Antenna inspection method, device and storage medium for communication base station

Abstract

The present invention discloses an antenna inspection method, device and storage medium for a communication base station, the method comprising: extracting target image data from an inspection request; creating an inspection task according to the inspection request, and dividing the inspection task into a first inspection item, a second inspection item and a third inspection item; classifying the obstructions above the antenna in the target image data to generate first inspection information; sequentially detecting and classifying the labels and packages below the antenna in the target image data to generate second inspection information; sequentially detecting and measuring the distance between antennas in the target image data to generate third inspection information; combining the first inspection information, the second inspection information and the third inspection information to generate an inspection result for the inspection task; and sending an inspection response carrying the inspection result to a client. The antenna of a satellite positioning system is remotely inspected and accepted, and the efficiency of the antenna of the satellite positioning system is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision, and in particular to an antenna inspection method, device and storage medium for a communication base station. Background Art

[0002] With the rapid popularization of communication technology, the communication industry is in a stage of rapid development. There is a demand for the construction of communication base stations. One of the acceptance links for the construction of communication base stations is to ensure the correct installation of the antenna of the satellite positioning system, so as to ensure that the communication base station can accurately and stably receive satellite signals, and realize accurate positioning services and reliable data transmission.

[0003] At present, the antenna installation of the satellite positioning system supports remote acceptance. On-site construction personnel collect image data to be inspected in the area where the antenna is constructed, upload the image data to the remote quality inspection system, and the acceptance personnel manually inspect the image data, analyze it according to the acceptance specifications, and give the analysis results.

[0004] However, the installation scale of satellite positioning system antennas varies and the distribution area is wide. The workload of manual inspection increases dramatically. The manual inspection method is subjective. Different inspection personnel may have different understandings and judgments on the antenna of the same satellite positioning system. This subjectivity leads to inconsistent inspection results. In addition, some minor quality problems may be ignored due to negligence of manual inspection, affecting the final project quality, thus resulting in low efficiency of inspection. Summary of the invention

[0005] In view of this, the present invention provides an antenna inspection method, device and storage medium for a communication base station, so as to improve the efficiency of inspecting the antenna of a satellite positioning system in a communication base station.

[0006] A first aspect of the present invention provides an antenna inspection method for a communication base station, comprising:

[0007] When receiving an inspection request sent by a client, extracting target image data from the inspection request; the target image data is image data collected by an antenna for receiving signals of a satellite positioning system in a communication base station;

[0008] Creating an inspection task according to the inspection request, and dividing the inspection task into a first inspection item, a second inspection item, and a third inspection item;

[0009] For the first inspection item, classifying the obstructions located above the antenna in the target image data to generate first inspection information;

[0010] For the second inspection item, sequentially detecting and classifying the labels and packages located under the antenna in the target image data to generate second inspection information;

[0011] For the third inspection item, sequentially detecting and measuring the distances between the antennas in the target image data to generate third inspection information;

[0012] generating an inspection result for the inspection task in combination with the first inspection information, the second inspection information and the third inspection information;

[0013] The inspection response carrying the inspection result is sent to the client for business processing.

[0014] A second aspect of the present invention provides an antenna inspection device for a communication base station, comprising:

[0015] An inspection request processing module is used to extract target image data from the inspection request when receiving the inspection request sent by the client; the target image data is image data collected by an antenna for receiving signals of a satellite positioning system in a communication base station;

[0016] An inspection task division module, used to create an inspection task according to the inspection request, and divide the inspection task into a first inspection item, a second inspection item and a third inspection item;

[0017] a first inspection module, configured to classify, for the first inspection item, an obstruction located above the antenna in the target image data to generate first inspection information;

[0018] a second inspection module, for detecting and classifying the labels and packages located under the antenna in the target image data in sequence for the second inspection item, so as to generate second inspection information;

[0019] a third inspection module, configured to detect and measure the distances between the antennas in sequence in the target image data for the third inspection item, so as to generate third inspection information;

[0020] An inspection result generating module, used to generate an inspection result for the inspection task by combining the first inspection information, the second inspection information and the third inspection information;

[0021] The inspection response processing module is used to send the inspection response carrying the inspection result to the client for business processing.

[0022] A third aspect of the present invention provides an electronic device, the electronic device comprising:

[0023] at least one processor; and

[0024] a memory communicatively connected to the at least one processor; wherein,

[0025] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the antenna inspection method of the communication base station as described in the first aspect above.

[0026] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the antenna inspection method of a communication base station as described in the first aspect above.

[0027] A fifth aspect of the present invention provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the antenna inspection method of the communication base station as described in the first aspect above.

[0028] In this embodiment, when an inspection request sent by a client is received, target image data is extracted from the inspection request; the target image data is image data collected from an antenna for receiving signals of a satellite positioning system in a communication base station; an inspection task is created based on the inspection request, and the inspection task is divided into a first inspection item, a second inspection item, and a third inspection item; for the first inspection item, obstructions located above the antenna are classified in the target image data to generate first inspection information; for the second inspection item, labels and packages located below the antenna are detected and classified in turn in the target image data to generate second inspection information; for the third inspection item, the distance between antennas is detected and measured in turn in the target image data to generate third inspection information; an inspection result is generated for the inspection task by combining the first inspection information, the second inspection information, and the third inspection information; and an inspection response carrying the inspection result is sent to the client for business processing. While maintaining the integrity of the inspection task as much as possible, one inspection task is logically divided into a first inspection item, a second inspection item and a third inspection item. The first inspection item, the second inspection item and the third inspection item are respectively inspected using independent business logic, so that the feature ranges concerned by the first inspection item, the second inspection item and the third inspection item converge, which can effectively reduce the impact of various factors, improve the accuracy and reliability of the inspection, meet the requirements of actual construction, and realize remote acceptance of the antenna of the satellite positioning system, which greatly reduces the workload of manual inspection and reduces the impact of the subjectivity of manual inspection, thereby effectively improving the efficiency of accepting the antenna of the satellite positioning system.

[0029] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0031] Figure 1 It is a flow chart of an antenna inspection method for a communication base station provided in Embodiment 1 of the present invention.

[0032] Figure 2 This is a schematic diagram of interaction between a client and an engine provided in Embodiment 1 of the present invention.

[0033] Figure 3 This is a flow chart of a first inspection item provided in the first embodiment of the present invention.

[0034] Figure 4 and Figure 5 This is a flow chart of a second inspection item provided in the first embodiment of the present invention.

[0035] Figure 6 This is a flow chart of a third inspection item provided in the first embodiment of the present invention.

[0036] Figure 7 , Figure 8 and Fig. 9 This is an example diagram for calculating antenna spacing provided in Embodiment 1 of the present invention.

[0037] Fig.10 It is a structural diagram of an antenna inspection device for a communication base station provided in Embodiment 2 of the present invention.

[0038] Fig.11 It is a structural schematic diagram of an electronic device provided in Embodiment 3 of the present invention. DETAILED DESCRIPTION

[0039] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0040] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present invention described herein can cover sequential implementations other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0041] Embodiment 1

[0042] See also Figure 1 , shows a flow chart of a communication base station antenna inspection method provided by Embodiment 1 of the present invention, the method can be executed by an antenna inspection device of a communication base station, the antenna inspection device of the communication base station can be implemented in the form of hardware and / or software, and the antenna inspection device of the communication base station can be configured in an electronic device. Figure 1 As shown, the method includes:

[0043] Step 101: When receiving an inspection request sent by a client, extract target image data from the inspection request.

