Airport fence state detection image method of unmanned vehicle-mounted terminal based on machine learning

Abstract

The invention discloses an airport fence state detection image method for an unmanned vehicle-mounted terminal based on machine learning, and the method comprises the steps: obtaining the original image data of an airport fence collected by the unmanned vehicle-mounted terminal, carrying out the preprocessing of the original image data of the airport fence, and obtaining the standardized image data; constructing a fence state detection model based on a deep learning algorithm; deploying a fence state detection model to an unmanned vehicle-mounted terminal, and performing real-time detection on the standardized image data to obtain a fence state detection result; and a fence state detection result is displayed through a display screen of the unmanned vehicle-mounted terminal or a remote monitoring center, and when an abnormal state is detected, an alarm mechanism is triggered to notify related personnel. The airport fence state can be automatically, accurately and efficiently detected, the problems existing in traditional manual inspection are solved, and powerful technical support is provided for guaranteeing airport safety.

Description

Technical Field

[0001] The present invention relates to the field of computer vision technology, and in particular to an airport fence status detection image method for an unmanned vehicle-mounted terminal based on machine learning. Background Art

[0002] As an important facility to ensure airport safety, the integrity of airport fences is directly related to the overall safety protection effect of the airport. Traditionally, the status detection of airport fences mainly relies on manual inspections. Inspectors check the fences regularly or irregularly to ensure that the fences are not damaged or have loopholes. However, this traditional manual inspection method has many significant defects.

[0003] First, manual inspection is inefficient. Airports are vast and fences are often long. Manual inspection is not only time-consuming and laborious, but also cannot achieve full coverage. Inspectors need to spend a lot of time and energy to check the fences one by one in a large area. Especially in complex terrain or bad weather conditions, the efficiency of this inspection method will be further reduced. Secondly, manual inspection is easily affected by human factors. Inspectors may fail to detect in place due to fatigue, negligence, lack of experience or other subjective factors, which directly affects the accuracy of detection. For example, during night inspections, the line of sight is limited, which is easy to miss or misjudge, and long working hours may also cause the inspectors to lose concentration, thereby increasing the safety risks of the fence. In addition, the timeliness of manual inspection cannot be guaranteed. During the operation of the airport, the fence may be damaged by external forces at any time, such as bad weather, collision with foreign vehicles, etc. Since the inspection has a certain periodicity, damage may occur during the inspection interval, which provides opportunities for criminals and seriously threatens the safety of the airport.

[0004] In the existing technology, although some automated detection systems have been proposed, most of them rely on fixed sensors or cameras. Although these systems can improve the timeliness of detection, their coverage is limited and they are restricted by the installation location, and they cannot flexibly respond to the status detection of various areas of the fence. More importantly, the existing automatic detection systems are usually based on simple image processing algorithms, which are difficult to cope with complex environmental changes such as lighting, shadows, weather, etc., and are prone to false alarms or missed alarms.

[0005] Therefore, there is an urgent need for an unmanned vehicle-mounted airport fence status detection image method based on machine learning. Summary of the invention

[0006] The present invention provides an airport fence status detection image method for an unmanned vehicle-mounted terminal based on machine learning to solve the above-mentioned problems existing in the prior art.

[0007] In order to achieve the above object, the present invention provides the following technical solutions:

[0008] The airport fence status detection image method based on machine learning on the unmanned vehicle side includes:

[0009] S101: acquiring original image data of the airport fence collected by the unmanned vehicle terminal, preprocessing the original image data of the airport fence, and acquiring standardized image data;

[0010] S102: Building a fence status detection model based on deep learning algorithm;

[0011] S103: deploying the fence status detection model to the unmanned vehicle, and obtaining the fence status detection result by performing real-time detection on the standardized image data;

[0012] S104: The fence status detection result is displayed on the display screen of the unmanned vehicle or the remote monitoring center. When an abnormal state is detected, an alarm mechanism is triggered to notify relevant personnel.

[0013] Wherein, step S101 includes:

[0014] S1011: Capture images of the airport fence using a dual-spectrum high-definition camera mounted on the unmanned vehicle to obtain original image data of the airport fence. The original image data includes fence images captured under different lighting conditions, different angles, and different weather conditions.

[0015] S1012: pre-processing the original image data of the airport fence, cropping the original image of the airport fence according to a preset cropping parameter, scaling the cropped image according to a preset scaling ratio, and normalizing the scaled image data;

[0016] S1013: Acquire the preprocessed standardized image data, and save the standardized image data into an image database.

[0017] Wherein, step S102 includes:

[0018] S1021: Use deep learning algorithms to perform preliminary feature extraction on standardized image data. In the deep learning algorithm, multiple convolutional layers are stacked to extract low-level and high-level features. After the convolutional layer, a pooling layer is added to perform feature dimension reduction. In the feature extraction process, a residual structure is introduced to solve the gradient vanishing problem in model training, and an attention mechanism is introduced to enhance the model's ability to recognize key feature areas.

[0019] S1022: Input the extracted image features into the fully connected layer for feature integration and classification, use a number of image data labeled with normal and abnormal fence states as training sets, train the fence state detection model, and obtain the trained fence state detection model. During the training process, use the back propagation algorithm and optimizer to adjust the parameters of the fence state detection model.

[0020] Among them, obtaining the fence status detection result includes:

[0021] Compressing the fence state detection model to obtain compressed fence state detection model data;

[0022] Pruning the compressed fence state detection model data to reduce model parameters and calculation amount, and obtaining a pruned fence state detection model;

[0023] Deploy the pruned fence status detection model to the unmanned vehicle to generate a model to be detected;

[0024] Obtain fence image data captured in real time by the dual-spectrum high-definition camera on the unmanned vehicle, and pre-process the fence image data as standardized image data;

[0025] Input the standardized image data into the model to be detected for real-time detection;

[0026] Get fence status detection results from real-time detection.

[0027] Among them, the fence status detection results are displayed on the display screen of the unmanned vehicle or the remote monitoring center, including:

[0028] Determine the fence status detection result output by the display screen of the unmanned vehicle and transmit it to the remote monitoring center. The detection result includes normal status result and abnormal status result;

[0029] Determine and output a plurality of new fence status detection results, determine the status of each new detection result and transmit it to the remote monitoring center until the number of new abnormal status results reaches a preset number threshold, or the current detection status result is determined to be the last detection result;

[0030] Control the unmanned vehicle to record and save the image data of abnormal status, and return to execute the output of the initial fence status detection result through the display screen or the remote monitoring center until the number of new abnormal status results reaches the preset number threshold, or the current detection status result is determined to be the last detection result;

[0031] Generate and output a list based on each test result and the status of the test result;

[0032] Determine the fence status detection results received by the unmanned vehicle terminal or remote monitoring center and transmit them to the alarm mechanism.

