Aerial optical fiber sensor event identification method and system

Abstract

The invention relates to an overhead optical fiber sensor event identification method and system, and belongs to the field of cable online monitoring, and the method comprises the steps: obtaining a vibration signal at a corresponding position through an optical fiber sensor; decomposing the vibration signal through a wavelet packet decomposition method to obtain a plurality of signal components of different frequency bands, weighting each signal component, and performing inverse transformation to obtain a reconstructed signal; extracting a vibration energy value at each sampling point from the reconstructed signal to generate a thermodynamic diagram; and carrying out image identification and classification on the thermodynamic diagram to obtain a sensor event and an event category. Vibration signals are processed through a wavelet packet decomposition method to obtain signal components of different frequency bands, environment noise in a low-frequency band is weakened through weighting to highlight event noise in a high-frequency band, a reconstructed signal is obtained through inverse transformation, a thermodynamic diagram is generated by using the reconstructed signal, and the performance of the thermodynamic diagram is improved. The area corresponding to the event in the thermodynamic diagram is identified and analyzed through the image identification algorithm, so that event identification is realized, a threshold value does not need to be set, the adaptability is better, and the efficiency is higher.

Description

Technical Field

[0001] The present application relates to the field of online cable monitoring, and in particular to an overhead optical fiber sensor event recognition method and system. Background Art

[0002] With the rapid development of fiber optic sensing technology, distributed fiber optic acoustic wave sensing technology has been widely used by researchers in the field of cable online monitoring due to its advantages such as long distance, high sensitivity, and anti-interference. However, in field applications, due to the complex application environment and variable intrusion methods, false alarms and missed alarms often occur, especially in the overhead fiber optic scenario. Compared with the buried fiber optic data, the background noise of the overhead fiber optic data is more. The simple extraction of time-frequency features or the use of conventional thresholds and other methods to filter noise cannot meet the actual needs. The main event features are difficult to extract, and the adaptability to the recognition of different events in different scenarios is poor, resulting in false alarms and missed alarms for events occurring around the cable.

[0003] In addition, on the one hand, since the sensors collect data quickly and the data volume is relatively large, especially the sensors on long-distance optical fibers have a sampling rate of 4000 Hz or even higher, which can collect 4000 vibration signal data per second or even more. Data reception and preprocessing (such as signal noise reduction) are time-consuming. On the other hand, due to the complex data collection background of overhead data, the collected vibration signal data contains a lot of noise (for example, natural environment interference noise such as wind and rain will be collected by sensors). Later, algorithms are needed to analyze event feature data from these vibration signal data, such as feature data of events such as animal groups invading optical fibers and human excavation of optical fibers. The calculation amount is large and the calculation process is time-consuming.

[0004] Conventional methods such as manually setting thresholds, database comparison, deep learning network recognition and classification, etc., traverse all event data for comparison or recognition and classification, resulting in a huge amount of calculation, reducing recognition efficiency and making it impossible to guarantee recognition accuracy. Based on the above two aspects, the current sensor event recognition technology for overhead optical fiber scenes faces efficiency and accuracy issues, resulting in high alarm delays, slow speeds, false alarms, missed alarms and other problems.

[0005] Fiber optic sensor event recognition technology is a technology that combines the high-precision measurement capabilities of fiber optic sensors with the intelligent analysis capabilities of pattern recognition algorithms to identify and classify different physical phenomena or events. This technology has applications in many fields, such as perimeter security, pipeline monitoring, and railway safety monitoring. Currently, existing fiber optic sensor event recognition technology methods are mostly targeted at buried optical fibers. After denoising using methods such as Fourier transform or wavelet transform combined with manually set thresholds, events are identified and classified through database comparison or deep learning networks.