[0044] This embodiment can be applied to a quality inspection platform, in which an engine is deployed. The engine is responsible for maintaining the business logic of accepting the antenna of the satellite positioning system. The business logic is integrated into the platform in the form of a remote call service and provided to construction personnel in the form of an API (Application Programming Interface), so that on-site construction personnel can use the remote call service through the client to accept the antenna of the satellite positioning system.

[0045] Among them, the satellite positioning system includes GPS (Global Positioning System), Beidou Navigation Satellite System (BDS), etc. The antenna of the satellite positioning system is used to receive the signal sent by the satellite positioning system.

[0046] At this time, the construction personnel control the client Clinet to collect target image data, which is the image data collected by the antenna used to receive signals from the satellite positioning system in the communication base station. The target image data is encapsulated into an inspection request and the inspection request is sent to the quality inspection platform. The quality inspection platform parses the structure of the inspection request and extracts the target image data from the inspection request.

[0047] In actual implementation, Figure 2 As shown, the client Clinet can create a check request based on the JsonRPC (remote procedure call based on Josn (JavaScript Object Notation, JS object notation)) protocol. The JsonRPC protocol defines a simple message format to implement communication between the client and the server (such as the engine).

[0048] Generally, the inspection request includes fields such as the version of the JsonRPC protocol, the method name, the request parameters, and the first identifier (such as ID), wherein the request parameters include the target image data.

[0049] For example, the format of the check request is:

[0050] {"jsonrpc":

[0051] "2.0","method","floor_AD_detection","params":{"image":exp.jpg},"id":1}

[0052] Among them, the version number of JosnRPC is 2.0, the method name is method, it specifies the remote method (ie, API interface) floor_AD_detection to be called, the request parameter params is used as the input parameter of the method floor_AD_detection, and the target image data image is provided, whose name is exp.jpg, and the first identifier id is 1.

[0053] The client sends the check request to the engine request queue (Engine-Request) through the push socket of the message queue library ZeroMQ.

[0054] Among them, the message queue library ZeroMQ provides a series of socket types to support different messaging modes. Push sockets are used to send messages to one or more recipients without worrying about how the messages are distributed.

[0055] Then, the check request created based on the JsonRPC protocol and sent by the client through the Push socket of the message queue library ZeroMQ is received in the engine request queue.

[0056] The engine acts as a server and listens to the inspection request from the client. When it receives the inspection request, it calls the engine to read the inspection request from the engine request queue through the pull socket of the message queue library ZeroMQ.

[0057] The Pull socket corresponds to the Push socket, which receives messages from the sender.

[0058] Thereafter, the engine Engine is called to process the inspection request, and the request parameters and the first identifier are read from the inspection request; wherein the request parameters include the target image data, and the business logic of the antenna of the acceptance satellite positioning system is run on the target image data.

[0059] Step 102: Create an inspection task according to the inspection request, and divide the inspection task into a first inspection item, a second inspection item, and a third inspection item.

[0060] The engine Engine responds to the inspection request, creates a new inspection task, and divides the inspection task into a first inspection item, a second inspection item, and a third inspection item.

[0061] Among them, the first inspection item is used to check whether there is any obstruction above the antenna, so as to improve the stability and reliability of navigation, positioning and other services provided by the satellite positioning system.

[0062] The second inspection item is to check whether the label and packaging of the antenna meet the installation specifications; among them, the label uses text to record the key information of the antenna, such as "5G-GPS signal line", and the transition area under the antenna is often wrapped with tape to achieve waterproofing.

[0063] The third inspection item is to check whether the distance between antennas meets the installation specifications.

[0064] The deep learning-based method can be applied to the acceptance of satellite positioning system antennas. The deep learning method is mainly based on image classification, which includes multiple key steps such as data collection, model training, testing and post-processing.

[0065] First, the image data of the satellite positioning system antenna is used as sample data, including whether there is an obstruction above or not, whether the label is clear or not, whether there is a wrapping or not, and whether the spacing between antennas meets the specifications or not.

[0066] These sample data rely on professionals to mark them, clearly indicating which part of the installation position of the antenna in the sample data has a problem and whether it passes.

[0067] Then, the sample data is divided into data sets, and deep learning-based image classification models are trained according to the divided data sets, such as ResNet, EfficientNet, DenseNet, etc. These image classification models can learn the image feature information of positive and negative samples during the training process.

[0068] Finally, the trained image classification model is deployed to the remote quality inspection platform.

[0069] In actual applications, the image classification model will analyze the image data transmitted from the construction site in real time and quickly identify the problems with the satellite positioning system antenna. After analyzing the specific problems, the relevant information will be automatically fed back to the on-site construction personnel and maintenance personnel.

[0070] Since the environment for installing satellite positioning system antennas is relatively complex and the number of detection tasks is large, if one image classification model is used to learn all tasks, it is not conducive to directly learning the common visual features of positive and negative samples.

[0071] Most of the collected sample data are obtained by taking photos by on-site construction workers after the construction is completed. This makes the sample data used to train the image classification model almost normal sample data (i.e. positive samples). The problematic sample data (i.e. negative samples), such as those without waterproof tape and labels, are not only scarce but also unevenly distributed, which will make the image classification model tend to judge these negative samples as normal.

[0072] In addition, due to different shooting angles and distances, the presentation of antennas, tags, packages and the distance between antennas in the image data will be significantly different. These differences may be reflected in the size, direction, and spacing of the antennas, the clarity of the tags, the size and type of the packages, and even the visual effects caused by lighting conditions and background environment.

[0073] These factors will increase the recognition difficulty of a single image classification model, which may lead to misjudgment or missed detection, reducing the accuracy and reliability of the inspection.

[0074] In this embodiment, the process of accepting the antenna of the satellite positioning system is analyzed, and while maintaining the integrity of the inspection task as much as possible, one inspection task is logically divided into a first inspection item, a second inspection item, and a third inspection item. The first inspection item, the second inspection item, and the third inspection item are inspected respectively using independent business logic, so that the feature ranges concerned by the first inspection item, the second inspection item, and the third inspection item converge, which can effectively reduce the impact of various factors, improve the accuracy and reliability of the inspection, and meet the requirements of actual construction.

[0075] Step 103: for the first inspection item, classify the obstructions located above the antenna in the target image data to generate first inspection information.

[0076] For the first inspection item, based on the deep learning classification method, the obstructions above the antenna can be classified in the target image data to identify whether there are any obstructions above the antenna, and the first inspection information is generated based on the identification result.

[0077] In one embodiment of the present invention, step 103 may include the following steps:

[0078] Step 1031: Load a first classification network for a first inspection item.

[0079] In this embodiment, a first classification network may be constructed and trained in advance based on a deep learning classification method for the task of identifying whether there is an obstruction above the antenna.

[0080] Among them, Figure 3 As shown, in the first classification network, the pre-trained residual network ResNet (such as ResNet18) is used as the backbone structure Backbone, and the fully connected layer (Fully Connected, FC) is used as the head structure Head.

[0081] If an image classification model is retrained as the first classification network, the image classification model is prone to overfitting when the sample data (the antenna of the satellite positioning system) is limited.

[0082] Considering that the pre-trained residual network ResNet has been trained on a wide range of data sets, the large-scale image data in the data set can extract rich visual features and better learn and identify various occlusions, greatly reducing training time and resource consumption. At the same time, the extensive knowledge acquired by the pre-trained residual network ResNet helps reduce the risk of overfitting and improve the generalization ability of new data, so that more accurate occlusion detection can be achieved in scenarios with limited sample data.

