A full-acoustic optical fiber monitoring method and system based on machine learning
Through the analysis of fiber-optic sensing signals through the GAS host combined with machine learning technology, the accurate identification of sound in full-sound fiber monitoring technology and rich classification of alarm information is achieved, which solves the shortcomings of sound recognition and monitoring in the existing technology and improves multiple Monitoring efficiency of application scenarios.
Patent Information
- Application Number
- CN202211498031.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-11-28
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2042-11-28
AI Technical Summary
The existing full-sound fiber monitoring technology has shortcomings in sound identification, analysis and alarm, and it is difficult to achieve accurate alarm information classification and real-time monitoring.
The optical fiber sensing signal is collected through the GAS host, combined with machine learning technology, and the acoustic wave vibration situation is analyzed using CNN audio classifier and neural network model, and the trained neural network model is generated to restore the live sound and monitor it in actual scenarios.
It has achieved richer alarm classification information, improved the accuracy and real-time monitoring capabilities of alarm information, and is suitable for multiple application scenarios such as oil and gas pipelines, national defense borders and natural disaster monitoring, greatly improving patrol efficiency.
Smart Images

Figure CN115713827B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of machine learning technology, and more specifically, to a full-acoustic optical fiber monitoring method and system based on machine learning. The method collects signals sent back by optical fiber sensing through a GAS host, and combines machine learning technology to analyze the acoustic vibration of the environment at the location, thereby restoring the on-site sound. Background Art
[0002] The all-acoustic fiber optic monitoring technology (GAS) uses the single-mode fiber core in ordinary optical cables to form a fiber optic sensor, collects the vibration signal around the optical fiber as the main parameter, realizes frequency collection of 0.5Hz~20kHz, and achieves sound collection and restoration. However, the all-acoustic fiber optic monitoring technology currently on the market still needs to be improved in terms of sound recognition, analysis, and alarm.
[0003] The present invention collects the signal sent back by the optical fiber sensor through the GAS host, and combines the machine learning technology to analyze the acoustic vibration of the environment field where the location is located to restore the on-site sound, which can realize richer alarm classification information and more accurate alarm information classification. The GAS host designed by the present invention and combined with machine learning can be widely used in oil and gas pipelines, national defense borders, communication optical cables, mine search and rescue, power cables, railway security, highways, natural disaster monitoring and other fields. In the perimeter application scenario, the optical cable laid on the soil, sand, pond, wall or fence is used as a sensing unit to monitor and alarm the peripheral intrusion events such as illegal intrusion, climbing, abnormal vehicle passing, and impact on fences and walls in real time. In the long-distance pipeline monitoring application scenario, beside the pipeline to be monitored, the accompanying ordinary optical cable can be used to realize real-time monitoring of the entire line, collect the sound and vibration signals of the entire line, and combine the Al intelligent algorithm to distinguish the event type to find problems and warn in the first time, greatly improving the inspection efficiency. Summary of the invention
[0004] The purpose of the present invention is to provide a full-acoustic optical fiber monitoring method and system based on machine learning, which collects the signals sent back by optical fiber sensing through the GAS host, and combines machine learning technology to analyze the acoustic vibration conditions of the environmental field at the location, thereby restoring the on-site sound.
[0005] The present invention is implemented by the following technical solution: a full acoustic optical fiber monitoring method based on machine learning, comprising the following steps:
[0006] Step S1, using the GAS host to collect on-site sound data, denoising the sound data, and generating a model training set;
[0007] Step S2, in the GAS host, capture sound data according to the CNN audio classifier, use the spectrogram to represent the sound data set, build a neural network model, use the neural network model to capture the sound data in the spectrogram, train the neural network model with the model training set, and preset the number of iterations and the learning rate;
[0008] The neural network model is composed of a point convolution combination module of the upper layer and a double convolution pooling combination module of the lower layer connected in sequence, wherein the point convolution combination module of the upper layer includes a first convolution layer, a second convolution layer, a first batch of normalization layers, a second batch of normalization layers, a first activation function layer, a second activation function layer, a maximum pooling layer and a feature multiplication layer;
[0009] The dual convolution pooling combination module includes a third convolution layer, a fourth convolution layer, a fifth convolution layer, a fully connected layer, a third batch one normalization layer, a third activation function layer, a first regularization layer and a loss function layer;
[0010] Step S3, inputting the sound data set into the neural network model in the structure of a data cube to train the neural network model, thereby generating a trained neural network model;
[0011] Step S4, in the actual scenario, use the GAS host to collect on-site sound data to generate a corresponding sound data set represented by a spectrogram, input the corresponding sound data set into the trained neural network model, and restore the sound wave vibration conditions of the on-site environment of the location through deep mining and analysis, restore the on-site sound and monitor it.