[0033] Among them, the back propagation algorithm and optimizer are used to adjust the parameters of the fence status detection model, including:

[0034] Obtain a preset airport fence standard image dataset, and extract multiple image features from the image dataset;

[0035] Input the extracted multiple image features into the fully connected layer to integrate and classify the image features;

[0036] Obtain a preset image dataset that annotates the normal state and abnormal state of the fence, and use the image dataset as a training set to train the fence state detection model;

[0037] During the training process, the back propagation algorithm is used to calculate the error of the fence state detection model, and the parameters of the fence state detection model are updated according to the error;

[0038] Obtain a preset optimizer, select the optimal parameter adjustment strategy from the optimizer, and iteratively adjust the parameters of the fence status detection model;

[0039] Repeat the back propagation algorithm and optimizer parameter adjustment steps until the error value of the fence status detection model is less than the preset error threshold;

[0040] Obtain the trained fence status detection model and verify the detection accuracy of the fence status detection model under different lighting conditions, angles, and weather conditions.

[0041] Among them, the standardized image data is input into the model to be detected for real-time detection, including:

[0042] Inputting the first frame of standardized image data into the model to be detected to obtain a first detection result;

[0043] Starting from the second frame of standardized image data, the detection result of the previous frame and the standardized image data of the current frame are input into the model to be detected to obtain the detection result of the current frame;

[0044] Integrate the first detection result and the detection results of all current frames to obtain a fence status detection result;

[0045] According to the fence status detection results, the status of each area of ​​the airport is detected, including:

[0046] Acquire airport fence standard image data, and extract airport area division information from the airport fence standard image data;

[0047] According to the airport area division information, the fence status detection results are mapped to the corresponding airport areas;

[0048] Evaluate the fence status of each airport area and obtain the airport area status evaluation results;

[0049] Generate an airport safety status report based on the airport area status assessment results.

[0050] Among them, multiple new fence status detection results are determined and output, including:

[0051] Identifying and extracting fence status detection data from the standardized image data, wherein the fence status detection data includes fence detection parameters and image environment parameters;

[0052] According to the fence detection parameters and image environment parameters, the fence status detection data is processed in layers;

[0053] Compare the airport fence standard image data with the fence status detection data to determine the fence status;

[0054] Compare and analyze each new fence status detection result, including normal status results and abnormal status results, match the new fence status detection results with the standard status in the airport fence standard image data one by one, and output the status of each detection result.

[0055] Among them, determining the fence status detection result received by the unmanned vehicle terminal or the remote monitoring center and transmitting it to the alarm mechanism includes:

[0056] Obtain the image data and transmission time of the fence status detection result received by the unmanned vehicle terminal or the remote monitoring center. The transmission time is the time between the output of the current fence status detection result and the acquisition of the alarm signal;

[0057] Determine the current fence status detection result based on the image data and transmit it to the alarm mechanism, the detection result includes normal status result and abnormal status result;

[0058] If the detection result is an abnormal state result, an alarm mechanism is triggered based on the transmission time and image data;

[0059] If the test result is normal, continue to monitor and save the test result.

[0060] Among them, according to the fence detection parameters and image environment parameters, the fence status detection data is layered processed, including:

[0061] Get the environment parameter type from the image environment parameter, and get the fence detection parameter type from the fence detection parameter;

[0062] Obtain the fence detection parameter type associated with each environmental parameter type as a data group to be processed;

[0063] After sorting each data group to be processed according to the environmental parameter type, the data corresponding to each environmental parameter type and fence detection parameter type is stored in the corresponding data group to be processed, and the fence status detection data is obtained.

[0064] Compared with the prior art, the present invention has the following advantages:

[0065] A method for detecting the status of airport fences on an unmanned vehicle based on machine learning includes: obtaining the original image data of the airport fence collected by the unmanned vehicle, preprocessing the original image data of the airport fence, and obtaining standardized image data; constructing a fence status detection model based on a deep learning algorithm; deploying the fence status detection model to the unmanned vehicle, and obtaining the fence status detection result by real-time detection of the standardized image data; displaying the fence status detection result through the display screen of the unmanned vehicle or the remote monitoring center, and triggering an alarm mechanism to notify relevant personnel when an abnormal state is detected. It can automatically, accurately and efficiently detect the status of the airport fence, solve the problems existing in traditional manual inspections, and provide strong technical support for ensuring airport safety.

[0066] Other features and advantages of the present invention will be set forth in the description which follows, and in part will be apparent from the description, or may be learned by practice of the present invention.

[0067] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0069] Figure 1 It is a flow chart of an airport fence status detection image method based on machine learning on an unmanned vehicle-mounted terminal in an embodiment of the present invention;

[0070] Figure 2 A flowchart of obtaining standardized image data in an embodiment of the present invention;

[0071] Figure 3 This is a flow chart of constructing a fence status detection model in an embodiment of the present invention. DETAILED DESCRIPTION

[0072] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0073] The embodiment of the present invention provides an airport fence status detection image method based on machine learning on an unmanned vehicle, comprising:

[0074] S101: acquiring original image data of the airport fence collected by the unmanned vehicle terminal, preprocessing the original image data of the airport fence, and acquiring standardized image data;

[0075] S102: Building a fence status detection model based on deep learning algorithm;

[0076] S103: deploying the fence status detection model to the unmanned vehicle, and obtaining the fence status detection result by performing real-time detection on the standardized image data;

[0077] S104: The fence status detection result is displayed on the display screen of the unmanned vehicle or the remote monitoring center. When an abnormal state is detected, an alarm mechanism is triggered to notify relevant personnel.

[0078] The working principle of the above technical solution is as follows: data collection and preprocessing. First, the visible light and thermal imaging high-definition cameras on the unmanned vehicle are used to collect images of the airport fence. In order to ensure the generalization ability of the model, the collected images should include fence images under different lighting conditions, different angles, and different weather conditions. The collected image data needs to be preprocessed, including image cropping, scaling, normalization and other operations, so as to facilitate the subsequent machine learning model processing; build a fence state detection model, and use deep learning technology to build a fence state detection model. Specifically, a convolutional neural network (CNN) can be used as the basic model, and image features are extracted and classified by stacking multiple convolutional layers, pooling layers and fully connected layers. In order to improve the performance of the model, advanced technologies such as residual structure and attention mechanism can also be introduced. In the process of model training, a large amount of annotated fence image data is used as the training set. The annotation information should include the normal state and various abnormal states of the fence (such as damage, deformation, displacement, etc.). The model is trained by back propagation algorithm and optimizer (such as Adam, SGD, etc.) so that it can accurately identify various states of the fence. The trained machine learning model needs to be deployed on the unmanned vehicle for real-time reasoning in practical applications. In order to ensure the speed and accuracy of reasoning, the model can be compressed and pruned to reduce model parameters and calculations. In the unmanned vehicle process, the fence image is captured in real time by a dual-spectrum high-definition camera and input into the deployed model for reasoning to obtain the fence status detection result. The detection results can be displayed on the display screen of the unmanned vehicle or the remote monitoring center. Once an abnormal state of the fence is found, the system immediately triggers the alarm mechanism and notifies the relevant personnel to handle it. At the same time, the system can also record and save the image data of the abnormal state for subsequent analysis and tracing.