[0083] At this point, the pre-trained weights (PretrainedWeights) of the residual network ResNet on a large-scale data set can be obtained, and the pre-trained residual network ResNet is used as the backbone structure Backbone of the first classification network. The backbone structure Backbone is the basic structure of the first classification network and is responsible for extracting the features of the image data. By using the pre-trained residual network ResNet as the backbone structure Backbone of the first classification network, its powerful feature extraction capability can be used in the classification task of whether there is occlusion above the satellite positioning system.

[0084] The original residual network ResNet is designed for multi-class classification. The classification task of whether there is an occlusion above the satellite positioning system is a binary classification problem, that is, judging whether there is an occlusion above the satellite positioning system antenna. Therefore, the head structure Head of the first classification network will be modified to adapt to binary classification. To this end, a fully connected layer FC and an activation function (such as Softmax) are added to the end of the first classification network to output the probabilities of two first categories, where the first category includes the existence of an occlusion above the satellite positioning system antenna (that is, the satellite positioning system antenna is occluded) and the absence of an occlusion above the satellite positioning system antenna (that is, the satellite positioning system antenna is not occluded).

[0085] The first classification network is fine-tuned using a data set collected for a classification task of whether there is an occlusion above the satellite positioning system. During the fine-tuning process, trainable parameters are continuously adjusted to better adapt to the classification task of whether there is an occlusion above the satellite positioning system.

[0086] In this embodiment, the pre-trained and fine-tuned residual network ResNet is used as the backbone structure Backbone, and the fine-tuned fully connected layer (Fully Connected, FC) is used as the head structure Head, which can improve the simplicity and efficient reasoning speed of the first classification network, and can process a large amount of target image data in a short time. In addition, the use of the pre-trained residual network ResNet as the basis can reduce the time and computing resources for training the first classification network.

[0087] Exemplarily, in the process of training the first classification network, the data set contains 7716 positive samples and 1093 negative samples. 80% of the overall data set is divided into training data. On this basis, a cross-validation strategy is implemented, that is, in each round, 20% of the training data is divided as the verification data of this round, and the remaining 80% is used for training. In this way, the verification data of each round is unique, ensuring that the first classification network can cover the entire training data for learning. In the testing phase, the image data is input into the first classification network that performs best in the verification data in each round for prediction. For the best first classification network output in each round, the results are accumulated and the softmax function is applied to obtain the predicted probability value, which will be used as the output of the first classification network.

[0088] In this example, the backbone structure Backbone selects the residual network ResNet18, and adopts pre-training and non-pre-training training methods respectively, with a total of 20 rounds of training. The highest probability value is the first category of the prediction.

[0089] When the engine runs the business logic of the antenna of the satellite positioning system, the first classification network is loaded into the memory for execution.

[0090] Step 1032: Input the target image data into the residual network to extract image features.

[0091] like Figure 3 As shown, for the current first inspection item, the current target image data (GPS Images) is input into the residual network ResNet serving as the backbone structure Backbone, and the residual network ResNet extracts image features from the current target image data according to its logic.

[0092] Step 1033: Map the image features into the fully connected layer as logical values ​​of the first category.

[0093] like Figure 3 As shown, the image features features output by the residual network ResNet are input into the fully connected layer FC. The fully connected layer FC classifies the image features according to its logic Classifier and maps them into logical values ​​Logits of each first category.

[0094] Step 1034: Activate the logical value as the probability value of the first category.

[0095] like Figure 2 As shown, the logical values ​​of each first category output by the fully connected layer FC are input into activation functions such as Softmax for operation to obtain the probability values ​​Probilities of each first category.

[0096] Step 1035: Write the first category with the highest probability value into the first inspection information.

[0097] The probability values ​​of the first categories are compared, and the first category with the highest probability value is screened out, and written into the first inspection information as the inspection result of the first inspection item.

[0098] Step 104 : For the second inspection item, sequentially detect and classify the labels and packages located below the antenna in the target image data to generate second inspection information.

[0099] For the second inspection item, the target detection method based on deep learning can be used to detect the labels and packages under the antenna in the target image data. In addition, the classification method based on deep learning can be used to classify the label and package obstructions, identify whether the labels and packages under the antenna meet the specifications, and generate the second inspection information based on the identification results.

[0100] In one embodiment of the present invention, step 104 may include the following steps:

[0101] Step 1041 : loading a first object detection network, a second object detection network, a second classification network, and a third classification network for a second inspection item.

[0102] In this embodiment, a target detection method based on deep learning can be pre-constructed and trained for the task of detecting labels under the antenna, and a second target detection network can be constructed and trained for the task of detecting packages under the antenna.

[0103] The tasks of detecting the label under the antenna and detecting whether the package under the antenna meets the specifications are multi-target detection tasks in an industrial environment. While ensuring the detection accuracy, there are also high requirements for the detection speed. Therefore, the first target detection network and the second target detection network can be selected as networks such as YOLOv7. YOLOv7 meets the requirements of real-time processing and can detect multiple categories and positions at the same time during a single forward pass. The detection is fast and efficient, which meets the requirements of the task of detecting whether the label under the antenna and the package meet the specifications.

[0104] In one embodiment of the present invention, step 1041 may further include the following steps:

[0105] Step 10411, collect sample image data.

[0106] When training the first target detection network and the second target detection network, sample image data can be collected as a data set by taking photos on site, etc. The content of the sample image data includes a label located under the antenna and a wrapping (such as tape, metal, etc.) located in the transition area below the antenna.

[0107] Step 10412: Divide the sample image data into a training set and a validation set.

[0108] Among the sample image data, part or all of the sample image data can be selected for annotation. For example, in a certain training, the data set contained approximately 10,000 sample image data, all of which were image data of the satellite positioning system antenna in working condition taken on-site by construction workers. 2,366 of the sample image data were selected for annotation as training sets and verification sets, and the remaining sample image data will be used directly for detection.

[0109] Step 10413: In the training set, perform data augmentation operations on the sample image data to generate new sample image data.

[0110] In the training set, the sample image data is adjusted to a uniform size, such as 640×640. According to the environment in which the antenna of the satellite positioning system is installed in the communication base station, data enhancement operations are performed on the sample image data to generate new sample image data, thereby improving the diversity of the sample image data and helping to optimize the performance of the first target detection network.

[0111] The data enhancement operation includes at least one of the following:

[0112] 1. Adjust at least one of brightness, contrast, saturation and hue for sample image data, which helps maintain performance under different lighting conditions and color deviations.

[0113] 2. Perform a slight rotation and / or affine transformation on the sample images, which helps the first object detection network learn to recognize objects from different angles.

[0114] 3. Fill the mosaic of the sample image data, and splice multiple frames of sample image data that fill the mosaic, such as splicing 4 frames of sample image data into a new frame of sample image data, so that sample image data containing multiple contexts can be formed.

[0115] 4. Linearly superimpose multiple frames of sample image data, such as adding two frames of sample image data according to a certain weight ratio, mixing parts or attributes of the sample image data, and promoting the first target detection network to learn more generalized features.

[0116] 5. Writing some object instances in the current sample image data to reasonable locations in other sample image data helps increase the first target detection network's ability to recognize the diversity of objects in different scenarios, especially helping to improve the accuracy of small sample categories and bounding box positioning.

[0117] Step 10414: If the data enhancement operation is completed, the sample image data in the training set is used to train the first target detection network or the second target detection network.

