic feature extraction module of a trained base station antenna target detection model to obtain a semantic feature map set of the frame image; performing target detection on the semantic feature map set of the frame image by using a feature map processing and fusion module of the trained base station antenna target detection model to obtain a bounding box set of the frame image; performing prediction on the bounding box set of the frame image by using the trained base station antenna target detection model to obtain the antenna quantity and antenna attribute information in each frame image; and performing antenna deduplication on the video file of the target base station according to the antenna quantity and antenna attribute information in each frame image by using a predefined judgment condition and a predefined intersection-over-union function to obtain the actual antenna quantity of the target base station.

Description

Technical Field

[0001] This invention relates to the fields of communication tower equipment inspection and communication base station maintenance technology, and in particular to a method for counting the number of antennas of a target base station, a training method for a base station antenna target detection model, electronic equipment, and storage medium. Background Technology

[0002] With the rapid development of communication technology, the number of communication towers and base stations has also increased rapidly. To ensure the stable operation of these massive numbers of communication towers and base stations and to guarantee the provision of high-quality communication services, relevant communication infrastructure service providers and / or suppliers need to perform routine maintenance on these base station equipment. The verification and inspection of the number of base station antennas is a significant and challenging issue that these service providers or suppliers face during routine maintenance. Because base station antennas are often mounted around the towers at considerable heights, in large numbers, and arranged in a 360° circle, current technologies and methods often struggle to accurately count the number of base station antennas using automated and non-manual methods, and also suffer from high costs and low efficiency. Summary of the Invention

[0003] In view of the above problems, the present invention provides a method for counting the number of antennas of a target base station, a training method for a base station antenna target detection model, an electronic device, and a storage medium, in order to at least solve one of the above problems.

[0004] According to a first aspect of the present invention, a method for counting the number of antennas of a target base station is provided, comprising:

[0005] Video files of the target base station are obtained through drone inspection, and each frame of the video file is augmented to obtain an enhanced set of frame images.

[0006] The semantic feature extraction module of the trained base station antenna target detection model is used to extract features from the enhanced frame image set to obtain a set of semantic feature maps of the frame images.

[0007] The feature map processing and fusion module of the trained base station antenna target detection model is used to perform target detection on the semantic feature map set of the frame image to obtain the bounding box set of the frame image;

[0008] The trained base station antenna target detection model is used to predict the bounding box set of frame images to obtain the number of antennas and antenna attribute information in each frame image;

[0009] Based on the number of antennas and antenna attribute information in each frame of the image, the video file of the target base station is deduplicated using predefined judgment conditions and a predefined cross-parallel ratio function to obtain the actual number of antennas of the target base station.

[0010] According to an embodiment of the present invention, the base station antenna target detection model trained above includes a single-stage base station antenna target detection model and a two-stage base station antenna target detection model.

[0011] According to embodiments of the present invention, the above-mentioned single-stage base station antenna target detection model includes a YOLOv5-based single-stage base station antenna target detection model, an SSD-based single-stage base station antenna target detection model, and a RetinaNet-based single-stage base station antenna target detection model.

[0012] The two-stage base station antenna target detection models include a two-stage base station antenna target detection model based on Fast R-CNN, a two-stage base station antenna target detection model based on Faster R-CNN, and a two-stage base station antenna target detection model based on Mask R-CNN.

[0013] According to an embodiment of the present invention, the semantic feature extraction module includes a downsampling convolution unit, a multi-layer convolution unit with a blocker, and a cascaded pooling unit;

[0014] The feature map processing and fusion module includes an instance segmentation unit, an upsampling convolution unit, and a feature map fusion unit.

[0015] According to an embodiment of the present invention, the above-mentioned method of deduplicating antennas in the video file of the target base station based on the number of antennas and antenna attribute information in each frame of the image, using predefined judgment conditions and a predefined cross-parallel ratio function, to obtain the actual number of antennas of the target base station includes:

[0016] Vertical auxiliary lines are drawn at preset positions in each frame of the image, and the positional relationship between all target objects in the current frame and all target objects in the previous frame is determined by the vertical auxiliary lines based on the number of antennas and antenna attribute information in each frame of the image.

