Fiber Optic Intrusion Detection Method and System Based on Deep Learning

Abstract

The present disclosure provides a fiber optic intrusion detection method and system based on deep learning, including: real-time collecting fiber optic signals in the area to be detected and performing corresponding preprocessing; converting the preprocessed fiber optic signals into images by using Gramian Angular Field; inputting the images into a pre-trained classifier to obtain the recognition results of the fiber optic signals; wherein, the classifier adopts a residual network, and its training process includes: constructing a training set, wherein the samples in the training set include the images corresponding to the collected historical fiber optic signals and the discrimination results of the intrusion events corresponding to the fiber optic signals; based on the images in the training set and their corresponding discrimination results, pre-training the CoAtNet model by using the knowledge distillation method, and using the privileged features, ordinary features obtained from the training, as well as the images in the training set and their corresponding judgment results as the input of the residual network to train it, so as to obtain a trained classifier.

Description

Technical Field

[0001] The present disclosure belongs to the technical field of intrusion detection, and particularly relates to a fiber optic intrusion detection method and system based on deep learning. Background Art

[0002] The statements in this part only provide background technical information related to the present disclosure and do not necessarily constitute prior art.

[0003] In recent years, as deep learning has become one of the latest trends in machine learning and artificial intelligence research, the research on fiber optic intrusion system algorithms based on deep learning has become the main breakthrough direction in the research of border security systems.

[0004] Traditional border security systems mainly include high-end imaging devices at night such as night vision devices, thermal imagers, and infrared cameras, and intelligent video surveillance devices. Although the former has advantages such as long night vision distance and strong concealment, the imaging system (lens), CCD, and DSP technology patents of infrared integrated machines are all in the hands of well-known international companies. It is not so easy for domestic security companies to get a share in the fierce bidding; for the latter, although there is no technical barrier, the effective monitoring range of intelligent monitoring devices is limited. If used on the border, a large number of intelligent monitoring devices need to be installed and used together, and personnel are required to monitor the screen at all times, and the human resources consumed are quite considerable.

[0005] In recent years, border security systems based on deep learning have gradually become popular, but the misjudgment rate of traditional algorithm-based border security systems is relatively high. This is because under the influence of environmental factors, convolutional neural networks cannot filter out the influence of environmental noise and cannot extract and classify time-domain features well. Summary of the Invention

[0006] In order to solve the above problems, the present disclosure provides a fiber optic intrusion detection method and system based on deep learning. The solution extracts the time-domain features of the collected fiber optic signals with environmental noise by using the Gram angle field. At the same time, the CoAtNet model is used to train the intrusion signals, and the trained weights are distilled to the resnet-18 model with low computing power requirements through a feature-based knowledge distillation method for intrusion detection, which can effectively extract time-domain features and effectively improve the accuracy of intrusion detection.

[0007] According to the first aspect of the embodiments of the present disclosure, a fiber optic intrusion detection method based on deep learning is provided, including:

[0008] Collect the fiber optic signals in the area to be detected in real time and perform corresponding preprocessing;

[0009] Convert the preprocessed fiber optic signals into images by using the Gram angle field;

[0010] Input the image into a pre-trained classifier to obtain the recognition result of the optical fiber signal. Among them, the classifier uses a residual network, and its training process includes: constructing a training set, where the samples in the training set include the images corresponding to the collected historical optical fiber signals and the discrimination results of the intrusion events corresponding to the optical fiber signals; based on the images in the training set and their corresponding discrimination results, use the knowledge distillation method to pre-train the CoAtNet model in advance, and use the privileged features, ordinary features obtained from the training, and the images in the training set and their corresponding judgment results as the input of the residual network to train it, and obtain a trained classifier.

[0011] Further, the conversion of the preprocessed optical fiber signal into an image using the Gram angular field is specifically as follows:

[0012] Normalize the preprocessed optical fiber signal sequence;

[0013] Convert the elements in the normalized optical fiber signal sequence into the polar coordinate space to obtain the polar coordinate representation;

[0014] Generate a Gram angular field based on the cosine value of the sum of the polar angles corresponding to each element in the optical fiber signal sequence in the polar coordinate space;

[0015] Obtain the corresponding image based on the element values of the Gram angular field.