[0118] In this embodiment, multiple rounds (eg, 300 rounds) of training are performed on the training set, and the first object detection network or the second object detection network with the highest mAP is selected.

[0119] Step 10415: If the training is completed, the first object detection network or the second object detection network is verified using the sample image data in the verification set.

[0120] In the validation set, the sample image data is adjusted to a uniform size, such as 640×640, and data augmentation operations on the sample image data can be ignored.

[0121] The sample image data in the verification set is used to verify the various performance indicators of the first target detection network or the second target detection network. When the verification passes, the first target detection network or the second target detection network is deployed to the quality inspection platform.

[0122] In addition, a second classification network can be constructed and trained in advance based on a deep learning classification method for the task of identifying whether the label under the antenna meets the specifications, and a third classification network can be constructed and trained for the task of identifying whether the package under the antenna meets the specifications.

[0123] In this embodiment, the type of the first classification network is the same as the type of the second classification network (such as ResNet18 in the ResNet series). When the first classification network and the second classification network are facing similar environments and the tasks are relatively simple, not only can the classification accuracy be guaranteed, but also the cost of training, deployment and application can be reduced.

[0124] Among them, the task of identifying whether the label under the antenna meets the specifications is a binary classification task. Accordingly, the second category output by the second classification network includes the text on the label being clear and the text on the label being unclear.

[0125] When training the second classification network, the positive sample labels are sample image data with clear text, and the negative sample labels are sample image data with unclear text.

[0126] In a certain training, in the training set, the positive samples contained 1063 image data, and the negative samples contained 669 sample image data; in the validation set, the positive samples contained 141 image data, and the negative samples contained 85 sample image data.

[0127] The task of identifying whether the package under the antenna meets the specifications is a multi-classification task. Accordingly, the third category output by the third classification network includes the transition area is not wrapped by tape or the transition area is incompletely wrapped by tape (category 0), the transition area is completely wrapped by tape (category 1), and the transition area is wrapped by a metal protective shell (category 2).

[0128] When training the third classification network, class 0 is a negative sample, and class 1 and class 2 are positive samples.

[0129] In a certain training, in the training set, there are 609 sample image data of class 0, 879 sample image data of class 1, and 499 sample image data of class 2; in the validation set, there are 85 sample image data of class 0, 75 sample image data of class 1, and 79 sample image data of class 2.

[0130] When training the second classification network and the third classification network, the image data in the two data sets (training set and validation set) are adjusted to a uniform size, such as 224×224. During the training process, data enhancement operations are performed on the training set. Specifically, the sample image data in the training set of the second classification network are randomly flipped horizontally and / or vertically with a probability of 0.5, and a random rotation of 45° is used. The sample image data in the training set of the third classification network is randomly flipped horizontally with a probability of 0.5. During the verification process, the data enhancement operation of the validation set can be ignored.

[0131] In one example, the third classification network can be selected as a network of the type such as the B0 architecture of EfficientNet. In the B0 architecture of EfficientNet, the image data is passed through a standard convolution layer to obtain a preliminary feature map to obtain a feature map.

[0132] The feature map will pass through a series of MBConv layers; each MBConv layer contains an expansion layer, a depthwise separable convolution, a SE (Squeeze-and-Excitation) layer, and an output layer. The feature map is expanded in the expansion layer to provide a basis for extracting more complex features. In the depthwise separable convolution, the feature map is first subjected to a depthwise convolution operation, that is, each channel uses a convolution kernel independently to process the spatial relationship; then a point-by-point convolution is performed to linearly combine the outputs of all channels at each spatial position to learn the dependencies between channels. The feature map output by the depthwise separable convolution will be processed by the SE layer, which first passes through a global average pooling layer to compress the feature map and generate a scalar representation of the global context for each channel. The scalar representation captures the global information of the channel and helps to strengthen important features in the subsequent steps. Next, these scalar representations of the global context are fed into a module consisting of a two-layer fully connected network, which usually includes a compression layer and an expansion layer. The compression layer first reduces the number of parameters and reduces the dimensionality, and then the expansion layer expands it back to the original number of channels. This process generates the weights for each channel. These weights strengthen the focus on useful features and suppress unimportant feature representations. The feature map is then adjusted to the size required for subsequent operations through a 1×1 convolution.

[0133] After passing through all MBConv layers, the feature map passes through a fully connected layer and outputs the classification result (i.e., the second category and the third category).

[0134] EfficientNet has the advantages of small model parameters and fast inference speed, and can be quickly deployed in industrial detection environments.

[0135] When the engine runs the business logic of the antenna of the satellite positioning system, the first target detection network, the second classification network and the third classification network are loaded into the memory for execution.

[0136] Step 1042: Input the target image data into the first target detection network to generate a first detection box and a second detection box.

[0137] like Figure 4 and Figure 5 As shown, for the current second inspection item, the current target image data is input into the first target detection network (such as YOLOv7), and the first target detection network performs the target detection operation according to its logic and generates a first detection frame on the target image data, wherein the content of the first detection frame is the label under the antenna.

[0138] Step 1043: input the target image data into a second target detection network to generate a second detection frame.

[0139] like Figure 5As shown, for the current second inspection item, the current target image data is input into the second target detection network (such as YOLOv7), and the second target detection network performs the target detection operation according to its logic and generates a second detection frame on the target image data, wherein the content of the second detection frame is the package in the transition area below the antenna.

[0140] Step 1044: crop the content within the first detection frame from the target image data to obtain first area data.

[0141] like Figure 4 As shown, if there is a first detection box, the content within the first detection box (i.e., the label under the antenna) is cropped out along the edge of the first detection box in the target image data and recorded as the first area data, thereby increasing the proportion of the label under the antenna in the entire image data (i.e., the first area data), which helps to improve the accuracy of classifying the label under the antenna.

[0142] Step 1045: Crop the content within the second detection frame from the target image data to obtain second area data.

[0143] like Figure 5 As shown, if there is a second detection frame, the content within the second detection frame (i.e., the package in the transition area below the antenna) is cropped out along the edge of the second detection frame in the target image data and recorded as the second area data, thereby increasing the proportion of the package in the transition area below the antenna in the entire image data (i.e., the second area data), which helps to improve the accuracy of classifying the package in the transition area below the antenna.

[0144] Step 1046: input the first region data into the second classification network for classification to obtain a second category.

[0145] like Figure 4 As shown, the first region data is input into the second classification network, and the second classification network performs binary classification (Image Classifier) ​​on the first region data according to its logic, outputs probability values ​​(Probilities) of two second categories, and outputs the second category with the highest probability value.

[0146] Among them, the second category includes clear text on the label and unclear text on the label.

[0147] Step 1047: input the second region data into a third classification network for classification to obtain a third category.

[0148] like Figure 5As shown, the second region data is input into the third classification network, and the third classification network performs multi-classification (Image Classifier) ​​on the second region data according to its logic, outputs probability values ​​(Probilities) of three third categories, and outputs the third category with the highest probability value.

[0149] Among them, the third category includes the transition area not being wrapped by tape or the transition area being incompletely wrapped by tape, the transition area being completely wrapped by tape, and the transition area being wrapped by a metal protective shell.

[0150] Step 1048: Write the second category and the third category into the second inspection information.

[0151] In this embodiment, the second category and the third category are written together into the second inspection information as the inspection result of the second inspection item.

[0152] Step 105 : For the third inspection item, sequentially detect and measure the distances between antennas in the target image data to generate third inspection information.