[0079] The beneficial effects of the above technical solution are: by proposing an airport fence status detection image method based on machine learning for an unmanned vehicle, this method can automatically, accurately and efficiently detect the airport fence status, solve the problems existing in traditional manual inspections, and provide strong technical support for ensuring airport safety.

[0080] In another embodiment, step S101 includes:

[0081] S1011: Capture images of the airport fence using a dual-spectrum high-definition camera mounted on the unmanned vehicle to obtain original image data of the airport fence. The original image data includes fence images captured under different lighting conditions, different angles, and different weather conditions.

[0082] Obtaining original image data of the airport fence, including: controlling the dual-spectrum high-definition camera on the unmanned vehicle to collect images of the airport fence and obtaining original image data of the airport fence;

[0083] Acquire fence images under different lighting conditions, including image data during daytime and nighttime, strong light and weak light conditions;

[0084] Acquire fence images at different angles, including image data taken at horizontal, vertical, and oblique angles;

[0085] Obtain fence images under different weather conditions, including image data under sunny, cloudy, rainy, foggy and snowy conditions;

[0086] Control the camera to collect images under complex environmental changes and adjust camera parameters to cope with changes in lighting, angles and weather;

[0087] Classify and mark the collected raw image data, and identify the image data under different lighting conditions, different angles and different weather conditions;

[0088] S1012: pre-processing the original image data of the airport fence, cropping the original image of the airport fence according to a preset cropping parameter, scaling the cropped image according to a preset scaling ratio, and normalizing the scaled image data;

[0089] S1013: Acquire the preprocessed standardized image data, and save the standardized image data into an image database.

[0090] The working principle of the above technical solution is: a dual-spectrum high-definition camera is installed on the unmanned vehicle, which can capture visible light and infrared spectrum images at the same time, thereby providing comprehensive monitoring capabilities of the airport fence under different lighting conditions. The unmanned vehicle drives around the airport fence and collects image data of the fence through the camera. During the collection process, the camera will obtain fence images under different lighting conditions (such as daytime, night, cloudy days, etc.), different angles (such as front, side, top view, etc.) and different weather conditions (such as sunny days, rainy days, foggy days, etc.) to ensure the diversity and comprehensiveness of the data.

[0091] Image preprocessing: The original image is cropped according to the preset cropping parameters. The purpose of cropping is to remove irrelevant parts of the image and only retain the key areas of the fence. The cropping parameters include the boundary coordinates of the image, the width and height of the crop, etc. After cropping, the image is scaled according to the preset scaling ratio. This step ensures that all images are of the same size, which facilitates subsequent analysis and processing. Scaling is performed using the following formula:

[0092] I scaled (x,y)=I cropped (αx,βy)

[0093] Among them, I scaled is the scaled image, I cropped is the cropped image, α and β are the scaling ratios of the x-axis and y-axis respectively.

[0094] The scaled image data is normalized to reduce the impact of illumination changes and camera noise. The preprocessed standardized image data is saved in the image database according to a specific naming and classification method. After the standardized processing, these image data can be effectively retrieved and analyzed by the system.

[0095] Among them, the original image data of the airport fence is preprocessed, including:

[0096] Recognize and obtain current fence status data from the original image data of the airport fence, wherein the current fence status data includes fence structure parameters and ambient lighting parameters;

[0097] The beneficial effects of the above technical solution are as follows: through the use of dual-spectrum high-definition cameras, the system can collect high-quality image data under different lighting conditions, angles and weather conditions, covering a variety of fence monitoring scenarios, thereby improving the diversity and practicality of the data; through cropping and scaling, irrelevant parts of the image are removed, and the size of all images is unified, laying the foundation for subsequent automated analysis. Normalization processing further reduces the impact of environmental factors on image data and improves the consistency and contrast of image data; standardized image data is stored in the database in an orderly manner, which is convenient for subsequent rapid retrieval and analysis. This efficient data management method ensures that image data can quickly respond to monitoring needs and improves the response speed and monitoring accuracy of the overall system.

[0098] In another embodiment, step S102 includes:

[0099] S1021: Use deep learning algorithms to perform preliminary feature extraction on standardized image data. In the deep learning algorithm, multiple convolutional layers are stacked to extract low-level and high-level features. After the convolutional layer, a pooling layer is added to perform feature dimension reduction. In the feature extraction process, a residual structure is introduced to solve the gradient vanishing problem in model training, and an attention mechanism is introduced to enhance the model's ability to recognize key feature areas.

[0100] S1022: Input the extracted image features into the fully connected layer for feature integration and classification, use a number of image data labeled with normal and abnormal fence states as training sets, train the fence state detection model, and obtain the trained fence state detection model. During the training process, use the back propagation algorithm and optimizer to adjust the parameters of the fence state detection model.

[0101] The working principle of the above technical solution is as follows: First, the convolutional neural network (CNN) in deep learning is used to extract features from the standardized image data. The convolution layer slides the image through the convolution kernel (filter) to extract low-level features (such as edges, textures, etc.) and high-level features (such as shape, structure, etc.) of the image. The stacking of multiple convolutional layers enables the model to extract more and more complex features from the image layer by layer. The convolution operation is expressed by the following formula:

[0102]

[0103] Among them, X is the input image, K is the convolution kernel, and Y is the output feature map.

[0104] After the convolutional layer, a pooling layer (usually a max pooling layer) is added to perform feature dimensionality reduction. The role of the pooling layer is to reduce the size of the feature map by taking the maximum value (or average value) of the local area, thereby reducing the amount of data and computational complexity while retaining important features. The pooling operation is expressed by the following formula:

[0105]

[0106] Among them, X is the input feature map and Y is the output after pooling.

[0107] In order to avoid the problem of gradient vanishing when the model increases in depth, the residual structure (ResNet) is introduced. The residual structure introduces "shortcut connections" between layers, so that the input can bypass certain layers and pass directly to the output, thereby alleviating the problem of gradient vanishing in deep networks. The output of the residual structure is expressed as:

[0108] Y=F(X)+X

[0109] Among them, F(X) is the output of the convolutional layer and X is the input feature.