[0016] Further, the conversion of the elements in the normalized signal sequence into the polar coordinate space is specifically that the time stamp of the element in the sequence is used as the radius, and the arccosine value of the element value is used as the cosine angle in the polar coordinate space.

[0017] Further, for the intrusion events identified by the trained classifier, update the images corresponding to their optical fiber signals and the intrusion event discrimination results to the training set, and re-train the classifier using the updated training set when the preset time period is satisfied.

[0018] Further, when using the knowledge distillation method for training, use the CoAtNet model as the Teacher module and the ResNet-18 model as the Student module, and supervise the training of the Student module based on the training results of CoAtNet, and finally obtain a classifier for optical fiber signal recognition.

[0019] According to the second aspect of the embodiments of the present disclosure, an optical fiber intrusion detection system based on deep learning is provided, including:

[0020] A data acquisition unit, which is used to collect the optical fiber signals in the area to be detected in real time and perform corresponding preprocessing;

[0021] A format conversion unit, which is used to convert the preprocessed optical fiber signal into an image by using the Gram angular field;

[0022] An identification unit, which is used to input the image into a pre-trained classifier to obtain the identification result of the optical fiber signal; wherein, the classifier adopts a residual network, and its training process includes: constructing a training set, wherein the samples in the training set include the images corresponding to the collected historical optical fiber signals and the discrimination results of the intrusion events corresponding to the optical fiber signals; based on the images in the training set and their corresponding discrimination results, the CoAtNet model is pre-trained by using the knowledge distillation method, and the privileged features, ordinary features obtained from the training, and the images in the training set and their corresponding judgment results are used as the inputs of the residual network to train it, and a trained classifier is obtained.

[0023] According to the third aspect of the embodiments of the present disclosure, an electronic device is provided, including a memory, a processor, and a computer program running on the memory. When the processor executes the program, the above-mentioned fiber intrusion detection method based on deep learning is implemented.

[0024] According to the fourth aspect of the embodiments of the present disclosure, a non-transitory computer-readable storage medium is provided, on which a computer program is stored. When the program is executed by a processor, the above-mentioned fiber intrusion detection method based on deep learning is implemented.

[0025] Compared with the prior art, the beneficial effects of the present disclosure are:

[0026] (1) The present disclosure provides a fiber intrusion detection method and system based on deep learning. The solution is based on the collected optical fiber signals with environmental noise, and uses the Gram angular field to fully extract their time-domain features. At the same time, the CoAtNet model is used to train the intrusion signals, and the trained weights are distilled to the resnet-18 model with low computing power requirements through the feature-based knowledge distillation method for intrusion detection, effectively reducing the hardware requirements. At the same time, it can effectively extract the time-domain features and effectively improve the accuracy of intrusion detection.

[0027] (2) By updating the recognized data to the training set and training the classifier periodically, the solution of the present disclosure can effectively ensure the robustness and accuracy of the classifier.

[0028] (3) The solution of the present disclosure can effectively solve the problems of long boundary lines, large personnel consumption, and high false positive rate.

[0029] The advantages of the additional aspects of the present disclosure will be partially given in the following description, partially become obvious from the following description, or be understood through the practice of the present disclosure. Description of the Drawings

[0030] The accompanying drawings forming a part of this disclosure are used to provide a further understanding of the disclosure. The schematic embodiments and descriptions thereof of the disclosure are used to explain the disclosure and do not constitute an improper limitation of the disclosure.

[0031] Figure 1 It is a schematic structural diagram of a system corresponding to a fiber optic intrusion detection method based on deep learning described in an embodiment of the present disclosure;

[0032] Figure 2 It is a flowchart of a fiber optic intrusion detection method based on deep learning described in an embodiment of the present disclosure. Detailed implementation manners

[0033] The following further describes the present disclosure in conjunction with the accompanying drawings and embodiments.

[0034] It should be noted that the following detailed descriptions are all exemplary and are intended to provide further explanations of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present disclosure belongs.