[0153] For the third inspection item, the target image data mainly includes the antennas of the satellite positioning system, and most of the antennas of the satellite positioning system in the target image data can be guaranteed to have at least two antennas. The background of the target image data is relatively complex. The installation environments of the satellite positioning system antennas range from cities to rural areas, from plains to mountains, and even on the roofs or wires outside the balconies of residential buildings. The shooting angles of the satellite positioning system antennas vary, including frontal shooting, close-up shots, wide-angle shots, and bird's-eye views. Multiple shooting angles bring complexity to the mapping from plane distance to space distance.

[0154] In addition, in some target image data, the satellite positioning system antennas are not fully captured, and there is even an extreme case where the exposed part of the satellite positioning system antenna is less than one-tenth; at the same time, in a small number of target image data, the satellite positioning system antennas block each other, resulting in the obscured satellite positioning system antennas being unable to expose most of the antennas. These factors also lead to uncertainty in the estimation and detection of the spacing between satellite positioning system antennas.

[0155] Most satellite positioning system antennas are photographed from the front, and most of them are conical in shape and have similar features, which is suitable for target detection. However, the shooting angles in the target image data are complex, especially the shapes taken from the top and bottom angles are quite different from those taken from the front; some satellite positioning system antennas taken from a distance account for a small proportion in the template image data and belong to the category of small objects. At the same time, the satellite positioning system antennas that are partially blocked or at the edge of the target image data may only be exposed by less than half.

[0156] Generally, if the target image data is taken from the front and the plane formed by the two satellite positioning system antennas is perpendicular to the camera device, it is relatively easy to judge whether the distance exceeds the threshold (such as 20 cm). If the scales of the two satellite positioning system antennas in the target image data are inconsistent and disparate, due to the perspective principle, there is another dimension of depth distance, and the two-dimensional distance in the target image data cannot be completely proportional to the distance in the three-dimensional. If the antenna of the satellite positioning system is incomplete or even blocked in the target image data, it is difficult to effectively use the prior knowledge that the diameter of the satellite positioning system antenna is 10 cm.

[0157] Therefore, if Figure 6 As shown, the inspection for the third inspection item can be divided into two stages. In the first stage, the antenna can be detected in the target image data based on a deep learning target detection method (such as YOLOv7). In the second stage, the antenna can be post-processed using an antenna distance classification algorithm customized for the antenna of the satellite positioning system to measure whether the spacing between the antennas meets the specifications, such as whether it exceeds 20 cm, and the second inspection information is generated based on the processing results.

[0158] In one embodiment of the present invention, step 105 may include the following steps:

[0159] Step 1051: Load a third object detection network for the third inspection item.

[0160] In this embodiment, a third target detection network may be constructed and trained in advance for the task of detecting antennas based on a target detection method based on deep learning.

[0161] Furthermore, the type of the first target detection network, the type of the second target detection network and the type of the third target detection network are the same (such as YOLOv7). When the environments faced by the first target detection network, the second target detection network and the third target detection network are similar and the tasks are relatively simple, not only the accuracy of target detection can be guaranteed, but also the cost of training, deployment and application can be reduced.

[0162] When training the third object detection network, the training set contains 338 sample image data and the validation set contains 91 sample image data.

[0163] Step 1052: Input the target image data into a third target detection network to generate multiple third detection frames.

[0164] The target image data is input into the second target detection network. The second target detection network performs target detection operations on the target image data according to its logic, and generates multiple (at least two) third detection frames on the target image data, wherein the content of each third detection frame is the antenna itself.

[0165] Step 1053 , determine whether the ratio between the areas of the two third detection frames is outside a preset interval; if so, execute step 1054 , if not, execute step 1055 .

[0166] In this embodiment, combined detection is performed on pairs of third detection frames in multiple third detection frames. For any two third detection frames, the areas of the third detection frames can be counted, and the ratio between the areas of any two third detection frames can be calculated to determine whether the ratio is outside a preset interval, where the interval includes 1.

[0167] Step 1054: Determine whether the distance between the two antennas is greater than the installation distance as distance information.

[0168] If the ratio between the areas of any two third detection frames is outside the preset range, it means that the size ratios of the antennas of any two satellite positioning systems are quite different. According to the perspective principle that objects are larger when they are near and smaller when they are far away, it can be considered that the distance between the antennas of any two satellite positioning systems is greater than the installation distance (such as 20 cm). This conclusion is set as the distance information.

[0169] Step 1055: Measure whether the distance between the two antennas is greater than the installation distance based on the two third detection frames as distance information.

[0170] If the ratio between the areas of any two third detection frames is within a preset range, it means that the size ratio of the antennas of any two satellite positioning systems is relatively small, especially when the size ratio of the antennas of any two satellite positioning systems is close to 1:1. It can be considered that the camera device is perpendicular to the plane formed by the antennas of any two satellite positioning systems, and there is basically no depth distance between the antennas of any two satellite positioning systems in the target image data.

[0171] At this time, any two third detection frames can be compared to measure whether the distance between the two antennas is greater than the installation distance, and the result is recorded as distance information.

[0172] In one embodiment of the present invention, step 1055 may further include the following steps:

[0173] Step 10551: Use two third detection frames to calculate the shortest separation distance between two antennas.

[0174] In this embodiment, the information of the two third detection frames can be used to estimate the shortest spacing distance between the antennas of any two satellite positioning systems, and the spacing between the antennas of any two satellite positioning systems can be approximately considered.

[0175] In one example, if Figure 7 As shown, the distance between the center points of the two third detection frames is calculated as the first center distance, and half of the sum of the horizontal dimensions of the two third detection frames is taken as the first correction value. Assuming that half of the horizontal dimension of one third detection frame is d1, and half of the horizontal dimension of the other third detection frame is d2, the first correction value is (d1+d2).

[0176] The difference between the first center distance distance and the first correction value (d1+d2) is equivalent to the shortest separation distance min_distance between the two antennas, that is, min_distance=distance-d1-d2.

[0177] In this example, the method for calculating the shortest separation distance between two antennas is simple and universal.

[0178] In another example, considering that the above example involves the positional relationship of two third detection frames (rectangles), directly using the sum of the lateral dimensions of the two third detection frames may result in a correction value that is too large and the shortest spacing distance between the two antennas being too close under certain conditions. Therefore, a correction is made in this example to improve accuracy.

[0179] In this example, if Figure 7 and Figure 8 As shown, the distance between the center points of the two third detection frames is calculated as the second center distance distance.

[0180] The difference between the horizontal sizes of the two third detection frames is calculated as the horizontal gap, and the difference between the vertical sizes of the two third detection frames is calculated as the vertical gap, so as to compare the horizontal gap with the vertical gap.

[0181] If the horizontal difference is greater than the vertical difference, half of the sum of the horizontal sizes of the two third detection frames is taken to obtain the second correction value. Suppose half of the horizontal size of one third detection frame is d1, and half of the horizontal size of the other third detection frame is d2, then the second correction value is (d1+d2).

[0182] At this time, the difference between the second center distance distance and the second correction value (d1+d2) is equivalent to the shortest spacing distance min_distance between the two antennas, that is, min_distance=distance-d1-d2.

[0183] If the horizontal difference is less than or equal to the vertical difference, half of the sum of the vertical dimensions of the two third detection frames is taken to obtain the second correction value. Suppose half of the vertical dimension of one third detection frame is h1, and half of the vertical dimension of the other third detection frame is h2, then the second correction value is (h1+h2).