[0006] However, for overhead optical fibers, the background noise is complex and has a great impact on the data collected by optical fiber sensors. For example, background noises such as natural wind, snow, and ice have a great impact on event sounds. It is difficult to eliminate complex background noises in different environments by combining Fourier transform or wavelet transform with manually set thresholds. Especially with the changes in seasonal environment, background noise has a greater impact on the event recognition of overhead optical fibers. When there are many different events in different scenes around the optical fiber, the event database feature matching method can only identify the event features already in the database, which has great limitations and a very high false alarm rate. When the amount of scene data is large, the comparison efficiency is low and the alarm delay is greatly increased.

[0007] Using deep learning networks to identify events has a large number of parameters and takes up a lot of resources. It also requires converting fiber optic sensor data into images and dividing the images into multiple windows, which need to be traversed one by one for identification and classification. In actual scenarios, events may only occur at a certain location at a certain time. When no events occur, the deep learning network will generate a lot of useless work, waste resources and have low efficiency. Summary of the invention

[0008] The present application provides an overhead optical fiber sensor event recognition method and system, which can solve the technical problem of low efficiency of overhead optical fiber sensor event recognition in the prior art.

[0009] In a first aspect, an embodiment of the present application provides an aerial optical fiber sensor event recognition method, the method comprising:

[0010] Acquiring vibration signals at corresponding positions using optical fiber sensors arranged at preset sampling points;

[0011] Decomposing the vibration signal by wavelet packet decomposition method to obtain a plurality of signal components of different frequency bands, weighting each signal component and then performing inverse transformation to obtain a reconstructed signal; the weight value increases with the increase of the frequency band, the weight value corresponding to the minimum frequency band is 0, and the weight value corresponding to the maximum frequency band is 1;

[0012] Extracting the vibration energy value at each sampling point from the reconstructed signal, using the vibration energy value as a thermal value, and generating a thermal map;

[0013] Image recognition and classification are performed on the heat map to extract sensor events and classify them into event categories.

[0014] In combination with the first aspect, in one implementation, the step of acquiring the vibration signal at the corresponding position by using the optical fiber sensor arranged at the preset sampling point specifically includes the following steps:

[0015] Real-time collection of sensor data collected by optical fiber sensors set at preset sampling points;

[0016] The sensor data is analyzed by multi-threaded parallel processing to obtain the vibration signal;

[0017] The vibration signals are stored in the order of acquisition time.

[0018] In combination with the first aspect, in one implementation, when the vibration signal is decomposed by a wavelet packet decomposition method, a multi-threaded parallel processing method is used to obtain the signal components.

[0019] In combination with the first aspect, in one implementation, the total number of the frequency bands obtained by decomposition is proportional to the number of types of medium and high frequency noise in the environment where the overhead optical fiber is located.

[0020] In combination with the first aspect, in one implementation, the weighting of each signal component and then performing inverse transformation to obtain a reconstructed signal specifically includes the following steps:

[0021] A new time-vibration energy value sequence is obtained by extracting the maximum value of the vibration energy value of the sampling point in a preset time period from each signal component;

[0022] An inverse transformation is performed according to the new time-vibration energy value sequence to obtain the reconstructed signal.

[0023] In combination with the first aspect, in one implementation, the generation of the heat map specifically includes the following steps:

[0024] Extracting the original vibration energy value at each sampling point from the reconstructed signal;

[0025] Normalize the original vibration energy value to obtain the final vibration energy value;

[0026] According to the final vibration energy value, a heat map including time, sampling point location, and final vibration energy value information is generated.

[0027] In combination with the first aspect, in one implementation, the obtaining of the sensor event and the event category specifically includes the following steps:

[0028] Performing edge detection on the heat map to obtain relevant information of a preset color area, wherein the relevant information includes time information, location information, and color information;

[0029] Event identification and classification are performed according to the relevant information to obtain corresponding sensor events and event categories.

[0030] In combination with the first aspect, in one implementation, the Sobel operator is used to perform edge detection on the heat map to obtain the relevant information.

[0031] In combination with the first aspect, in one implementation, a ResNet18 network is used to perform event recognition and classification on the relevant information to obtain corresponding sensor events and event categories.