[0035] It should be noted that the terms used herein are only for describing specific implementation manners and are not intended to limit the exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0036] In the case of no conflict, the embodiments in the present disclosure and the features in the embodiments may be combined with each other.

[0037] Embodiment 1:

[0038] The purpose of this embodiment is to provide a fiber optic intrusion detection method based on deep learning.

[0039] A fiber optic intrusion detection method based on deep learning includes:

[0040] Real-time collect the fiber optic signals in the area to be detected and perform corresponding preprocessing;

[0041] Convert the preprocessed fiber optic signals into images using the Gramian angular field;

[0042] Input the image into a pre-trained classifier to obtain the recognition result of the optical fiber signal. Among them, the classifier uses a residual network, and its training process includes: constructing a training set, where the samples in the training set include the images corresponding to the collected historical optical fiber signals and the discrimination results of the intrusion events corresponding to the optical fiber signals; based on the images in the training set and their corresponding discrimination results, use the knowledge distillation method to pre-train the CoAtNet model in advance, and use the obtained privileged features, ordinary features, and the images in the training set and their corresponding judgment results as the input of the residual network to train it, and obtain a trained classifier.

[0043] In a specific implementation, the conversion of the preprocessed optical fiber signal into an image using the Gram angular field is specifically as follows:

[0044] Normalize the preprocessed optical fiber signal sequence;

[0045] Convert the elements in the normalized optical fiber signal sequence to the polar coordinate space to obtain a polar coordinate representation;

[0046] Generate a Gram angular field based on the cosine value of the sum of the polar angles corresponding to each element in the optical fiber signal sequence in the polar coordinate space;

[0047] Obtain the corresponding image based on the element values of the Gram angular field.

[0048] In a specific implementation, when converting the elements in the normalized signal sequence to the polar coordinate space, specifically, use the timestamp of the element in the sequence as the radius, and the arccosine value of the element value as the cosine angle in the polar coordinate space.

[0049] In a specific implementation, for the intrusion events identified by the trained classifier, update the images corresponding to their optical fiber signals and the intrusion event discrimination results to the training set, and when the preset time period is met, use the updated training set to retrain the classifier.

[0050] In a specific implementation, when using the knowledge distillation method for training, use the CoAtNet model as the Teacher module and the ResNet-18 model as the Student module, and supervise the training of the Student module based on the training results of CoAtNet, and finally obtain a classifier for optical fiber signal recognition; the CoAtNet model includes a first convolutional module, a second convolutional module, a first self-attention module, and a second self-attention module arranged in sequence.

[0051] Regarding the deficiencies existing in the prior art, such as Figure 2As shown, this embodiment provides a fiber optic intrusion detection method based on deep learning for fiber optic data containing environmental noise to achieve accurate detection of intrusion signals. The specific steps are as follows:

[0052] S1. Obtain the fiber optic signals in the area to be monitored and perform corresponding preprocessing;

[0053] Among them, a Distributed Optical Fiber Vibration Sensing System (DVS) is used to collect fiber optic signals. Among them, an optical cable is buried underground in the area to be monitored for obtaining vibration signals; at the same time, the signals are processed into new data with a time length of 5 seconds (10,000 sampling points) and a step size of 1.25 seconds (2,500 sampling points), and then preprocessing such as filtering and decimating by a factor of ten is performed;

[0054] S2. Input the preprocessed fiber optic signals into the Gramian Angular Field (GAF) to convert the time series data into pictures, which can better extract the time domain features in the fiber optic signals into the pictures and improve the recognition accuracy of the model. The steps of using the Gramian Angular Field for data processing include the following parts:

[0055] S201: Normalization

[0056] The first step of GAF is to normalize the time series data to the interval [-1, +1]. Assume the time series data is X = x1, x2, x3, x4,... x N , and the normalized value is denoted as The normalized Gram Matrix is as follows:

[0057]

[0058] Here, φ i,j represents the angle between vector i and vector j. G represents the correlation degree between vectors, and this correlation degree is actually determined by the angle between vectors. Since the time series data points are not vectors, polar coordinates are introduced.