[0110] In order to enhance the model's ability to identify key feature areas in the image, the attention mechanism is introduced. The attention mechanism assigns weights to different areas, allowing the model to pay more attention to the key areas in the image. Specifically, the attention mechanism weights the feature map, and the formula is as follows:

[0111] Y attn =αF(X)

[0112] Among them, α is the weight matrix calculated by the attention mechanism, and F(X) is the feature map.

[0113] The features extracted from the convolutional layer and the pooling layer are input into the fully connected layer. The fully connected layer integrates these features and generates the final classification result. The output of the fully connected layer is expressed as:

[0114] Y fc =W×X+b

[0115] Among them, W is the weight matrix, X is the input feature, and b is the bias.

[0116] Several airport fence image data marked as normal and abnormal states are used as training sets to train the fence state detection model. During the training process, the backpropagation algorithm is used to calculate the gradient of the loss function relative to each parameter, and the optimizer (such as Adam or SGD) is used to adjust the model parameters to gradually reduce the value of the loss function. The formula is as follows:

[0117]

[0118] Among them, θ is the model parameter, η is the learning rate, and L(θ) is the loss function.

[0119] After training is completed, the optimized model is saved and deployed in the actual fence monitoring system to detect the status of the airport fence in real time.

[0120] In the deep learning algorithm model, multiple convolutional layers are stacked, including:

[0121] Extract low-level features from the first convolutional layer;

[0122] The low-level features are fed into the subsequent convolutional layers to extract high-level features;

[0123] Perform feature mapping on the output of the convolutional layer to obtain a convolutional feature map.

[0124] Add a pooling layer after the convolutional layer, including:

[0125] According to the preset pooling parameters, the convolution feature map is pooled;

[0126] Perform feature dimensionality reduction processing on the pooled feature map to obtain the reduced-dimensional feature data.

[0127] In the feature extraction process, the residual structure is introduced, including:

[0128] Add the input data of the previous layer and the output data of the next layer to get the residual output;

[0129] The residual output is used as the input of the next layer to continue the convolution and pooling operations.

[0130] The attention mechanism is introduced in the feature extraction process, including:

[0131] According to the preset attention mechanism model, feature weighting is performed on the feature data after dimensionality reduction;

[0132] The weight of the key feature area in the feature data is strengthened to obtain the enhanced feature data.

[0133] The beneficial effects of the above technical solution are as follows: through multi-layer convolution operations, the system can effectively extract low-level and high-level features of airport fence images. The introduction of residual structure solves the common gradient vanishing problem in deep networks, and the attention mechanism enhances the model's ability to recognize key areas in the image, thereby improving the accuracy of feature extraction; the fully connected layer integrates and classifies the extracted image features, so that the system can accurately distinguish between normal and abnormal states of the fence. By using annotated training sets for supervised learning, the model continuously optimizes parameters during the training process, and finally obtains a high-precision fence status detection model; the trained model can be deployed in the actual fence monitoring system to achieve real-time monitoring of the airport fence. Once an abnormal situation is detected, the system can respond quickly, improving the safety and operational efficiency of the airport.

[0134] In another embodiment, obtaining the fence status detection result includes:

[0135] Compressing the fence state detection model to obtain compressed fence state detection model data;

[0136] Pruning the compressed fence state detection model data to reduce model parameters and calculation amount, and obtaining a fence state detection model after pruning;

[0137] Deploy the pruned fence status detection model to the unmanned vehicle to generate a model to be detected;

[0138] Obtain fence image data captured in real time by the dual-spectrum high-definition camera on the unmanned vehicle, and pre-process the fence image data as standardized image data;

[0139] Input the standardized image data into the model to be detected for real-time detection;

[0140] Get fence status detection results from real-time detection.

[0141] The working principle of the above technical solution is as follows: First, the trained fence status detection model is compressed. The purpose of model compression is to reduce the storage space and computing resource usage of the model while retaining the accuracy of the model as much as possible. Common compression methods include weight quantization, low-rank decomposition, etc. The compressed model data is more suitable for running in resource-constrained environments, such as unmanned vehicle terminals. Based on the compressed model, the model is further pruned. The purpose of pruning is to further reduce the number of parameters and the amount of calculation of the model by deleting redundant network connections or neurons. This usually involves deleting weights or neurons that have little effect on the final output. The pruned model is more computationally efficient and suitable for real-time applications. The pruning operation can be achieved by adjusting the sparsity in the model weight matrix, and the formula is expressed as:

[0142] W′=P(W)

[0143] Among them, W is the original weight matrix, P is the pruning operation, and W′ is the pruned weight matrix.

[0144] The fence status detection model that has been compressed and pruned is deployed to the computing unit of the unmanned vehicle. A dual-spectrum high-definition camera is installed on the unmanned vehicle to capture real-time image data of the fence. Because the model has been optimized, it can run efficiently under the limited computing resources of the on-board equipment. The dual-spectrum high-definition camera on the unmanned vehicle captures image data of the airport fence in real time. The dual-spectrum camera can capture visible light and infrared spectrum images at the same time, thereby improving the accuracy and robustness of detection. The acquired image data needs to be preprocessed to be converted into a standardized input format. The preprocessing steps may include graying, resizing, and noise removal of the image. The standardized image data is more suitable for input into the deep learning model for analysis. The preprocessed standardized image data is input into the model to be detected. The model to be detected is based on the convolutional neural network structure and can analyze the input fence image data in real time to detect the status of the fence (normal or abnormal). The model to be detected will output the detection results based on the input image data to determine the status of the fence. If an abnormal fence is detected, the system can immediately issue an alarm so that corresponding countermeasures can be taken.

[0145] The beneficial effects of the above technical solution are as follows: by compressing and pruning the fence status detection model, the number of parameters and the amount of calculation of the model are greatly reduced, so that it can run efficiently on the resource-constrained unmanned vehicle side and maintain good real-time detection performance. The optimized model can process the fence image data captured by the dual-spectrum camera in real time on the unmanned vehicle side and detect the status of the fence in time. This real-time detection capability ensures the safe monitoring of the airport fence and can respond quickly to potential security threats. The image data captured by the dual-spectrum high-definition camera is used, combined with standardized preprocessing steps, to ensure the data quality of the input model. This improves the accuracy and robustness of the detection, especially under complex environmental conditions. Through model compression and pruning, the system effectively utilizes the limited computing resources of the unmanned vehicle side, realizes efficient fence status detection, and provides strong technical support for the airport's security monitoring.