[0184] At this time, the difference between the second center distance distance and the second correction value (h1+h2) is equivalent to the shortest spacing distance min_distance between the two antennas, that is, min_distance=distance-h1-h2.

[0185] In yet another example, considering that the above two examples have some errors in certain business scenarios, another approximation algorithm is proposed in this example to improve applicability.

[0186] In this example, if Fig. 9 As shown, the midpoints (coordinates) of each side in the third detection frame are obtained, and the distance (such as Euclidean distance) between the midpoints of any side in the two third detection frames is calculated to obtain the midpoint distance.

[0187] The third detection frame is a rectangle, and each of the third detection frames has four sides and four midpoints. The distances between the midpoints of each side in the third detection frame are calculated pairwise, and a total of 16 midpoint distances are calculated and recorded as (disatance1, disatance2, disatance3, ..., disatance16).

[0188] Use the minimum function min() to take the smallest one from all midpoint distances as the shortest distance min_distance between the two antennas, that is, min_distance=min(disatance1, disatance2, disatance3, ..., disatance16).

[0189] Step 10552: enlarge the smallest of the horizontal dimensions of the two third detection frames and use it as the distance threshold.

[0190] The smallest lateral size of any two third detection frames is taken as approximately the diameter of the antenna of the satellite positioning system, which is multiplied by an amplification factor (such as 2) to implement a magnification process and obtain a distance threshold.

[0191] Step 10553: If the shortest interval distance is greater than the distance threshold, it is determined that the distance between the two antennas is greater than the installation distance.

[0192] Step 10554: If the shortest interval distance is less than or equal to the distance threshold, it is determined that the distance between the two antennas is not greater than the installation distance.

[0193] The shortest separation distance between antennas of any two satellite positioning systems is compared with the distance threshold.

[0194] If the shortest spacing distance between antennas of any two satellite positioning systems is greater than the distance threshold, it can be considered that the spacing between antennas of any two satellite positioning systems is greater than the installation distance.

[0195] If the shortest spacing distance between antennas of any two satellite positioning systems is less than or equal to the distance threshold, it can be considered that the spacing between antennas of any two satellite positioning systems is not greater than the installation distance.

[0196] Step 1056: Write the distance information into the third inspection information.

[0197] In this embodiment, the distance information is written into the third inspection information as the inspection result of the third inspection item.

[0198] Step 106: Generate an inspection result for the inspection task by combining the first inspection information, the second inspection information, and the third inspection information.

[0199] For the first inspection item, if there is no obstruction above the antenna, that is, there is no obstruction above the antenna, it can be considered that the first inspection item has been passed; if there is an obstruction above the antenna, that is, there is obstruction above the antenna, it can be considered that the first inspection item has not been passed.

[0200] For the second inspection item, if the font on the label under the antenna is clear, and the transition area under the antenna is completely wrapped with tape or the transition area is wrapped by a metal protective shell, it can be considered that the second inspection item has been passed; if the font on the label under the antenna is not clear, and / or the transition area under the antenna is not wrapped with tape or the transition area is incompletely wrapped with tape, it can be considered that the second inspection item has not been passed.

[0201] For the third inspection item, if the distance between the two antennas is greater than the installation distance, it can be considered that the third inspection item has been passed; if the distance between the two antennas is not greater than the installation distance, it can be considered that the third inspection item has not been passed.

[0202] When the first inspection information, the second inspection information and the third inspection information are all passed, an inspection result may be generated for the inspection task as passed acceptance.

[0203] When any one of the first inspection information, the second inspection information and the third inspection information fails, an inspection result may be generated for the inspection task as failing the acceptance.

[0204] Step 107: Send the inspection response carrying the inspection result to the client for business processing.

[0205] In this embodiment, an inspection response carrying the inspection results can be sent to the client. The client parses the inspection response, reads the inspection results therefrom, and performs corresponding business processing based on the inspection results, for example, updating the webUI (web user interface) of the quality inspection platform, storing inspection results, and so on.

[0206] In the specific implementation, Figure 2 As shown, the engine creates a check response based on the JsonRPC protocol.

[0207] Generally, the check response contains fields such as the version of the JsonRPC protocol, the response result and the second identifier (such as ID). The second identifier in the check response should be the same as the first identifier in the check request to ensure that the check response is for a specific check request, and the response result includes the check result.

[0208] Exemplarily, the format of the check response is as follows:

[0209] {"jsonrpc":"2.0","result":"pass","id":1}

[0210] Among them, the version number of JosnRPC is 2.0, the check result is pass, and the first identifier id is 1.

[0211] The calling engine Engine writes the inspection response into the engine response queue (Engine-Response) through the push socket of the message queue library ZeroMQ to push the inspection response to the client; the client Client is used to read the inspection response from the engine response queue (Engine-Response) through the pull socket of the message queue library ZeroMQ, and perform business processing according to the inspection result when the second identifier is the same as the first identifier.

[0212] This embodiment uses ZeroMQ and JosnRPC to design an AI quality inspection interface for remote call services, which greatly reduces the frequency of on-site inspection of satellite positioning system antennas by maintenance personnel. It not only improves acceptance efficiency and reduces costs, but also reduces risks caused by human factors.

[0213] In this embodiment, when an inspection request sent by a client is received, target image data is extracted from the inspection request; the target image data is image data collected from an antenna for receiving signals of a satellite positioning system in a communication base station; an inspection task is created based on the inspection request, and the inspection task is divided into a first inspection item, a second inspection item, and a third inspection item; for the first inspection item, obstructions located above the antenna are classified in the target image data to generate first inspection information; for the second inspection item, labels and packages located below the antenna are detected and classified in turn in the target image data to generate second inspection information; for the third inspection item, the distance between antennas is detected and measured in turn in the target image data to generate third inspection information; an inspection result is generated for the inspection task by combining the first inspection information, the second inspection information, and the third inspection information; and an inspection response carrying the inspection result is sent to the client for business processing. While maintaining the integrity of the inspection task as much as possible, one inspection task is logically divided into a first inspection item, a second inspection item and a third inspection item. The first inspection item, the second inspection item and the third inspection item are respectively inspected using independent business logic, so that the feature ranges concerned by the first inspection item, the second inspection item and the third inspection item converge, which can effectively reduce the impact of various factors, improve the accuracy and reliability of the inspection, meet the requirements of actual construction, and realize remote acceptance of the antenna of the satellite positioning system, which greatly reduces the workload of manual inspection and reduces the impact of the subjectivity of manual inspection, thereby effectively improving the efficiency of accepting the antenna of the satellite positioning system.

[0214] Embodiment 2

[0215] See also Fig.10 , shows a schematic diagram of the structure of an antenna inspection device for a communication base station provided by Embodiment 2 of the present invention. Fig.10 As shown, the device comprises:

[0216] The inspection request processing module 1001 is used to extract target image data from the inspection request when receiving the inspection request sent by the client; the target image data is image data collected by the antenna for receiving the signal of the satellite positioning system in the communication base station;

[0217] An inspection task division module 1002 is used to create an inspection task according to the inspection request, and divide the inspection task into a first inspection item, a second inspection item, and a third inspection item;

[0218] A first inspection module 1003, configured to classify, for the first inspection item, obstructions located above the antenna in the target image data to generate first inspection information;

[0219] A second inspection module 1004 is used to detect and classify the labels and packages located under the antenna in the target image data in order for the second inspection item to generate second inspection information;

[0220] A third inspection module 1005 is used to detect and measure the distances between the antennas in the target image data in sequence for the third inspection item to generate third inspection information;

[0221] An inspection result generating module 1006, configured to generate an inspection result for the inspection task in combination with the first inspection information, the second inspection information and the third inspection information;

[0222] The inspection response processing module 1007 is used to send the inspection response carrying the inspection result to the client for business processing.