[0032] In a second aspect, an embodiment of the present application provides an overhead optical fiber sensor event recognition system, the system comprising:

[0033] A data acquisition module, which uses an optical fiber sensor arranged at a preset sampling point to acquire a vibration signal at a corresponding position;

[0034] A data processing module, which is used to decompose the vibration signal by wavelet packet decomposition method to obtain a plurality of signal components of different frequency bands, perform inverse transformation after weighting each signal component, and obtain a reconstructed signal; the weight value increases with the increase of the frequency band, the weight value corresponding to the minimum frequency band is 0, and the weight value corresponding to the maximum frequency band is 1; extract the vibration energy value at each sampling point from the reconstructed signal, use the vibration energy value as the thermal value, and generate a thermal map;

[0035] The data analysis module is used to perform image recognition and classification on the thermal map to extract sensor events and classify them into event categories.

[0036] The beneficial effects brought by the technical solution provided in the embodiments of the present application include:

[0037] The vibration signal is processed by wavelet packet decomposition method to obtain signal components in different frequency bands. The environmental noise in the low frequency band is weakened by weighting to highlight the event noise in the high frequency band. The reconstructed signal is obtained by inverse transformation. The reconstructed signal is used to generate a heat map. The area corresponding to the event in the heat map is identified and analyzed by the image recognition algorithm, thereby realizing event recognition. There is no need to set a threshold, which has better adaptability and higher efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a flow chart of an embodiment of an overhead optical fiber sensor event identification method of the present application;

[0039] Figure 2 for Figure 1 Schematic diagram of the process of step S1;

[0040] Figure 3 for Figure 1 Schematic diagram of the process of step S3;

[0041] Figure 4 for Figure 1 Schematic diagram of the process of step S4;

[0042] Figure 5 for Figure 1 Schematic diagram of the process of step S5;

[0043] Figure 6 This is a schematic diagram of the architecture of an embodiment of an overhead optical fiber sensor event identification system of the present application. DETAILED DESCRIPTION

[0044] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0045] In the prior art, the fiber vibration signal data collected by the fiber sensor is generally subjected to Fourier transform or traditional wavelet transform, and then noise is reduced by manually setting a threshold. Then, the signal waveform drawn by the time domain diagram is compared with the event signal waveform defined in the database, so as to identify the event according to the degree of similarity. Both Fourier transform and traditional wavelet transform methods need to be combined with manually setting thresholds to remove noise, and the effect of removing noise in overhead complex environments is poor. Fourier transform only converts the signal into time domain or frequency domain and loses a lot of signal feature information. Wavelet transform is based on time-frequency analysis method, and convolves the signal with wavelet functions of different scales and positions to extract the signal features. It only decomposes the signal into low-frequency and high-frequency parts, and cannot focus on the more detailed frequency features of the signal. The resolution of the high-frequency part is not as good as that of the low-frequency part, and it is especially difficult to capture the finer features of the high-frequency part. Moreover, the noise is removed by manually setting thresholds, which cannot adapt to various environments, especially the field of overhead optical fibers. In addition, the use of large models for traversing and identifying a large amount of fiber sensor data is inefficient, and it is impossible to quickly identify and achieve the effect of real-time alarm. Moreover, the computing resources are more occupied, and certain computing resource conditions are required for practical applications.

[0046] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0047] In a first aspect, an embodiment of the present application provides an overhead optical fiber sensor event recognition method.

[0048] In one embodiment, referring to Figure 1 , Figure 1 This is a flow chart of an embodiment of the overhead optical fiber sensor event recognition method of the present application. Figure 1 As shown, the overhead optical fiber sensor event recognition method includes:

[0049] Step S1: Acquire a vibration signal at a corresponding position using an optical fiber sensor arranged at a preset sampling point.

[0050] Step S2: Decompose the vibration signal by wavelet packet decomposition method to obtain signal components of multiple different frequency bands.

[0051] Step S3: weight each signal component and then perform inverse transformation to obtain a reconstructed signal. The weight value increases with the increase of the frequency band. The weight value corresponding to the minimum frequency band is 0, and the weight value corresponding to the maximum frequency band is 1.