[0146] In another embodiment, the fence status detection result is displayed on a display screen of an unmanned vehicle or a remote monitoring center, including:

[0147] Determine the fence status detection result output by the display screen of the unmanned vehicle and transmit it to the remote monitoring center. The detection result includes normal status result and abnormal status result;

[0148] Determine and output a plurality of new fence status detection results, determine the status of each new detection result and transmit it to the remote monitoring center until the number of new abnormal status results reaches a preset number threshold, or the current detection status result is determined to be the last detection result;

[0149] Control the unmanned vehicle to record and save the image data of abnormal status, and return to execute the output of the initial fence status detection result through the display screen or the remote monitoring center until the number of new abnormal status results reaches the preset number threshold, or the current detection status result is determined to be the last detection result;

[0150] Generate and output a list based on each test result and the status of the test result;

[0151] Determine the fence status detection results received by the unmanned vehicle terminal or remote monitoring center and transmit them to the alarm mechanism.

[0152] The working principle of the above technical solution is as follows: the display screen of the unmanned vehicle is used to display the fence status detection results in real time, including the results of normal state and abnormal state. When the detection is completed, the detection results will be immediately displayed on the screen of the vehicle for the operator or the automatic system to view. At the same time, the detection results are transmitted to the remote monitoring center through the wireless communication module. The remote monitoring center can receive the fence status detection results sent by the unmanned vehicle in real time, thereby realizing remote monitoring and management. The system continuously generates new fence status detection results on the unmanned vehicle. Whenever a new detection result is generated, the system will immediately determine its status (normal or abnormal) and transmit the result to the remote monitoring center. The system will continue to detect the fence status until the number of abnormal status results detected reaches the preset number threshold, or when it is determined that the current detection result is the last detection result, the system will stop the output and transmission of new detection results. Whenever an abnormal state of the fence is detected, the unmanned vehicle will automatically record and save the image data corresponding to the abnormal state. These image data contain abnormal conditions captured by the dual-spectrum high-definition camera, which can be used for further analysis and evidence preservation. While recording and saving abnormal state image data, the system will continue to output the initial fence state detection results through the display screen or remote monitoring center until the number of new abnormal states reaches the preset threshold, or the current detection state is the last detection result. The system generates a result list based on each detection result and its corresponding state (normal or abnormal). The list is arranged in chronological order and contains the state changes of all detections. The generated detection result list will be displayed on the vehicle end and transmitted to the remote monitoring center, so that the operator can view all historical detection data. The system will continue to monitor the fence state detection results received by the unmanned vehicle end or the remote monitoring center. Once an abnormal state result is detected and the preset alarm condition (such as continuous abnormal detection or severe abnormality) is met, the system will automatically trigger the alarm mechanism. The alarm signal is sent to the remote monitoring center through the wireless communication module, and the sound and light alarm on the unmanned vehicle end can also be used to sound an alarm to remind relevant personnel to deal with it immediately.

[0153] The beneficial effects of the above technical solution are: through the real-time reception and display of the unmanned vehicle-mounted display screen and the remote monitoring center, the system can monitor the fence status in real time and promptly alarm when an abnormality is detected to ensure the safety of the airport fence. The system has a continuous detection function and can continuously generate and transmit new fence status detection results. Through the preset abnormal threshold control, the system can take immediate action when necessary to avoid potential safety hazards. The image data in the abnormal state is automatically recorded and saved, providing important on-site evidence, which is helpful for post-analysis and problem tracing. This is of great significance for airport safety management. The generation and output of the test result list facilitates operators to view and manage historical test data. Through the list, the changing trend of the fence status can be tracked, which is helpful for comprehensive analysis and decision-making. The remote monitoring center can receive the test results of the unmanned vehicle-mounted terminal in real time, realize remote management and control, reduce the dependence on manual inspections, and improve the automation of safety management.

[0154] In another embodiment, a back propagation algorithm and an optimizer are used to adjust the parameters of the fence status detection model, including:

[0155] Obtain a preset airport fence standard image dataset, and extract multiple image features from the image dataset;

[0156] Input the extracted multiple image features into the fully connected layer to integrate and classify the image features;

[0157] Obtain a preset image dataset that annotates the normal state and abnormal state of the fence, and use the image dataset as a training set to train the fence state detection model;

[0158] During the training process, the back propagation algorithm is used to calculate the error of the fence state detection model, and the parameters of the fence state detection model are updated according to the error;

[0159] Obtain a preset optimizer, select the optimal parameter adjustment strategy from the optimizer, and iteratively adjust the parameters of the fence status detection model;

[0160] Repeat the back propagation algorithm and optimizer parameter adjustment steps until the error value of the fence status detection model is less than the preset error threshold;

[0161] Obtain the trained fence status detection model and verify the detection accuracy of the fence status detection model under different lighting conditions, angles, and weather conditions.

[0162] The working principle of the above technical solution is as follows: First, multiple images representing the normal and abnormal states of the airport fence are extracted from the preset airport fence standard image dataset. These image datasets are collected based on the actual airport fence environment, including images under various conditions such as daytime, nighttime, sunny days, and rainy days. A convolutional neural network (CNN) is used to extract multiple features of the image. The feature extraction process involves multiple convolutional layers and pooling layers, which are responsible for capturing key features such as edges and textures in the fence image, and finally forming a set of representative image feature vectors. The extracted image feature vectors are input into one or more fully connected layers. In the fully connected layer, these features are further integrated and processed to form a comprehensive feature representation for classification. Through the last fully connected layer, the system will classify the integrated image features to determine whether the fence state shown in the image is normal or abnormal. Image data with the normal and abnormal states of the fence annotated are obtained from the preset image dataset as a training set. These annotated datasets are used to guide the model to identify what is a normal state and what is an abnormal state. The fence state detection model is trained using the training set data. During the training process, the system uses the image's feature vectors and the corresponding annotated data to adjust the model's parameters to improve its classification accuracy. In each training iteration, the system calculates the predicted results of the model output through forward propagation, compares them with the true label (i.e., the annotated fence status), and calculates the error value. Common error calculation methods include mean square error (MSE). Based on the error value, the system calculates the gradient of each parameter through the backpropagation algorithm and updates the model's parameters to minimize the error. The system selects the optimal parameter adjustment strategy (such as Adam optimizer, SGD, etc.) from the preset optimizer to further improve the training efficiency and accuracy of the model. The model parameters are iteratively adjusted using the selected optimizer, and backpropagation and optimizer adjustments are repeatedly performed until the model's error value drops below the preset error threshold, indicating that the model has converged and achieved good detection results. After obtaining the trained fence status detection model, the system will verify it under different lighting conditions, angles, and weather conditions to ensure the detection accuracy of the model in the actual application environment. For example, tests are conducted in strong light, low light, different angles, and rainy and snowy weather to verify the robustness and accuracy of the model.