[0223] In one embodiment of the present invention, the first checking module 1003 includes:

[0224] A first model loading module is used to load a first classification network for the first inspection item; the first classification network uses a residual network as a backbone structure and a fully connected layer as a head structure;

[0225] An image feature extraction module, used for inputting the target image data into the residual network to extract image features;

[0226] A logic value mapping module, used to map the image feature input into the fully connected layer into a logic value of a first category; the first category includes whether there is an occluder above the antenna or no occluder above the antenna;

[0227] A probability value calculation module, used for activating the logic value as the probability value of the first category;

[0228] The first inspection information generating module is configured to write the first category having the highest probability value into the first inspection information.

[0229] In one embodiment of the present invention, the second checking module 1004 includes:

[0230] A second network loading module, used for loading the first target detection network, the second target detection network, the second classification network and the third classification network for the second inspection item;

[0231] A first target detection module, used for inputting the target image data into the first target detection network to generate a first detection frame; the content of the first detection frame is the label under the antenna;

[0232] A second target detection module is used to input the target image data into the second target detection network to generate a second detection frame; the content of the second detection frame is the package in the transition area below the antenna;

[0233] A first cropping module, used for cropping the content located in the first detection frame from the target image data to obtain first area data;

[0234] A second cropping module, used for cropping the content located in the second detection frame from the target image data to obtain second area data;

[0235] A label classification module, used for inputting the first area data into the second classification network for classification to obtain a second category; the second category includes the text on the label being clear and the text on the label being unclear;

[0236] a wrapping classification module, used for inputting the second area data into the third classification network for classification to obtain a third category; the third category includes that the transition area is not wrapped by the tape or the transition area is incompletely wrapped by the tape, the transition area is completely wrapped by the tape, and the transition area is wrapped by a metal protective shell;

[0237] The second inspection information generating module is used to write the second category and the third category into the second inspection information.

[0238] In one embodiment of the present invention, the second network loading module includes:

[0239] A sample image data acquisition module, used to acquire sample image data; the content of the sample image data includes a label located below the antenna and a package located in a transition area below the antenna;

[0240] A sample image data division module, used for dividing the sample image data into a training set and a verification set;

[0241] A data enhancement operation execution module, used to perform a data enhancement operation on the sample image data in the training set to generate new sample image data;

[0242] An object detection network training module, configured to train the first object detection network or the second object detection network using the sample image data in the training set if the data enhancement operation is completed;

[0243] An object detection network verification module, configured to verify the first object detection network or the second object detection network using the sample image data in the verification set if training is completed;

[0244] The data enhancement operation includes at least one of the following:

[0245] adjusting at least one of brightness, contrast, saturation and hue of the sample image data;

[0246] Performing a slight rotation and / or affine transformation on the sample image;

[0247] Filling mosaics on the sample image data, and splicing multiple frames of the sample image data filled with mosaics;

[0248] Linearly superimposing multiple frames of sample image data;

[0249] Writing some object instances in the current sample image data onto other sample image data.

[0250] In one embodiment of the present invention, the third checking module 1005 includes:

[0251] A third network loading module, used for loading a third object detection network for the third inspection item;

[0252] A third target detection module, used for inputting the target image data into the third target detection network to generate a plurality of third detection frames; the content of the third detection frame is the antenna;

[0253] an area ratio determination module, configured to determine whether the ratio between the areas of the two third detection frames is outside a preset interval; if so, calling the first distance information generation module, and if not, calling the second distance information generation module;

[0254] A first distance information generating module, used to determine that the distance between the two antennas is greater than the installation distance as distance information;

[0255] A second distance information generating module, used for measuring whether the distance between the two antennas is greater than the installation distance according to the two third detection frames as distance information;

[0256] The third inspection information generating module is used to write the distance information into the third inspection information.

[0257] In one embodiment of the present invention, the second distance information generating module includes:

[0258] A spacing distance calculation module, used to calculate the shortest spacing distance between the two antennas using the two third detection frames;

[0259] A distance threshold generating module, used for enlarging the smallest of the lateral dimensions of the two third detection frames as a distance threshold;

[0260] A first relationship determination module, configured to determine that the distance between the two antennas is greater than the installation distance if the shortest interval distance is greater than the distance threshold;

[0261] The second relationship determination module is used to determine that the distance between the two antennas is not greater than the installation distance if the shortest spacing distance is less than or equal to the distance threshold.

[0262] In one embodiment of the present invention, the interval distance calculation module includes:

[0263] A first center distance calculation module, used to calculate the distance between the center points of the two third detection frames as a first center distance;

[0264] A first correction value calculation module, used for taking half of the sum of the lateral dimensions of the two third detection frames as a first correction value;

[0265] A first spacing distance equivalent module, used for equating the difference between the first center distance and the first correction value to the shortest spacing distance between the two antennas;

[0266] or,

[0267] A second center distance calculation module, used to calculate the distance between the center points of the two third detection frames as a second center distance;

[0268] A lateral difference calculation module, used for calculating the difference between the lateral sizes of the two third detection frames as the lateral difference;

[0269] A longitudinal difference calculation module, used for calculating the difference between the longitudinal dimensions of the two third detection frames as the longitudinal difference;

[0270] A second correction value calculation module, configured to obtain a second correction value by taking half of the sum of the horizontal dimensions of the two third detection frames if the horizontal difference is greater than the vertical difference; and obtain a second correction value by taking half of the sum of the vertical dimensions of the two third detection frames if the horizontal difference is less than or equal to the vertical difference;

[0271] A second spacing distance equivalent module, used for equating the difference between the second center distance and the second correction value to the shortest spacing distance between the two antennas;

[0272] or,

[0273] A midpoint distance calculation module, used to calculate the distance between the midpoints of any sides of the two third detection frames to obtain the midpoint distance;

[0274] The third spacing distance equivalent module is used to select the smallest one from all the midpoint distances as the shortest spacing distance between the two antennas.

[0275] In one embodiment of the present invention, the inspection request processing module 1001 includes:

[0276] The engine request queue storage module is used to receive in the engine request queue the inspection request created based on the JsonRPC protocol and sent by the client through the Push socket of the message queue library ZeroMQ;

[0277] An inspection request pull module, used for calling the engine to read the inspection request from the engine request queue through the pull socket of the message queue library ZeroMQ;

[0278] An inspection request parsing module, used for calling an engine to read a request parameter and a first identifier from the inspection request; the request parameter includes target image data;

[0279] The inspection response processing module 1007 includes:

[0280] An inspection response creation module, used for creating an inspection response based on the JsonRPC protocol; the inspection response has a response result and a second identifier; the response result includes the inspection result;

[0281] The check response push module is used to call the engine to write the check response into the engine response queue through the push socket of the message queue library ZeroMQ to push the check response to the client; the client is used to read the check response from the engine response queue through the pull socket of the message queue library ZeroMQ, and perform business processing according to the check result when the second identifier is the same as the first identifier.

[0282] The antenna inspection device for a communication base station provided in an embodiment of the present invention can execute the antenna inspection method for a communication base station provided in any embodiment of the present invention, and has corresponding functional modules and beneficial effects for executing the antenna inspection method for a communication base station.