[0059] S202: Coordinate transformation

[0060] Convert the normalized values into polar coordinates

[0061]

[0062]

[0063] where t i ∈N represents the point x iThe timestamp, where N is the number of all time points included in the time series data. Each time series point data contains two pieces of information: one is the normalized value of this data point The other is its time series position t i , the above polar coordinate transformation encoding includes both of these two pieces of information and does not lose any information. Mathematically, it is a bijective function, that is, the independent variable and the dependent variable of this function have a one-to-one correspondence relationship, and it is the same in both forward and reverse directions. The polar axis r i , retains the relationship in time; the polar angle φ i retains the relationship in value; performing polar coordinates on the time series data can be interpreted as understanding the data from another perspective.

[0064] S203: Customized inner product

[0065] GAF defines its own special inner product: <x1,x2> = cos(φ1 + φ2), that is to say, the "inner product" between two time series points is the cosine of the sum of the polar angles after the polar coordinate transformation of these two time series points, in the form of:

[0066]

[0067] The above matrix with the customized inner product is the result of GAF. We can see that this is an n×n square matrix, and time is encoded into the geometric dimension 1...n of the square matrix. Compared with the original time series data, GAF adds one-dimensional information. GAF transforms each point of the time series data into the correlation relationship between this point and other points, and this relationship is measured using the customized inner product of GAF.

[0068] S3. First, input the feature image output by the Gram angle field into the Teacher module in the knowledge distillation network for training. The Teacher module uses the CoAtNet algorithm, which is composed of two network architectures, transformer + CNN. The transformer solves the limitation that RNN cannot perform parallel computing. Compared with CNN, the number of operations required to calculate the correlation between two positions does not increase with the distance, and the transformer can generate a more interpretable model. However, the transformer lacks an ideal inductive bias, so a large amount of data and computing resources are required to compensate. Therefore, consider combining the transformer and CNN so that the model has a good inductive bias under low data conditions, making CoAtNet not only have the scalability of the Transformer model but also a faster convergence speed, thereby improving the efficiency. The solution in this embodiment is designed based on the principle of CoAtNet for its network structure:

[0069] First, the convolutional module and the self-attention module are combined and compared through experiments to determine an optimal combination method. A simple idea obtained after comparison is to simply sum the global static convolutional kernel and the adaptive attention matrix before softmax normalization.

[0070] The formula is as follows, where x i , y i ∈R D are the input and output at position i respectively, and R D is the set of D-dimensional real numbers, where represents the global space.

[0071]

[0072] Among them, is the global space, i, j, and k are the positions in the global space respectively, and W is the depth convolutional kernel, which is an input-independent parameter of static values.

[0073] Secondly, after finding a concise method to combine convolution and attention, we next stack the entire network using a vertical layout design. To obtain better performance, a five-level network simulating the construction of ConvNets is used, and finally, through experimental comparison of generalization and model capabilities, the network structure is determined to be C-C-T-T (C represents the convolutional network, and T represents the self-attention network). After stacking, a variant model of the transformer, CoAtNet, is obtained.

[0074] S4. Use the method based on feature distillation to input the privileged features and ordinary features learned by the teacher module into the student model to supervise the training of the student model and improve the training effect of the model. The main difference between privileged feature distillation and model distillation is that the knowledge of the former teacher model comes from privileged features and ordinary features, while the knowledge source of the latter mainly comes from a more complex model. Among them, the loss function of privileged feature distillation is as follows:

[0075] minw s (1 - λ)*L s (y, f s (X; W s )) + λ*L d (f t (X, X * ; W t ), f s (X; W s )) (6)

[0076] Among them, f t and f s represent the teacher model and the student model respectively, and Ls Denote the loss of the student model with respect to the "hard" labels as \(L\). d Denote the loss of the student model with respect to the distilled "soft" labels as \(L_d\), where \(\lambda\in[0,1]\) is a hyperparameter that balances the above two loss values; \(X\) is the input information, and \(W\) s are the parameters of the teacher model, which remain unchanged during the training process, and \(X\) * is the privileged information.