[0163] The beneficial effects of the above technical solution are as follows: the fence status detection model trained through the above steps can achieve high-precision detection under various environmental conditions and improve the safety monitoring capabilities of airport fences. Using machine learning technology, the system can automatically detect and judge the status of the fence, reducing the reliance on manual inspections and greatly improving the automation and intelligence level of airport security management. By verifying the model under different lighting, angles and weather conditions, ensuring its stability and accuracy in complex environments, the system can cope with various challenges in practical applications. Using backpropagation and optimizer strategies, the model parameters can be adjusted quickly and effectively to ensure that the training process is efficient and the final model has excellent performance.

[0164] In another embodiment, the standardized image data is input into the model to be detected, and real-time detection is performed, including:

[0165] Inputting the first frame of standardized image data into the model to be detected to obtain a first detection result;

[0166] Starting from the second frame of standardized image data, the detection result of the previous frame and the standardized image data of the current frame are input into the model to be detected to obtain the detection result of the current frame;

[0167] Integrate the first detection result and the detection results of all current frames to obtain a fence status detection result;

[0168] According to the fence status detection results, the status of each area of ​​the airport is detected, including:

[0169] Acquire airport fence standard image data, and extract airport area division information from the airport fence standard image data;

[0170] According to the airport area division information, the fence status detection results are mapped to the corresponding airport areas;

[0171] Evaluate the fence status of each airport area and obtain the airport area status evaluation results;

[0172] Generate an airport safety status report based on the airport area status assessment results.

[0173] The working principle of the above technical solution is as follows: First, obtain the first frame image data of the airport fence and standardize it to ensure data consistency between different frames. Standardization includes steps such as size adjustment and brightness equalization. The standardized first frame image data is input into the pre-trained fence status detection model, and the model analyzes the image and outputs preliminary detection results. Starting from the second frame image, when the system performs detection, in addition to inputting the standardized image data of the current frame, it also inputs the detection results of the previous frame into the model. By combining historical detection results with the current frame data, the model can more accurately judge the fence status of the current frame. The model outputs the detection results of the current frame and integrates them with the previous results to form a complete fence status detection sequence.

[0174] The system integrates the preliminary detection results of the first frame with the detection results of all current frames to generate a comprehensive fence status detection result. During the integration process, the system will consider the temporal correlation between each frame to improve the accuracy of the overall judgment. The division information of each area of ​​the airport is extracted from the standardized airport fence image data. This information includes the location, boundaries, etc. of different areas, which is convenient for subsequent status mapping.

[0175] The comprehensive fence status detection results are mapped to the corresponding airport areas. The status of each area will be updated according to the fence detection situation, ensuring that the system can reflect the actual security status of each area.

[0176] Status assessment: Conduct a detailed assessment of the fence status of each airport area, including determining whether there are any anomalies or potential safety hazards. The assessment results are generated for each area to ensure that the overall safety status of the airport can be reflected in a targeted manner. Safety status report generation: Based on the assessment results, the system generates an airport safety status report, which lists the safety status of each area and the corresponding recommended measures in detail to provide decision support for airport managers.

[0177] The beneficial effects of the above technical solution are: by combining each frame of the image with the detection results of the previous frame and inputting them into the model, the system can capture subtle changes in the image and improve the accuracy of detection, especially in complex or changing environments. The system can not only detect abnormal conditions of the fence, but also combine these results with the airport area division information to provide accurate status assessments for each area, providing strong support for security management. By integrating multi-frame detection results and regional assessments, the security status report generated by the system can fully reflect the security status of the airport and help managers to promptly discover and deal with potential safety hazards. The entire process is highly automated, reducing the need for manual intervention and greatly improving the efficiency and intelligence level of airport security monitoring. The system has strong dynamic adaptability and can maintain high-precision fence status detection under conditions such as lighting, angles and weather changes, ensuring the reliability and continuity of airport security monitoring.

[0178] In another embodiment, multiple new fence status detection results are determined and output, including:

[0179] Identifying and extracting fence status detection data from the standardized image data, wherein the fence status detection data includes fence detection parameters and image environment parameters;

[0180] According to the fence detection parameters and image environment parameters, the fence status detection data is processed in layers;

[0181] Compare the airport fence standard image data with the fence status detection data to determine the fence status;

[0182] Compare and analyze each new fence status detection result, including normal status results and abnormal status results, match the new fence status detection results with the standard status in the airport fence standard image data one by one, and output the status of each detection result.

[0183] The working principle of the above technical solution is as follows: Obtain standardized image data: The system first obtains standardized image data of the airport fence. These data are pre-processed, including size adjustment, color correction, etc., to ensure data consistency and comparability. Fence status detection data is identified and extracted from the standardized image data. The detection data includes two main parts: fence detection parameters and image environment parameters. These parameters include the location, shape, integrity, etc. of the fence, reflecting the physical state of the fence.

[0184] Image environment parameters: These parameters include light intensity, weather conditions, camera angle, etc., which reflect the external environmental factors of the image. Data layering processing: According to the extracted fence detection parameters and image environment parameters, the system performs layered processing on the fence status detection data. Specific layering operations include: Basic layer processing: First, the basic physical state parameters of the fence are processed, such as the shape of the fence, whether the boundary is complete, etc. Environmental layer processing: Then, the environmental parameters of the image are processed. These parameters may affect the accuracy of fence detection, such as shadows caused by changes in lighting, changes in camera angles, etc. Layered result combination: The system combines the processing results of each layer to form an overall fence status detection result.

[0185] Standard image data comparison: compare the fence status detection data after the current frame processing with the standard image data of the airport fence. The standard image data represents the normal state of the fence, including the fence detection parameters and image environment parameters under normal conditions. Status judgment: through comparison, the system can judge whether the current fence status is normal. If the detection result matches the standard image data, the fence status is considered normal; if it does not match, it is marked as abnormal.

[0186] Frame-by-frame comparison and analysis: For each new fence status detection result, the system will compare and analyze it with the standard status in the standard image data one by one. The purpose of the analysis is to confirm whether the current fence status is normal or abnormal. Status output: Based on the results of the comparison and analysis, the system outputs the specific status of each detection result, including whether there is an abnormality, the type of abnormality, and potential safety hazards.

[0187] The beneficial effects of the above technical solution are as follows: by layering the fence detection parameters and environmental parameters, the system can more accurately identify the state of the fence, especially in the case of complex environmental changes, to ensure the reliability of the detection results. By using standard image data for comparative analysis, the system can quickly identify the abnormal state of the fence, reduce the possibility of misjudgment, and improve the safety and accuracy of airport fence monitoring. The system can compare and analyze new fence status detection results frame by frame, promptly discover and report abnormal states, and ensure that the safety of the airport fence is continuously monitored and maintained. The system not only outputs the status of each detection result, but also provides a detailed abnormal analysis report to help airport managers quickly take corresponding measures and improve the overall safety management level of the airport. The system has good dynamic adaptability, can still maintain high-precision detection and status judgment under different environmental conditions, and provides real-time fence safety monitoring capabilities.