[0012] In order to better implement the present invention, further, the method for designing the point convolution combination module in step S2 includes:
[0013] Connecting the first convolution layer and the second convolution layer in parallel in the point convolution combination module;
[0014] The first convolution layer is connected to the first batch of normalization layers and the first activation function layer in sequence, and a Relu activation function is set in the first activation function layer;
[0015] The second convolutional layer is sequentially connected to the second batch of normalization layers and the second activation function layer, and a Relu activation function is set in the second activation function layer;
[0016] The first activation function layer and the second activation function layer are connected through a feature multiplication layer, and the feature multiplication layer and the maximum pooling layer are connected.
[0017] In order to better implement the present invention, further, the point convolution combination module is used to solve the disorder of the sound data set and increase the local features of the sound data set.
[0018] In order to better implement the present invention, further, the method for designing the double convolution pooling combination module in step S2 includes:
[0019] In the dual convolution pooling combination module, the third convolution layer is sequentially connected to the third convolution layer, the fourth convolution layer, the fifth convolution layer, the fully connected layer, the third batch one normalization layer, the third activation function layer, the first regularization layer and the loss function layer;
[0020] The Relu activation function is set in the third activation function layer.
[0021] In order to better implement the present invention, further, the dual convolution pooling combination module is used to classify and display the sound data set displayed in the form of a spectrogram.
[0022] In order to better implement the present invention, further, step S3 includes:
[0023] After the sound dataset is input into the neural network model, the point convolution combination module is used to extract the local information of the sound dataset. At the same time, the upper point convolution combination module upsamples the sound data, and then deeply mines the sound dataset. Finally, the lower double convolution pooling combination module aggregates the feature information of the sound dataset, and then uses the Relu activation function in the third activation function layer and the softmax loss function in the loss function layer to calculate the loss value of the predicted point of the sound dataset and the real point of the sound dataset in the spectrogram. The learning rate descent method is used to iteratively attenuate the loss value and optimize the weight parameters of the neural network model until the number of iterations is equal to the maximum number of iterations. The training is terminated to generate a trained neural network model.
[0024] In order to better implement the present invention, further, step S4 includes:
[0025] The GAS host performs fiber optic monitoring based on the optical cable itself to complete the sound verification.
[0026] In order to better implement the present invention, further, the present invention also provides a full acoustic optical fiber monitoring system based on machine learning, including a collection unit, a model building unit, a training unit and a monitoring unit, wherein:
[0027] The acquisition unit is used to collect on-site sound data using the GAS host, denoise the sound data, generate a model training set, capture the sound data according to the CNN audio classifier in the GAS host, and represent the sound data set using a spectrogram;
[0028] A model building unit is used to build a neural network model, use the neural network model to capture sound data in the spectrogram, train the neural network model with a model training set, and preset the number of iterations and learning rate;
[0029] The neural network model is composed of a point convolution combination module of the upper layer and a double convolution pooling combination module of the lower layer connected in sequence, wherein the point convolution combination module of the upper layer includes a first convolution layer, a second convolution layer, a first batch of normalization layers, a second batch of normalization layers, a first activation function layer, a second activation function layer, a maximum pooling layer and a feature multiplication layer;
[0030] The dual convolution pooling combination module includes a third convolution layer, a fourth convolution layer, a fifth convolution layer, a fully connected layer, a third batch one normalization layer, a third activation function layer, a first regularization layer and a loss function layer;
[0031] A training unit, used for inputting the sound data set into the neural network model in the structure of a data cube to train the neural network model and generate a trained neural network model;
[0032] The monitoring unit is used to use the GAS host to collect on-site sound data in actual scenarios to generate a corresponding sound data set represented by a spectrogram, input the corresponding sound data set into the trained neural network model, and restore the sound wave vibration conditions of the on-site environment of the location through deep mining and analysis, restore the on-site sound and monitor it.
[0033] In order to better implement the present invention, the present invention further provides a computer-readable storage medium on which computer program instructions are stored, and the program instructions implement the steps of the above method when executed by a processor.