[0283] Embodiment 3

[0284] See also Fig.11 , shows a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, blade servers, mainframe computers, and other suitable computers. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0285] like Fig.11 As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0286] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0287] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as the antenna inspection method of the communication base station.

[0288] In some embodiments, the antenna inspection method of the communication base station may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the antenna inspection method of the communication base station described above may be performed. Alternatively, in other embodiments, the processor 11 may be configured to execute the antenna inspection method of the communication base station in any other appropriate manner (e.g., by means of firmware).

[0289] Various implementations 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 chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0290] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0291] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0292] To provide interaction with a user, the systems and techniques described herein may 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 a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0293] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0294] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0295] Embodiment 4

[0296] An embodiment of the present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the antenna inspection method of the communication base station provided in any embodiment of the present invention is implemented.

[0297] In the process of implementation, the computer program product can be written in one or more programming languages ​​or a combination thereof to perform the computer program code of the present invention, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (for example, using an Internet service provider to connect through the Internet).

[0298] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0299] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for inspecting an antenna of a communication base station, characterized in that: include: When receiving the inspection request sent by the client, extracting the target image data from the inspection request; The target image data is image data collected by an antenna for receiving signals of a satellite positioning system in a communication base station; Creating an inspection task according to the inspection request, and dividing the inspection task into a first inspection item, a second inspection item, and a third inspection item; For the first inspection item, classifying the obstructions located above the antenna in the target image data to generate first inspection information; For the second inspection item, sequentially detecting and classifying the labels and packages located under the antenna in the target image data to generate second inspection information; For the third inspection item, sequentially detecting and measuring the distances between the antennas in the target image data to generate third inspection information; generating an inspection result for the inspection task in combination with the first inspection information, the second inspection information and the third inspection information; Sending the inspection response carrying the inspection result to the client for business processing; The step of sequentially detecting and measuring the distances between the antennas in the target image data for the third inspection item to generate third inspection information includes: For the third inspection item, loading a third object detection network; Inputting the target image data into the third target detection network to generate a plurality of third detection frames; the content of the third detection frame is the antenna; Determining whether a ratio between the areas of the two third detection frames is outside a preset interval; If yes, determining that the distance between the two antennas is greater than the installation distance is used as distance information; If not, measuring whether the distance between the two antennas is greater than the installation distance according to the two third detection frames as distance information; The distance information is written into the third inspection information.

2. The method according to claim 1, characterized in that The step of classifying, for the first inspection item, the obstruction located above the antenna in the target image data to generate first inspection information includes: Loading a first classification network for the first inspection item; the first classification network uses a residual network as a backbone structure and a fully connected layer as a head structure; Inputting the target image data into the residual network to extract image features; Inputting the image feature into the fully connected layer and mapping it into a first category of logical values; the first category includes whether there is an occluder above the antenna or there is no occluder above the antenna; activating the logical value as a probability value of the first category; The first category having the highest probability value is written into the first inspection information.

3. The method according to claim 1, characterized in that For the second inspection item, sequentially detecting and classifying the labels and packages located under the antenna in the target image data to generate second inspection information includes: Loading a first target detection network, a second target detection network, a second classification network, and a third classification network for the second inspection item; Inputting the target image data into the first target detection network to generate a first detection frame; the content of the first detection frame is the label under the antenna; Inputting the target image data into the second target detection network to generate a second detection frame; the content of the second detection frame is the package in the transition area below the antenna; Cut out the content within the first detection frame from the target image data to obtain first area data; Cut out the content within the second detection frame from the target image data to obtain second area data; Inputting the first area data into the second classification network for classification to obtain a second category; the second category includes the text on the label being clear and the text on the label being unclear; Inputting the second region data into the third classification network for classification to obtain a third category; the third category includes the transition region not being wrapped by tape or the transition region being incompletely wrapped by tape, the transition region being completely wrapped by tape, and the transition region being wrapped by a metal protective shell; The second category and the third category are written into the second inspection information.

4. The method according to claim 3, characterized in that The loading of the first target detection network, the second target detection network, the second classification network, and the third classification network for the second inspection item includes: Collecting sample image data; the sample image data includes a label located below the antenna and a package located in a transition area below the antenna; Dividing the sample image data into a training set and a validation set; In the training set, performing a data augmentation operation on the sample image data to generate new sample image data; If the data enhancement operation is completed, the sample image data in the training set is used to train the first object detection network or the second object detection network; If the training is completed, the first object detection network or the second object detection network is verified using the sample image data in the verification set; The data enhancement operation includes at least one of the following: adjusting at least one of brightness, contrast, saturation and hue of the sample image data; Performing a slight rotation and / or affine transformation on the sample image; Filling mosaics on the sample image data, and splicing multiple frames of the sample image data filled with mosaics; Linearly superimposing multiple frames of sample image data; Writing some object instances in the current sample image data onto other sample image data.

5. The method according to claim 1, characterized in that: The measuring, based on the two third detection frames, whether the distance between the two antennas is greater than the installation distance as distance information includes: Using the two third detection frames to calculate the shortest spacing distance between the two antennas; enlarging the smallest of the lateral dimensions of the two third detection frames as a distance threshold; If the shortest spacing distance is greater than the distance threshold, determining that the spacing between the two antennas is greater than the installation distance; If the shortest spacing distance is less than or equal to the distance threshold, it is determined that the spacing between the two antennas is not greater than the installation distance.

6. The method according to claim 5, characterized in that The using the two third detection frames to calculate the shortest spacing distance between the two antennas includes: Calculating the distance between the center points of the two third detection frames as a first center distance; Taking half of the sum of the horizontal dimensions of the two third detection frames as the first correction value; equating the difference between the first center distance and the first correction value to the shortest spacing distance between the two antennas; or, Calculating the distance between the center points of the two third detection frames as the second center distance; Calculating a difference between the horizontal sizes of the two third detection frames as a horizontal gap; Calculating the difference between the longitudinal dimensions of the two third detection frames as the longitudinal gap; If the horizontal difference is greater than the vertical difference, half of the sum of the horizontal sizes of the two third detection frames is taken to obtain a second correction value; If the horizontal difference is less than or equal to the vertical difference, taking half of the sum of the vertical dimensions of the two third detection frames to obtain a second correction value; equating the difference between the second center distance and the second correction value to the shortest spacing distance between the two antennas; or, Calculate the distance between the midpoints of any sides of the two third detection frames to obtain the midpoint distance; The smallest one among all the midpoint distances is taken as the shortest spacing distance between the two antennas.

7. The method according to any one of claims 1 to 6, characterized in that When receiving the inspection request sent by the client, extracting the target image data from the inspection request includes: Receive the check request created based on the JsonRPC protocol and sent by the client through the Push socket of the message queue library ZeroMQ in the engine request queue; The calling engine reads the inspection request from the engine request queue through the pull socket of the message queue library ZeroMQ; The calling engine reads a request parameter and a first identifier from the inspection request; the request parameter includes target image data; The step of sending the inspection response carrying the inspection result to the client for business processing includes: An inspection response created based on the JsonRPC protocol; the inspection response includes a response result and a second identifier; the response result includes the inspection result; The calling engine writes the check response into the engine response queue through the push socket of the message queue library ZeroMQ to push the check response to the client; the client is used to read the check response from the engine response queue through the pull socket of the message queue library ZeroMQ, and perform business processing according to the check result when the second identifier is the same as the first identifier.

8. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can perform the antenna inspection method of the communication base station according to any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the antenna inspection method of the communication base station according to any one of claims 1 to 7 is implemented.

Images