[0188] In another embodiment, determining the fence status detection result received by the unmanned vehicle terminal or the remote monitoring center and transmitting it to the alarm mechanism includes:

[0189] Obtain the image data and transmission time of the fence status detection result received by the unmanned vehicle terminal or the remote monitoring center. The transmission time is the time between the output of the current fence status detection result and the acquisition of the alarm signal;

[0190] Determine the current fence status detection result based on the image data and transmit it to the alarm mechanism, the detection result includes normal status result and abnormal status result;

[0191] If the detection result is an abnormal state result, an alarm mechanism is triggered based on the transmission time and image data;

[0192] If the test result is normal, continue to monitor and save the test result.

[0193] The working principle of the above technical solution is as follows: Receiving image data: The unmanned vehicle or remote monitoring center receives the image data of the fence status in real time through a camera or other sensor. These data are transmitted to the monitoring center through a wireless network or a wired network. Recording transmission time: When the system receives the image data of the detection result, it automatically records the transmission time. The transmission time refers to the time interval from the generation of the detection result to the acquisition of the image data and the completion of the analysis by the monitoring center.

[0194] Image data processing: The monitoring center analyzes the received image data and determines the current state of the fence through a preset algorithm. The detection results are divided into normal state and abnormal state. Detection result transmission: Once the fence status detection result is determined, the system transmits the result to the alarm mechanism. If it is an abnormal state, the system will further trigger an alarm.

[0195] Abnormal state processing: If the detection result is abnormal, the system will trigger an alarm mechanism based on the transmission time and image data. For example, when the fence is damaged or illegally invaded, the system will immediately send out an alarm signal to notify relevant personnel to take action. Normal state processing: When the detection result is normal, the system will continue to monitor the status of the fence and save the current detection result for subsequent analysis and comparison.

[0196] Monitoring loop: The system will continuously cycle the above process and continuously monitor the fence status. Each newly detected status will be recorded and analyzed to ensure that the fence is safe at all times. Abnormal status response: When an abnormal status is detected and the alarm mechanism is triggered, the system will issue different levels of alarms according to the specific situation so that you can respond quickly and take necessary measures.

[0197] The beneficial effects of the above technical solution are: the system can detect and analyze the changes in the fence status at the first time, and quickly trigger the alarm mechanism in the abnormal state. This ensures the safety of the airport fence and can respond to potential security threats in a timely manner. By recording the transmission time and real-time analysis of image data, the system can effectively distinguish between normal and abnormal states, thereby reducing the probability of false alarms and missed alarms and improving the accuracy of detection. The system can continuously monitor the fence status and save all detection results. Continuous monitoring data under normal conditions can be used for long-term analysis to help optimize the system's detection algorithm and improve adaptability to environmental changes. The alarm mechanism is based on actual transmission time and image data to ensure the timeliness and pertinence of the alarm. Different abnormal situations can trigger different levels of alarms, providing more detailed and intelligent security management.

[0198] In another embodiment, the fence status detection data is hierarchically processed according to the fence detection parameters and the image environment parameters, including:

[0199] Get the environment parameter type from the image environment parameter, and get the fence detection parameter type from the fence detection parameter;

[0200] Obtain the fence detection parameter type associated with each environmental parameter type as a data group to be processed;

[0201] After sorting each data group to be processed according to the environmental parameter type, the data corresponding to each environmental parameter type and fence detection parameter type is stored in the corresponding data group to be processed, and the fence status detection data is obtained.

[0202] The working principle of the above technical solution is: through the environmental sensors (such as temperature sensors, humidity sensors, light sensors, etc.) deployed around the airport fence, the system obtains the types of environmental parameters related to the fence status in real time. For example, temperature, humidity, wind speed, etc. At the same time, the fence detection system (such as infrared detectors, vibration sensors, etc.) obtains the fence status data, such as vibration frequency, stress changes, current intensity, etc.

[0203] According to the actual environmental conditions, each environmental parameter type is associated with the corresponding fence detection parameter type. For example, temperature affects the stress change of the fence material, and humidity may affect the current strength of the fence. In this way, the system will find the associated fence detection parameter for each environmental parameter to form a data group to be processed. Data sorting: For each data group to be processed, the system sorts according to the degree of influence of the environmental parameter type (such as the influence of temperature on stress). After sorting, the data corresponding to each environmental parameter type and fence detection parameter type is stored in the corresponding data group. Data group storage: These sorted data are stored as fence status detection data for subsequent analysis and status judgment. This step ensures the systematic and logical nature of data processing. The system forms the final fence status detection data by combining the data of environmental parameters and fence detection parameters. These data are used to determine whether the fence is in a normal state or whether there are potential risks.

[0204] Assume that the airport fence system is running in winter, when the ambient temperature (environmental parameter type) is low, which may cause the stress of the fence metal material (fence detection parameter type) to increase. The system first obtains the current temperature value and stress value, then associates the temperature data with the corresponding stress data, sorts them and stores them as a data group to be processed. The system analyzes these data and finds that the stress value exceeds the predetermined threshold, thereby generating an abnormal fence detection result and triggering an alarm mechanism.

[0205] The beneficial effects of the above technical solution are: by associating environmental parameters with fence detection parameters, the system can comprehensively analyze the fence status, avoid misjudgment due to ignoring environmental influences, and improve the accuracy of detection; by sorting and grouping environmental parameters and fence detection parameters, the system can more effectively manage and process data to ensure the stability and reliability of detection results; the system can dynamically adjust parameter associations according to actual environmental conditions, which enables it to accurately detect the fence status under different weather conditions and enhance the flexibility and adaptability of the system; by comprehensively analyzing the detection data of the environment and fence, the system can identify potential risks earlier, issue warnings in advance, and avoid the occurrence of emergencies.

[0206] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. An airport fence status detection image method based on machine learning for an unmanned vehicle, characterized in that: include: S101: acquiring original image data of the airport fence collected by the unmanned vehicle terminal, preprocessing the original image data of the airport fence, and acquiring standardized image data; S102: Building a fence status detection model based on deep learning algorithm; S103: deploying the fence status detection model to the unmanned vehicle, and obtaining the fence status detection result by performing real-time detection on the standardized image data; S104: The fence status detection result is displayed on the display screen of the unmanned vehicle or the remote monitoring center. When an abnormal state is detected, an alarm mechanism is triggered to notify relevant personnel.