[0034] Compared with the prior art, the present invention has the following advantages and beneficial effects:
[0035] (1) The full-acoustic optical fiber monitoring method designed in the present invention can perform real-time monitoring on every 2 to 8 meters (can be set according to demand) of the sensing optical fiber as a unit, and can flexibly set the monitoring interval for different applications and protection scenarios, so as to facilitate the construction of a multi-scenario, long-distance, integrated security system for perimeter protection and long-distance perimeter protection;
[0036] (2) The present invention collects the signals sent back by the optical fiber sensor through the GAS host, and combines machine learning technology to analyze the acoustic vibration of the environment field at the location to restore the on-site sound, which can achieve richer alarm classification information and more accurate alarm information classification. The GAS host combined with machine learning designed by the present invention can be widely used in oil and gas pipelines, national defense borders, communication optical cables, mine search and rescue, power cables, railway security, highways, natural disaster monitoring and other fields. In the perimeter application scenario, the optical cable laid on the soil, sand, pond, wall or fence is used as a sensing unit to monitor and alarm the peripheral intrusion events such as illegal intrusion, climbing, abnormal vehicle passing, and impact on fences and walls in real time. In the long-distance pipeline monitoring application scenario, beside the pipeline to be monitored, the accompanying ordinary optical cable can be used to realize real-time monitoring of the entire line, collect the sound and vibration signals of the entire line, and combine the Al intelligent algorithm to distinguish the event type to find problems and issue warnings in the first time, greatly improving the inspection efficiency. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] The present invention is further described in conjunction with the following drawings and embodiments, and all the concepts and innovations of the present invention should be regarded as the disclosed contents and the protection scope of the present invention.
[0038] Figure 1 This is a schematic diagram of the structure of the neural network model provided by the present invention. DETAILED DESCRIPTION
[0039] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. It should be understood that the described embodiments are only part of the embodiments of the present invention, not all of the embodiments, and therefore should not be regarded as limiting the scope of protection. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technical personnel in this field without making creative work are within the scope of protection of the present invention.
[0040] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "disposed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be an indirect connection through an intermediate medium, or it can be the internal communication of two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.
[0041] Embodiment 1:
[0042] A full acoustic optical fiber monitoring method based on machine learning in this embodiment, such as Figure 1As shown, it is a schematic diagram of the structure of the neural network model provided by the present invention. Among them, C1 is the first convolution layer, C2 is the second convolution layer, C3 is the third convolution layer, C4 is the fourth convolution layer, C5 is the fifth convolution layer, BN1 is the first batch of normalization layers, BN2 is the second batch of normalization layers, BN3 is the third batch of normalization layers, ReLU1 is the first activation function layer, ReLU2 is the second activation function layer, ReLU3 is the third activation function layer, Maxpooling is the maximum pooling layer, FC is the fully connected layer, Dropout is the first regularization layer, and Softmax is the loss function layer.
[0043] The full acoustic optical fiber monitoring method based on machine learning provided by the present invention comprises the following steps:
[0044] Use the GAS host to collect on-site sound data, denoise the sound data, and generate a model training set;
[0045] In the GAS host, capture sound data according to the CNN audio classifier, use the spectrogram to represent the sound data set, build a neural network model, use the neural network model to capture the sound data in the spectrogram, train the neural network model with the model training set, and preset the number of iterations and learning rate.
[0046] The neural network model is composed of a point convolution combination module of the upper layer and a double convolution pooling combination module of the lower layer connected in sequence. The point convolution combination module of the upper layer includes the first convolution layer, the second convolution layer, the first batch of normalization layers, the second batch of normalization layers, the first activation function layer, the second activation function layer, the maximum pooling layer and the feature multiplication layer.
[0047] The dual convolution pooling combination module includes a third convolution layer, a fourth convolution layer, a fifth convolution layer, a fully connected layer, a third batch-one normalization layer, a third activation function layer, a first regularization layer, and a loss function layer;
[0048] Input the sound data set into the neural network model in the structure of a data cube to train the neural network model, and generate a trained neural network model;
[0049] In actual scenarios, the GAS host is used to collect on-site sound data to generate a corresponding sound data set represented by a spectrogram. The corresponding sound data set is input into the trained neural network model, and the sound wave vibration conditions of the on-site environment of the location are restored through deep mining and analysis, and the on-site sound is restored for monitoring.
[0050] Connecting the first convolution layer and the second convolution layer in parallel in the point convolution combination module;
[0051] The first convolution layer is connected to the first batch of normalization layers and the first activation function layer in sequence, and a Relu activation function is set in the first activation function layer;
[0052] The second convolutional layer is sequentially connected to the second batch of normalization layers and the second activation function layer, and a Relu activation function is set in the second activation function layer;
[0053] The first activation function layer and the second activation function layer are connected through a feature multiplication layer, and the feature multiplication layer and the maximum pooling layer are connected.