[0017] Based on the positional changes and vertical auxiliary lines, the number of antennas detected in the current video frame is obtained, and each frame in the video file of the target base station is traversed to obtain the initial antenna objects in the video file.

[0018] Based on the initial antenna objects in the video file, the overlap between antenna objects in adjacent frames is calculated using a predefined cross-exchange-union (CEX) function.

[0019] Based on predefined judgment conditions, the initial antenna objects in the video file are deduplicated using the overlap ratio to obtain the actual number of antennas of the target base station.

[0020] According to a second aspect of the present invention, a training method for a base station antenna target detection model is provided, applied to the aforementioned method for counting the number of antennas of a target base station, comprising:

[0021] Video files of the target base station are obtained by using drones for 360-degree inspection, and each frame of the video file is labeled for the antenna object to obtain a video image sample set with truth labels.

[0022] Image data augmentation is performed on the video image sample set to obtain the augmented video image sample set, and the parameters of the base station antenna target detection model are initialized.

[0023] The base station antenna target detection model is used to perform downsampling, feature extraction, feature fusion and pooling operations on the enhanced video image sample set to obtain a semantic feature map set;

[0024] The base station antenna target detection model is used to perform feature map instance segmentation and feature map fusion operations on the semantic feature map set to obtain a fused feature map set;

[0025] The base station antenna target detection model is used to fuse feature maps for prediction, and the predicted values ​​and the corresponding true values ​​in the prediction set are input into a predefined loss function to calculate the loss value.

[0026] Based on the loss value, the parameters of the base station antenna target detection model are optimized using a predefined optimizer until the preset training conditions are met, thus obtaining the trained base station antenna target detection model.

[0027] According to an embodiment of the present invention, the above-mentioned method of acquiring video files of target base stations through 360-degree inspection by UAVs, and labeling each frame of the video file for antenna objects to obtain a video image sample set with truth labels includes:

[0028] Based on lighting conditions, weather conditions, and time information, a 360-degree drone inspection of the target base station was conducted, resulting in multiple types of video files of the target base station.

[0029] The antenna objects contained in each frame of a multi-category video file are labeled for the front and back sides of the antenna objects, resulting in a video image sample set with truth labels.

[0030] According to an embodiment of the present invention, the loss function includes a classification loss function and a bounding box regression loss function;

[0031] The predicted values ​​in the prediction set include the number of antennas of the target base station, the type of antenna, and the location information of the antennas in the image.

[0032] According to a third aspect of the present invention, an electronic device is provided, comprising:

[0033] One or more processors;

[0034] Storage device for storing one or more programs.

[0035] When one or more programs are executed by one or more processors, the one or more processors execute the above-mentioned method for counting the number of antennas of the target base station and the training method for the base station antenna target detection model.

[0036] According to a fourth aspect of the present invention, a computer-readable storage medium is provided having executable instructions stored thereon, which, when executed by a processor, cause the processor to perform the above-described method for counting the number of antennas of a target base station and the method for training a base station antenna target detection model.

[0037] The antenna count method for target base stations provided by this invention utilizes a trained base station antenna target detection model. It processes video images obtained from drone inspections of target communication base stations to obtain accurate values ​​for the number of antennas. Because this method uses drones for inspection during target base station acquisition, it avoids the dangers of manually climbing towers, significantly improving antenna inspection efficiency. Furthermore, by combining artificial intelligence methods and deep learning models into the communication base station inspection process, it replaces manual antenna counting, greatly reducing the possibility of incorrect or missed antenna counts. Moreover, the combination of drone operations and deep learning methods significantly improves upon existing technologies that only count antennas from one side of the tower, truly achieving efficient and complete antenna counting even when base station antennas are distributed across a 360-degree range. Attached Figure Description

[0038] Figure 1 This is a flowchart of a method for counting the number of antennas of a target base station according to an embodiment of the present invention;

[0039] Figure 2 This is a schematic diagram of vertical auxiliary lines in a frame image according to an embodiment of the present invention;