[0077] S5. When the classifier identifies an intrusion event, an alarm is given through a buzzer and a light alarm.

[0078] S6. Save the intrusion signal of the intrusion event to the training set, update the training set, and prepare for the subsequent regular update of the weight file.

[0079] Embodiment 2:

[0080] The purpose of this embodiment is to provide a fiber optic intrusion detection system based on deep learning.

[0081] A fiber optic intrusion detection system based on deep learning, comprising:

[0082] A data acquisition unit, which is used to collect fiber optic signals in the area to be detected in real time and perform corresponding preprocessing;

[0083] A format conversion unit, which is used to convert the preprocessed fiber optic signals into images using the Gramian Angular Field;

[0084] An identification unit, which is used to input the image into a pre-trained classifier to obtain the identification result of the fiber optic signal; wherein, the classifier adopts a residual network, and its training process includes: constructing a training set, where the samples in the training set include the images corresponding to the collected historical fiber optic signals and the discrimination results of the intrusion events corresponding to the fiber optic signals; based on the images in the training set and their corresponding discrimination results, the CoAtNet model is pre-trained using the knowledge distillation method, and the privileged features, ordinary features obtained from the training, as well as the images in the training set and their corresponding judgment results are used as the input to the residual network to train it, and a trained classifier is obtained.

[0085] As Figure 1 shown, a specific deployment method of a fiber optic intrusion detection system based on deep learning is presented. Specifically, in this embodiment, this set of systems is applied to the school fence for testing, where the teacher module of knowledge distillation is trained on the server in the laboratory, and the trained weight file is input into the student module in the on-site computing stick for testing, and the test result accuracy rate reaches over 90%.

[0086] The fiber optic intrusion detection system based on deep learning described in this embodiment can better extract time-domain features. In this system, we use the CoAtNet algorithm with better classification effect for classification and recognition. Finally, through the method of knowledge distillation, it is trained and tested on the resnet-18 model. This knowledge distillation method can reduce the computing power requirements of the model, enabling it to achieve high-precision prediction on computers with low computing power. This method can well solve the problems of long boundary lines, large personnel consumption, and high false positive rates.

[0087] Furthermore, the system described in this embodiment corresponds to the method described in Embodiment 1, and its technical details have been described in detail in Embodiment 1, so they will not be repeated here.

[0088] In more embodiments, there is also provided:

[0089] An electronic device includes a memory, a processor, and computer instructions stored on the memory and running on the processor. When the computer instructions are run by the processor, the method described in Embodiment 1 is completed. For the sake of brevity, it will not be repeated here.

[0090] It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA, or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0091] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A part of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0092] A computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by the processor, the method described in Embodiment 1 is completed.

[0093] The method in Embodiment 1 can be directly embodied as being executed and completed by a hardware processor, or by a combination of hardware and software modules in the processor. The software module may be located in a mature storage medium in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. This storage medium is located in the memory, and the processor reads the information in the memory and combines its hardware to complete the steps of the above method. To avoid repetition, it will not be described in detail here.

[0094] Those of ordinary skill in the art will realize that the units and algorithm steps of each example described in conjunction with this embodiment can be implemented by either electronic hardware or a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods for each specific application to implement the described functions, but such implementation should not be considered to exceed the scope of this disclosure.

[0095] The fiber intrusion detection method and system based on deep learning provided by the above embodiment can be implemented and have broad application prospects.

[0096] The foregoing is only a preferred embodiment of the present disclosure and is not intended to limit the present disclosure. For those skilled in the art, various changes and modifications can be made to the present disclosure. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present disclosure shall be included within the protection scope of the present disclosure.