[0054] The point convolution combination module is used to solve the disorder of the sound dataset and increase the local features of the sound dataset.
[0055] Methods for designing a dual convolutional pooling combination module include:
[0056] In the double convolution pooling combination module, the third convolution layer is sequentially connected to the third convolution layer, the fourth convolution layer, the fifth convolution layer, the fully connected layer, the third batch one normalization layer, the third activation function layer, the first regularization layer and the loss function layer;
[0057] The Relu activation function is set in the third activation function layer.
[0058] The dual convolutional pooling combination module is used to classify and display the sound dataset presented in the form of spectrograms.
[0059] After the sound dataset is input into the neural network model, the point convolution combination module is used to extract the local information of the sound dataset. At the same time, the upper point convolution combination module upsamples the sound data, and then deeply mines the sound dataset. Finally, the lower double convolution pooling combination module aggregates the feature information of the sound dataset, and then uses the Relu activation function in the third activation function layer and the softmax loss function in the loss function layer to calculate the loss value of the predicted point of the sound dataset and the real point of the sound dataset in the spectrogram. The learning rate descent method is used to iteratively attenuate the loss value and optimize the weight parameters of the neural network model until the number of iterations is equal to the maximum number of iterations. The training is terminated to generate a trained neural network model.
[0060] Embodiment 2:
[0061] This embodiment is further optimized on the basis of embodiment 1.
[0062] The other parts of this embodiment are the same as those of Embodiment 1, and thus will not be described in detail.
[0063] Embodiment 3:
[0064] This embodiment is further optimized on the basis of the above-mentioned embodiment 1 or 2. The GAS host designed by the present invention integrates machine learning technology, which can break through and solve the problems of multiple recognition types, high accuracy, and rapid response.
[0065] The other parts of this embodiment are the same as those of the above-mentioned embodiment 1 or 2, and thus will not be described in detail.
[0066] Embodiment 4:
[0067] This embodiment is further optimized on the basis of any one of the above embodiments 1-3. Given a large number of samples of a certain event of the GAS host, the GAS host uses a deep neural network model and a long short-term memory network algorithm to process the sample data, and after intelligent data analysis, accurately determines the event information. It has the ability of self-learning and self-optimization. The accuracy of GAS host recognition is optimized by the GAS host's continuous machine learning of large numbers of samples.
[0068] The other parts of this embodiment are the same as any one of the above embodiments 1-3, so they are not repeated here.
[0069] Embodiment 5:
[0070] This embodiment is further optimized on the basis of any one of the above embodiments 1-4. The full-acoustic optical fiber monitoring system provided by the present invention restores the signal sent back by the optical fiber sensor, analyzes the acoustic vibration of the environmental field at that location, and restores the on-site sound. The GAS host can achieve the purpose of reducing false alarms through continuous machine learning, thereby achieving the effect of reducing costs and increasing efficiency.
[0071] The other parts of this embodiment are the same as any one of the above embodiments 1-4, so they are not repeated here.
[0072] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Any simple modification or equivalent change made to the above embodiment based on the technical essence of the present invention shall fall within the protection scope of the present invention.
Claims
1. A full-acoustic optical fiber monitoring method based on machine learning, It is characterized in that The following steps are involved: Step S1, using the GAS host to collect on-site sound data, denoising the sound data, and generating a model training set; Step S2, in the GAS host, capture sound data according to the CNN audio classifier, use the spectrogram to represent the sound data set, build a neural network model, use the neural network model to capture the sound data in the spectrogram, train the neural network model with the model training set, and preset the number of iterations and the learning rate; The neural network model is composed of a point convolution combination module of the upper layer and a double convolution pooling combination module of the lower layer connected in sequence, wherein the point convolution combination module of the upper layer includes a first convolution layer, a second convolution layer, a first batch of normalization layers, a second batch of normalization layers, a first activation function layer, a second activation function layer, a maximum pooling layer and a feature multiplication layer; The dual convolution pooling combination module includes a third convolution layer, a fourth convolution layer, a fifth convolution layer, a fully connected layer, a third batch one normalization layer, a third activation function layer, a first regularization layer and a loss function layer; Step S3, inputting the sound data set into the neural network model in the structure of a data cube to train the neural network model, thereby generating a trained neural network model; Step S4, in the actual scenario, using the GAS host to collect on-site sound data to generate a corresponding sound data set represented by a spectrogram, inputting the corresponding sound data set into the trained neural network model, and restoring the sound wave vibration of the on-site environment where the on-site sound data is collected through deep mining and analysis, restoring the on-site sound and monitoring; The method for designing the point convolution combination module in step S2 includes: Connecting the first convolution layer and the second convolution layer in parallel in the point convolution combination module; The first convolution layer is connected to the first batch of normalization layers and the first activation function layer in sequence, and a Relu activation function is set in the first activation function layer; The second convolutional layer is sequentially connected to the second batch of normalization layers and the second activation function layer, and a Relu activation function is set in the second activation function layer; The first activation function layer and the second activation function layer are connected through a feature multiplication layer, and the feature multiplication layer and the maximum pooling layer are connected; The method for designing the dual convolution pooling combination module in step S2 includes: In the dual convolution pooling combination module, the third convolution layer is sequentially connected with the third convolution layer, the fourth convolution layer, the fifth convolution layer, the fully connected layer, the third batch one normalization layer, the third activation function layer, the first regularization layer, and the loss function layer; The Relu activation function is set in the third activation function layer.