[0040] Figure 3 This is a schematic diagram of the critical actions between adjacent frame images during the antenna number detection process according to an embodiment of the present invention;

[0041] Figure 4 This is a flowchart of a training method for a base station antenna target detection model according to an embodiment of the present invention;

[0042] Figure 5This is a schematic diagram of base station antenna target marking according to an embodiment of the present invention;

[0043] Figure 6 This is a schematic diagram of the YOLOv5 network structure according to an embodiment of the present invention;

[0044] Figure 7 A block diagram of an electronic device adapted to implement the above-described method for counting the number of antennas of a target base station and the method for training a base station antenna target detection model is shown schematically according to an embodiment of the present invention. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments and accompanying drawings.

[0046] Currently, the main inspection methods for communication tower base stations include manual inspection, remote monitoring, and automated inspection. Manual inspection requires personnel to climb the tower to count the antennas. Since communication towers are generally quite tall, it's impossible to see all the antennas from the ground. This often requires personnel to climb from the base to the top of the tower to manually count the antennas, a time-consuming, labor-intensive, and dangerous method with high labor costs. While existing automated inspection methods still require manual operation of drones for visual counting, with the drone circling the top of the tower and personnel visually counting the antennas via real-time video from the drone's remote controller, this method is prone to errors (over- or under-counting) due to the large number and uniform appearance of the antennas. This method also demands a high level of experience and technical skill from the inspection personnel. Other inspection methods, such as counting the number of base station antennas in a single image, can use drones combined with artificial intelligence to count the number of antennas appearing in a single image of a target communication tower base station antenna. However, this method cannot count the number of base station antennas installed in a 360-degree direction around the tower and therefore lacks practical application value.

[0047] To address various technical problems in existing technologies, this invention provides a method for counting the number of antennas of a target base station and a training method for a base station antenna target detection model. This method combines UAV inspection technology, artificial intelligence technology, and deep learning methods to solve the problem of not being able to perform complete and accurate counting of the antennas of a target communication base station.

[0048] Figure 1 This is a flowchart of a method for counting the number of antennas of a target base station according to an embodiment of the present invention.

[0049] like Figure 1 As shown, the method for counting the number of antennas of the target base station includes operations S110 to S150.

[0050] When operating the S110, video files of the target base station are obtained through drone inspection, and each frame of the video file is augmented to obtain an enhanced set of frame images.

[0051] In operation S120, the semantic feature extraction module of the trained base station antenna target detection model is used to extract features from the enhanced frame image set to obtain a set of semantic feature maps of the frame images.

[0052] According to an embodiment of the present invention, the base station antenna target detection model trained above includes a single-stage base station antenna target detection model and a two-stage base station antenna target detection model.

[0053] According to embodiments of the present invention, the above-mentioned single-stage base station antenna target detection model includes a YOLOv5-based single-stage base station antenna target detection model, an SSD-based single-stage base station antenna target detection model, and a RetinaNet-based single-stage base station antenna target detection model; wherein, the dual-stage base station antenna target detection model includes a Fast R-CNN-based dual-stage base station antenna target detection model, a Faster R-CNN-based dual-stage base station antenna target detection model, and a Mask R-CNN-based dual-stage base station antenna target detection model.

[0054] According to an embodiment of the present invention, the semantic feature extraction module includes a downsampling convolution unit, a multi-layer convolution unit with a blocker, and a cascaded pooling unit; wherein, the feature map processing and fusion module includes an instance segmentation unit, an upsampling convolution unit, and a feature map fusion unit.

[0055] In operation S130, the feature map processing and fusion module of the trained base station antenna target detection model is used to perform target detection on the semantic feature map set of the frame image to obtain the bounding box set of the frame image.

[0056] In operation S140, the trained base station antenna target detection model is used to predict the bounding box set of the frame image to obtain the number of antennas and antenna attribute information in each frame image.

[0057] In operation S150, based on the number of antennas and antenna attribute information in each frame of the image, the video file of the target base station is deduplicated using predefined judgment conditions and a predefined cross-parallel ratio function to obtain the actual number of antennas of the target base station.