Claims

1. A fiber optic intrusion detection method based on deep learning, characterized in that, Including: Collecting optical fiber signals in the area to be detected in real time and performing corresponding preprocessing; Converting the preprocessed optical fiber signals into images by using the Gram angular field; Inputting the image into a pre-trained classifier to obtain the recognition result of the optical fiber signal; wherein, the classifier adopts a residual network, and its training process includes: constructing a training set, where the samples in the training set include the images corresponding to the collected historical optical fiber signals and the discrimination results of the intrusion events corresponding to the optical fiber signals; based on the images in the training set and their corresponding discrimination results, pre-training the CoAtNet model by using the knowledge distillation method, and using the privileged features, ordinary features obtained from the training, and the images in the training set and their corresponding judgment results as the input of the residual network to train it, so as to obtain a trained classifier; When using the knowledge distillation method for training, taking the CoAtNet model as the Teacher module and the ResNet-18 model as the Student module, supervising the training of the Student module based on the training results of CoAtNet, and finally obtaining a classifier for optical fiber signal recognition; The loss function of knowledge distillation is as follows: ； where f t and f s represent the teacher model and the student model respectively, L s represents the loss of the student model with respect to the hard labels, and L d represents the loss of the student model with respect to the soft labels obtained by distillation. λ ∈ [0, 1] is a hyperparameter that weighs the above two loss values; X is the input information, and W s are the parameters of the teacher model, which remain unchanged during the training process, and X * is the privileged information.

2. The fiber optic intrusion detection method based on deep learning according to claim 1, characterized in that, The specific method of converting the preprocessed optical fiber signals into images by using the Gram angular field is: Normalizing the preprocessed optical fiber signal sequence; Converting the elements in the normalized optical fiber signal sequence into the polar coordinate space to obtain a polar coordinate representation; Generating a Gram angular field based on the cosine value of the sum of the polar angles corresponding to each element in the optical fiber signal sequence in the polar coordinate space; Obtaining the corresponding image based on the element values of the Gram angular field.

3. The fiber optic intrusion detection method based on deep learning according to claim 1, characterized in that, Converting the elements in the normalized signal sequence into the polar coordinate space, using the timestamp of the element in the sequence as the radius and the arccosine value of the element value as the cosine angle in the polar coordinate space.

4. The fiber optic intrusion detection method based on deep learning according to claim 1, characterized in that, For the intrusion events recognized by the trained classifier, updating the images corresponding to their optical fiber signals and the intrusion event discrimination results to the training set, and retraining the classifier by using the updated training set when a preset time period is met.

5. The fiber optic intrusion detection method based on deep learning according to claim 1, characterized in that, The CoAtNet model includes a first convolutional module, a second convolutional module, a first self-attention module, and a second self-attention module arranged in sequence.

6. The fiber optic intrusion detection method based on deep learning according to claim 1, characterized in that, The corresponding preprocessing is specifically: filtering and downsampling the collected optical fiber signals to obtain an optical fiber signal sequence arranged in chronological order.

7. A fiber optic intrusion detection system based on deep learning, characterized in that, Including: A data acquisition unit, which is used to collect optical fiber signals in the area to be detected in real time and perform corresponding preprocessing; A format conversion unit, which is used to convert the preprocessed optical fiber signals into images by using the Gram angular field; An identification unit, which is used to input the image into a pre-trained classifier to obtain the identification result of the optical fiber signal; wherein, the classifier adopts a residual network, and its training process includes: constructing a training set, wherein the samples in the training set include the images corresponding to the collected historical optical fiber signals and the discrimination results of the intrusion events corresponding to the optical fiber signals; based on the images in the training set and their corresponding discrimination results, the CoAtNet model is pre-trained by using the knowledge distillation method, and the privileged features, ordinary features obtained from the training, as well as the images in the training set and their corresponding judgment results are used as the input of the residual network to train it, and a trained classifier is obtained; The loss function of knowledge distillation is as follows: ； where f t and f s represent the teacher model and the student model respectively, L s represents the loss of the student model with respect to the hard labels, and L d represents the loss of the student model with respect to the soft labels obtained by distillation. λ ∈ [0, 1] is a hyperparameter that balances the above two loss values; X is the input information, and W s is the parameter of the teacher model, which remains unchanged during the training process, and X * is the privileged information.

8. An electronic device, including a memory, a processor, and a computer program running on the memory, where the processor, when executing the program, implements a method for optical fiber intrusion detection based on deep learning according to any one of claims 1-6.

9. A non-transitory computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements a method for optical fiber intrusion detection based on deep learning according to any one of claims 1-6.

Images