2. According to claim 1, a full acoustic optical fiber monitoring method based on machine learning, It is characterized in that The point convolution combination module is used to solve the disorder of the sound data set and increase the local features of the sound data set.
3. The full acoustic optical fiber monitoring method based on machine learning according to claim 1, It is characterized in that The dual convolutional pooling combination module is used to classify and display the sound data set presented in the form of a spectrogram.
4. The full acoustic optical fiber monitoring method based on machine learning according to claim 1, It is characterized in that The step S3 comprises: After the sound dataset is input into the neural network model, the point convolution combination module is used to extract the local information of the sound dataset. At the same time, the upper point convolution combination module upsamples the sound data, and then deeply mines the sound dataset. Finally, the lower double convolution pooling combination module aggregates the feature information of the sound dataset, and then uses the Relu activation function in the third activation function layer and the softmax loss function in the loss function layer to calculate the loss value of the predicted point of the sound dataset and the real point of the sound dataset in the spectrogram. The learning rate descent method is used to iteratively attenuate the loss value and optimize the weight parameters of the neural network model until the number of iterations is equal to the maximum number of iterations. The training is terminated to generate a trained neural network model.
5. The full acoustic optical fiber monitoring method based on machine learning according to claim 1, It is characterized in that The step S4 comprises: The GAS host performs fiber optic monitoring based on the optical cable itself to complete the sound verification.
6. A full acoustic fiber optic monitoring system based on machine learning, It is characterized in that It includes a collection unit, a model building unit, a training unit and a monitoring unit, among which: The acquisition unit is used to collect on-site sound data using the GAS host, denoise the sound data, generate a model training set, capture the sound data according to the CNN audio classifier in the GAS host, and represent the sound data set using a spectrogram; A model building unit is used to build a neural network model, use the neural network model to capture sound data in the spectrogram, train the neural network model with a model training set, and preset the number of iterations and learning rate; The neural network model is composed of a point convolution combination module of an upper layer and a double convolution pooling combination module of a lower layer connected in sequence, wherein the point convolution combination module of the upper layer includes a first convolution layer, a second convolution layer, a first batch of normalization layers, a second batch of normalization layers, a first activation function layer, a second activation function layer, a maximum pooling layer and a feature multiplication layer; the first convolution layer and the second convolution layer are connected in parallel; the first convolution layer is sequentially connected to the first batch of normalization layers and the first activation function layer, and a Relu activation function is set in the first activation function layer; the second convolution layer is sequentially connected to the second batch of normalization layers and the second activation function layer, and a Relu activation function is set in the second activation function layer; the first activation function layer and the second activation function layer are connected through a feature multiplication layer, and the feature multiplication layer and the maximum pooling layer are connected; The dual convolution pooling combination module includes a third convolution layer, a fourth convolution layer, a fifth convolution layer, a fully connected layer, a third batch one normalization layer, a third activation function layer, a first regularization layer, and a loss function layer connected in sequence; a Relu activation function is set in the third activation function layer; A training unit, used for inputting the sound data set into the neural network model in the structure of a data cube to train the neural network model and generate a trained neural network model; The monitoring unit is used to use the GAS host to collect on-site sound data in actual scenarios to generate a corresponding sound data set represented by a spectrogram, input the corresponding sound data set into the trained neural network model, and restore the sound wave vibration conditions of the on-site environment where the on-site sound data is collected through deep mining and analysis, restore the on-site sound and monitor it.
7. A computer-readable storage medium having computer program instructions stored thereon, It is characterized in that When the program instructions are executed by a processor, the steps of the method described in any one of claims 1 to 5 are implemented.