[0058] The antenna attribute information mentioned above includes the antenna's coordinate position information in the frame image.

[0059] The antenna count method for target base stations provided by this invention utilizes a trained base station antenna target detection model. It processes video images obtained from drone inspections of target communication base stations to obtain accurate values ​​for the number of antennas. Because this method uses drones for inspection during target base station acquisition, it avoids the dangers of manually climbing towers, significantly improving antenna inspection efficiency. Furthermore, by combining artificial intelligence methods and deep learning models into the communication base station inspection process, it replaces manual antenna counting, greatly reducing the possibility of incorrect or missed antenna counts. Moreover, the combination of drone operations and deep learning methods significantly improves upon existing technologies that only count antennas from one side of the tower, truly achieving efficient and complete antenna counts even when base station antennas are distributed across a 360-degree range.

[0060] According to an embodiment of the present invention, the above-described method of deduplicating antennas in the video file of the target base station based on the number of antennas and antenna attribute information in each frame image, using predefined judgment conditions and a predefined intersection-exchange-parallel ratio (IOP) function, to obtain the actual number of antennas of the target base station includes: drawing vertical auxiliary lines at preset positions in each frame image; determining the positional change relationship between all target objects in the current frame image and all target objects in the previous frame image based on the number of antennas and antenna attribute information in each frame image; obtaining the number of detected antennas in the current video frame image based on the positional change relationship and the vertical auxiliary lines; traversing each frame image in the video file of the target base station to obtain the initial antenna objects in the video file; calculating the overlap degree between antenna objects in adjacent frame images based on the initial antenna objects in the video file using a predefined IOP function; and deduplicating the initial antenna objects in the video file based on the overlap degree according to the predefined judgment conditions to obtain the actual number of antennas of the target base station.

[0061] The following specific embodiments, in conjunction with the appendix, demonstrate this process. Figure 2 and 3 The method for counting the number of antennas of the target base station provided by the present invention will be described in further detail.

[0062] Figure 2 This is a schematic diagram of vertical auxiliary lines in a frame image according to an embodiment of the present invention.

[0063] Figure 3 This is a schematic diagram of the critical actions between adjacent frame images during the antenna number detection process according to an embodiment of the present invention.

[0064] The trained base station antenna target detection model is used as the detection model of this invention. Assuming that the input of each scaled image in this invention is [3, 1280, 1280], and the output of the base station antenna target detection model is [n, c+5], the inference of the model can be as shown in (1):

[0065] Output = YOLOv5(Input)(1),

[0066] Where n is the number of detected targets in the output; c is the number of target types, which in this method marks the front and back of the base station antenna, and c is 2 in actual use; 5 is the position of the target in the image, including: cx, cy, w, h and conf, where cx and cy represent the position of the target center point in the image, w and h represent the width and height of the target, and conf represents the confidence level of the target's existence.

[0067] Based on the output of a single image in the object detection model, this invention creates a vertical auxiliary line at the vertical center of each image. Taking a 1920*1080 image input as an example, the coordinates of the two ends of this vertical center line L are (960, 0) and (960, 1080), respectively. Figure 2 The vertical center line is shown in the image. The location of the vertical center line is chosen mainly based on the following two considerations: (1) The center is located in the middle of the image, which can divide the image into two parts, making it easier to count the targets on the left and right sides of the center line; (2) When the UAV is flying in circles, the tower column is located in the middle of the image. At this time, the black vertical center auxiliary line can coincide with the tower column during the entire 360-degree circle flight of the UAV.

[0068] After selecting the vertical center line in the image, assume the input video sequence is I1, I2, I3, ..., I m Where m is the total number of frames in the video, which is also the total number of input images, and the input image I... i Let represent the input of the i-th image, where 1 ≤ i ≤ m; the output of each image on the model is denoted as O. i , where 1≤i≤m; define a set LeftBboxs={} to store frontal antenna targets appearing to the left of the black vertical center line; define WireCount as a statistical variable to count the number of base station antennas.

[0069] The method for counting the number of antennas of the target base station can be described by the following pseudocode.