2. The method for detecting airport fence status based on machine learning on an unmanned vehicle according to claim 1 is characterized in that: Step S101 includes: S1011: Capture images of the airport fence using a dual-spectrum high-definition camera mounted on the unmanned vehicle to obtain original image data of the airport fence. The original image data includes fence images captured under different lighting conditions, different angles, and different weather conditions. S1012: pre-processing the original image data of the airport fence, cropping the original image of the airport fence according to a preset cropping parameter, scaling the cropped image according to a preset scaling ratio, and normalizing the scaled image data; S1013: Acquire the preprocessed standardized image data, and save the standardized image data into an image database.

3. The method for detecting airport fence status based on machine learning on an unmanned vehicle according to claim 1 is characterized in that: Step S102 includes: S1021: Use deep learning algorithms to perform preliminary feature extraction on standardized image data. In the deep learning algorithm, multiple convolutional layers are stacked to extract low-level and high-level features. After the convolutional layer, a pooling layer is added to perform feature dimension reduction. In the feature extraction process, a residual structure is introduced to solve the gradient vanishing problem in model training, and an attention mechanism is introduced to enhance the model's ability to recognize key feature areas. S1022: Input the extracted image features into the fully connected layer for feature integration and classification, use a number of image data labeled with normal and abnormal fence states as training sets, train the fence state detection model, and obtain the trained fence state detection model. During the training process, use the back propagation algorithm and optimizer to adjust the parameters of the fence state detection model.

4. The method for detecting airport fence status based on machine learning on an unmanned vehicle according to claim 1 is characterized in that: Get the fence status detection results, including: Compressing the fence state detection model to obtain compressed fence state detection model data; Pruning the compressed fence state detection model data to reduce model parameters and calculation amount, and obtaining a fence state detection model after pruning; Deploy the pruned fence status detection model to the unmanned vehicle to generate a model to be detected; Obtain fence image data captured in real time by the dual-spectrum high-definition camera on the unmanned vehicle, and pre-process the fence image data as standardized image data; Input the standardized image data into the model to be detected for real-time detection; Get fence status detection results from real-time detection.

5. The method for detecting airport fence status based on machine learning on an unmanned vehicle according to claim 1 is characterized in that: The fence status detection results are displayed on the display screen of the unmanned vehicle or the remote monitoring center, including: Determine the fence status detection result output by the display screen of the unmanned vehicle and transmit it to the remote monitoring center. The detection result includes normal status result and abnormal status result; Determine and output a plurality of new fence status detection results, determine the status of each new detection result and transmit it to the remote monitoring center until the number of new abnormal status results reaches a preset number threshold, or the current detection status result is determined to be the last detection result; Control the unmanned vehicle to record and save the image data of abnormal status, and return to execute the output of the initial fence status detection result through the display screen or the remote monitoring center until the number of new abnormal status results reaches the preset number threshold, or the current detection status result is determined to be the last detection result; Generate and output a list based on each test result and the status of the test result; Determine the fence status detection results received by the unmanned vehicle terminal or remote monitoring center and transmit them to the alarm mechanism.

6. The method for detecting airport fence status based on machine learning on an unmanned vehicle according to claim 3 is characterized in that: The back propagation algorithm and optimizer are used to adjust the parameters of the fence status detection model, including: Obtain a preset airport fence standard image dataset, and extract multiple image features from the image dataset; Input the extracted multiple image features into the fully connected layer to integrate and classify the image features; Obtain a preset image dataset that annotates the normal state and abnormal state of the fence, and use the image dataset as a training set to train the fence state detection model; During the training process, the back propagation algorithm is used to calculate the error of the fence state detection model, and the parameters of the fence state detection model are updated according to the error; Obtain a preset optimizer, select the optimal parameter adjustment strategy from the optimizer, and iteratively adjust the parameters of the fence status detection model; Repeat the back propagation algorithm and optimizer parameter adjustment steps until the error value of the fence status detection model is less than the preset error threshold; Obtain the trained fence status detection model and verify the detection accuracy of the fence status detection model under different lighting conditions, angles, and weather conditions.

7. The method for detecting airport fence status based on machine learning on an unmanned vehicle according to claim 4 is characterized in that: Input standardized image data into the model to be detected for real-time detection, including: Inputting the first frame of standardized image data into the model to be detected to obtain a first detection result; Starting from the second frame of standardized image data, the detection result of the previous frame and the standardized image data of the current frame are input into the model to be detected to obtain the detection result of the current frame; Integrate the first detection result and the detection results of all current frames to obtain a fence status detection result; According to the fence status detection results, the status of each area of ​​the airport is detected, including: Acquire airport fence standard image data, and extract airport area division information from the airport fence standard image data; According to the airport area division information, the fence status detection results are mapped to the corresponding airport areas; Evaluate the fence status of each airport area and obtain the airport area status evaluation results; Generate an airport safety status report based on the airport area status assessment results.

8. The method for detecting airport fence status based on machine learning on an unmanned vehicle according to claim 5 is characterized in that: Determine and output multiple new fence status detection results, including: Identifying and extracting fence status detection data from the standardized image data, wherein the fence status detection data includes fence detection parameters and image environment parameters; According to the fence detection parameters and image environment parameters, the fence status detection data is processed in layers; Compare the airport fence standard image data with the fence status detection data to determine the fence status; Compare and analyze each new fence status detection result, including normal status results and abnormal status results, match the new fence status detection results with the standard status in the airport fence standard image data one by one, and output the status of each detection result.

9. The method for detecting airport fence status based on machine learning on an unmanned vehicle according to claim 5 is characterized in that: Determine the fence status detection results received by the unmanned vehicle or remote monitoring center and transmit them to the alarm mechanism, including: Obtain the image data and transmission time of the fence status detection result received by the unmanned vehicle terminal or the remote monitoring center. The transmission time is the time between the output of the current fence status detection result and the acquisition of the alarm signal; Determine the current fence status detection result based on the image data and transmit it to the alarm mechanism, the detection result includes normal status result and abnormal status result; If the detection result is an abnormal state result, an alarm mechanism is triggered based on the transmission time and image data; If the test result is normal, continue to monitor and save the test result.

10. The method for detecting airport fence status based on machine learning on an unmanned vehicle according to claim 8, characterized in that: According to the fence detection parameters and image environment parameters, the fence status detection data is processed in layers, including: Get the environment parameter type from the image environment parameter, and get the fence detection parameter type from the fence detection parameter; Obtain the fence detection parameter type associated with each environmental parameter type as a data group to be processed; After sorting each data group to be processed according to the environmental parameter type, the data corresponding to each environmental parameter type and fence detection parameter type is stored in the corresponding data group to be processed, and the fence status detection data is obtained.

Images