[0070]

[0071]

[0072] The pseudocode for the above method mainly uses the vertical auxiliary center line in the frame image to divide the image in two. When the images in the video are constantly switching (which is also when the drone is circling the tower), the front antenna target on the left side will show a tendency to move to the right. When the center point of the target is at the critical position of the black center line, its previous frame I n-1 Just when it is on the left side or center line of the image, and its next frame I n When the center point of the target in the image happens to be exactly to the right of the black center line, the critical action effect is as follows: Figure 3 As shown, this is recorded as counting one antenna target. This process is repeated continuously until the algorithm has traversed the entire 360-degree video sequence of the circular flight. Then, the total number of base station antennas in the video sequence (i.e., the total number of base station antennas on the tower) can be determined. As for how to determine whether the targets in two frames are the same target, the IOU function mentioned in the table above is used. Its function is to calculate the overlap between two target boxes in the image. If the overlap between two consecutive frames exceeds 0.5 (maximum 1, minimum 0), the two targets are generally considered to be the same target.

[0073] Figure 4 This is a flowchart of a training method for a base station antenna target detection model according to an embodiment of the present invention.

[0074] like Figure 4 As shown, the training method of the above-mentioned base station antenna target detection model is applied to the antenna number statistics method of the above-mentioned target base station, including operations S410 to S460.

[0075] When operating the S410, video files of the target base station are acquired through 360-degree inspection by drones. Each frame of the video file is labeled for the antenna object to obtain a video image sample set with truth labels.

[0076] During operation S420, image data augmentation is performed on the video image sample set to obtain the augmented video image sample set, and the parameters of the base station antenna target detection model are initialized.

[0077] When operating the S430, the base station antenna target detection model is used to perform downsampling, feature extraction, feature fusion, and pooling operations on the enhanced video image sample set to obtain a semantic feature map.

[0078] In operating S440, the base station antenna target detection model is used to perform feature map instance segmentation and feature map fusion operations on the semantic feature map set to obtain a fused feature map set.

[0079] When operating the S450, the base station antenna target detection model is used to fuse the feature map set for prediction, and the predicted values ​​and the corresponding true values ​​in the prediction set are input into a predefined loss function for calculation to obtain the loss value.

[0080] In the S460 operation, based on the loss value, the parameters of the base station antenna target detection model are optimized using a predefined optimizer until the preset training conditions are met, thus obtaining the trained base station antenna target detection model.

[0081] According to an embodiment of the present invention, the loss function includes a classification loss function and a bounding box regression loss function. The predicted values ​​in the prediction set include the number of antennas of the target base station, the antenna type, and the antenna location information in the image.

[0082] According to an embodiment of the present invention, the above-mentioned method of obtaining video files of the target base station by means of 360-degree inspection by UAV, and labeling each frame of the video file for antenna objects to obtain a video image sample set with truth labels includes: performing 360-degree inspection of the target base station by UAV based on lighting conditions, weather conditions and time information to obtain multi-category video files of the target base station; labeling the antenna objects contained in each frame of the multi-category video files, respectively labeling the front and back of the antenna objects, to obtain a video image sample set with truth labels.

[0083] The following specific embodiments, in conjunction with the appendix, demonstrate this process. Figure 5 and 6 The training method for the above-mentioned base station antenna target detection model will be explained in further detail.

[0084] Figure 5 This is a schematic diagram of the base station antenna target marking according to an embodiment of the present invention.

[0085] Figure 6 This is a schematic diagram of the YOLOv5 network structure according to an embodiment of the present invention.

[0086] During the training data acquisition phase, drone aerial imagery data was collected and labeled for the tower base station antenna scenario. When collecting data, as many base station antenna images as possible should be acquired, and data should be collected under different lighting conditions, weather conditions, and time periods. Furthermore, when labeling base station antenna targets, attention should be paid to making distinct marks on the front and back of the antenna, such as... Figure 5 The image shows the labeling of the data using a labeling tool. The black dashed boxes indicate targets in front of the antenna, and the black solid boxes indicate targets behind the antenna. After labeling the antenna dataset, the dataset can be packaged for use by subsequent deep learning object detection algorithms.

[0087] In the model training phase, (ii) a base station antenna target detection model is trained and optimized using a deep learning target detection model on the labeled base station antenna dataset. This method employs the YOLOv5 target detection model. YOLOv5 is currently the mainstream single-stage target detection model, with near-optimal model accuracy and inference speed, and is widely used in industrial fields such as smart security, smart transportation, and intelligent inspection. YOLOv5 performs target detection on feature maps at three scales: large, medium, and small. The YOLOv5 model architecture is as follows: Figure 6 As shown.

[0088] The training process of the YOLOv5 model mainly includes the preparation stage, the feature extraction stage, the feature processing stage, and the target prediction stage.

[0089] During the preparation phase, the images need to undergo data augmentation (especially Mosaic data augmentation) and a set of anchor presets need to be initialized (YOLOv5 provides different general presets for models with different numbers of parameters).

[0090] In the feature extraction (Backbone) stage, the basic structures Conv, C3, and SPPF were used to extract features from the input image. Conv was used to downsample the input (a total of 5 downsampling operations were performed); C3 was used to extract and fuse features from the input, enriching the semantic information of the features. In this process, Boottleneck was used to reduce the number of parameters and computational cost, and the CSPNet idea was borrowed to enhance the learning ability of CNN; SPPF used pooling and feature fusion to enrich the semantic information of the features, so that the deepest feature maps have extremely rich semantic information.

[0091] In the feature processing (Neck) stage, shallow features are fused between the feature maps of the three scales to be detected (shallow features are beneficial for detection). PANet fuses shallow features into the extracted feature maps, so that the feature maps have both rich semantic information and accurate object location information.

[0092] In the head prediction stage, the processed feature map is predicted, and the parameter weights are optimized based on the loss function (Classification Loss and Bounding Box Regeression Loss) and the optimizer.

[0093] The pre-trained model used in this invention is yolov5s.pt, the network input is set to 1280*1280, the confidence level is set to 0.5, and the IOU threshold is set to 0.5.

[0094] Figure 7 A block diagram of an electronic device adapted to implement the above-described method for counting the number of antennas of a target base station and the method for training a base station antenna target detection model is shown schematically according to an embodiment of the present invention.

[0095] like Figure 7 As shown, an electronic device 700 according to an embodiment of the present invention includes a processor 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage portion 708 into a random access memory (RAM) 703. The processor 701 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 701 may also include onboard memory for caching purposes. The processor 701 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present invention.

[0096] RAM 703 stores various programs and data required for the operation of electronic device 700. Processor 701, ROM 702, and RAM 703 are interconnected via bus 704. Processor 701 executes various operations of the method flow according to embodiments of the present invention by executing programs in ROM 702 and / or RAM 703. It should be noted that programs may also be stored in one or more memories other than ROM 702 and RAM 703. Processor 701 may also execute various operations of the method flow according to embodiments of the present invention by executing programs stored in one or more memories.

[0097] According to an embodiment of the present invention, the electronic device 700 may further include an input / output (I / O) interface 705, which is also connected to a bus 704. The electronic device 700 may also include one or more of the following components connected to the I / O interface 705: an input section 706 including a keyboard, mouse, etc.; an output section 707 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 708 including a hard disk, etc.; and a communication section 709 including a network interface card such as a LAN card, modem, etc. The communication section 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the I / O interface 705 as needed. A removable medium 711, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 710 as needed so that computer programs read from it can be installed into the storage section 708 as needed.

[0098] The present invention also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of the present invention.

[0099] According to embodiments of the present invention, a computer-readable storage medium may be a non-volatile computer-readable storage medium, such as including, but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of the present invention, a computer-readable storage medium may include ROM 702 and / or RAM 703 and / or one or more memories other than ROM 702 and RAM 703 described above.

[0100] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0101] The above specific embodiments further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for counting the number of antennas at a target base station, characterized in that, include: The video file of the target base station is obtained by drone inspection, and each frame of the video file is augmented to obtain an augmented frame image set. The semantic feature extraction module of the trained base station antenna target detection model is used to extract features from the enhanced frame image set to obtain a set of semantic feature maps of the frame images. The feature map processing and fusion module of the trained base station antenna target detection model is used to perform target detection on the semantic feature map set of the frame image to obtain the bounding box set of the frame image; The trained base station antenna target detection model is used to predict the bounding box set of the frame image to obtain the number of antennas and antenna attribute information in each frame image; Based on the number of antennas and antenna attribute information in each frame of the image, antenna deduplication is performed on the video file of the target base station using predefined judgment conditions and a predefined intersection-exchange-parallel ratio (IOP) function to obtain the actual number of antennas of the target base station. This includes: drawing vertical auxiliary lines at preset positions in each frame of the image; determining the positional change relationship between all target objects in the current frame and all target objects in the previous frame based on the number of antennas and antenna attribute information in each frame of the image; obtaining the number of detected antennas in the current video frame based on the positional change relationship and the vertical auxiliary lines; traversing each frame of the video file of the target base station to obtain the initial antenna objects in the video file; calculating the overlap between antenna objects in adjacent frames based on the initial antenna objects in the video file using the predefined IOP function; and deduplicating the initial antenna objects in the video file using the overlap based on the predefined judgment conditions to obtain the actual number of antennas of the target base station. The trained base station antenna target detection model includes a single-stage base station antenna target detection model and a two-stage base station antenna target detection model. The single-stage base station antenna target detection model includes a YOLOv5-based single-stage base station antenna target detection model, an SSD-based single-stage base station antenna target detection model, and a RetinaNet-based single-stage base station antenna target detection model. The dual-stage base station antenna target detection model includes a dual-stage base station antenna target detection model based on Fast R-CNN, a dual-stage base station antenna target detection model based on Faster R-CNN, and a dual-stage base station antenna target detection model based on Mask R-CNN.

2. The method according to claim 1, characterized in that, The semantic feature extraction module includes a downsampling convolutional unit, a multi-layer convolutional unit with a blocker, and a cascaded pooling unit. The feature map processing and fusion module includes an instance segmentation unit, an upsampling convolution unit, and a feature map fusion unit.

3. A training method for a base station antenna target detection model, applied to the method described in any one of claims 1-2, characterized in that, include: Video files of the target base station are obtained by using drones for 360-degree inspection, and each frame of the video file is labeled for the antenna object to obtain a video image sample set with truth labels. Image data augmentation is performed on the video image sample set to obtain an augmented video image sample set, and the parameters of the base station antenna target detection model are initialized. The base station antenna target detection model is used to perform downsampling, feature extraction, feature fusion, and pooling operations on the enhanced video image sample set to obtain a semantic feature map set. The base station antenna target detection model is used to perform feature map instance segmentation and feature map fusion operations on the semantic feature map set to obtain a fused feature map set. The base station antenna target detection model uses the fused feature map set to make predictions, and the predicted values ​​and the corresponding true values ​​in the prediction set are input into a predefined loss function to calculate the loss value. Based on the loss value, the parameters of the base station antenna target detection model are optimized using a predefined optimizer until the preset training conditions are met, thus obtaining a trained base station antenna target detection model.

4. The method according to claim 3, characterized in that, Video files of the target base station were acquired through 360-degree drone inspection, and each frame of the video file was labeled for the antenna object, resulting in a video image sample set with ground truth labels, including: Based on lighting conditions, weather conditions, and time information, a 360-degree drone inspection is conducted on the target base station to obtain multiple types of video files of the target base station. The antenna objects contained in each frame of the multi-category video files are labeled with the front and back sides of the antenna objects to obtain the video image sample set with truth labels.

5. The method according to claim 3, characterized in that, The loss function includes a classification loss function and a bounding box regression loss function; The predicted values ​​in the prediction set include the number of antennas of the target base station, the type of antenna, and the location information of the antennas in the image.

6. An electronic device, comprising: One or more processors; Storage device for storing one or more programs. Wherein, when the one or more programs are executed by the one or more processors, the one or more processors perform the method according to any one of claims 1 to 4.

7. A computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, cause the processor to perform the method according to any one of claims 1 to 4.

Images