[0052] Step S4: extract the vibration energy value at each sampling point from the reconstructed signal, use the vibration energy value as the thermal value, and generate a thermal map.

[0053] Step S5: perform image recognition and classification on the heat map to extract sensor events and classify them to obtain event categories.

[0054] In this embodiment, the present invention introduces a three-level detection concept: first-level noise reduction, second-level recognition, and third-level classification. The signal characteristics are gradually made obvious through step-by-step signal processing, and the calculation time at each level is reduced.

[0055] In the first-level noise reduction stage, based on the wavelet packet decomposition method, the vibration signal is decomposed into signal components of different frequency bands, and each frequency band is weighted according to the frequency, and a weight in the interval [0,1] is assigned to each frequency band. The weight value assigned from the lowest frequency band to the highest frequency band increases successively. Since background noise, such as wind and ice sound, is generally low-frequency noise, and event noise, such as animal damage to optical fiber and vehicle damage to optical fiber, is generally high-frequency noise, after assigning the above weights to each frequency band, the low-frequency environmental noise is multiplied by a smaller weight, which is equivalent to being weakened, while the high-frequency event noise is multiplied by a higher weight and is still retained. After the inverse transform of the wavelet packet is performed on the weighted frequency band components to obtain the reconstructed signal, it is equivalent to filtering the low-frequency signal, which is beneficial to subsequent event recognition.

[0056] Compared with methods such as Fourier transform or wavelet transform, the improved wavelet packet decomposition and reconstruction process can perform more detailed time-frequency feature analysis of the low-frequency and high-frequency parts of the signal, remove low-frequency noise through dynamic weighting, retain and reduce some low-frequency signal features, and extract signals with high-frequency detail information. There is no need to filter out noise by manually setting thresholds, so it has better adaptability and higher recognition accuracy.

[0057] In the secondary recognition stage, the vibration energy value at each sampling point is extracted from the above-mentioned reconstructed signal, and the vibration energy value is used as the thermal value to generate a thermal map. The thermal map can be a static map corresponding to a certain sampling moment, or a dynamic map corresponding to multiple sampling moments. Each sampling moment corresponds to a frame of image. Based on the image, the image recognition method is used to extract the area of ​​specific color from the image, and the event feature area can be identified. Compared with the technical solution of extracting the time-frequency characteristics of the signal through a large model in the prior art, constructing a sliding window to slide on each scanning frame, generating overlapping time-space windows respectively, and then detecting and classifying intrusion events based on the convolutional neural network model of S transformation, for complex overhead optical fiber sensor data, due to the presence of a large amount of complex background noise, the direct extraction of time-frequency features is not effective, and traversing all time-space windows for detection and classification is inefficient, and the accuracy needs to be considered. The present invention extracts specific color areas in the thermal map, and performs event recognition and classification according to the extraction results, which has less calculation and higher efficiency.

[0058] In the three-level classification stage, the event type can be analyzed based on the color and shape of the identified event feature area.

[0059] In summary, the vibration signal is processed by wavelet packet decomposition method to obtain signal components in different frequency bands. The environmental noise in the low frequency band is weakened by weighting to highlight the event noise in the high frequency band. The reconstructed signal is obtained by inverse transformation. The reconstructed signal is used to generate a heat map. The area corresponding to the event in the heat map is identified and analyzed by the image recognition algorithm, thereby realizing event recognition. There is no need to set a threshold, which has better adaptability and higher efficiency.

[0060] Further, in one embodiment, if Figure 2 As shown, the above step S1 specifically includes the following steps:

[0061] Step S11 : collecting sensor data collected by the optical fiber sensor set at the preset sampling point in real time.

[0062] Step S12: Analyze the sensor data using a multi-threaded parallel processing method to obtain a vibration signal.

[0063] Step S13: store the vibration signals in the order of acquisition time.

[0064] In this embodiment, the original optical fiber vibration signal data collected by the optical fiber sensor set at each sampling point is received in real time and read in real time, increasing the timeliness of data reception to reduce the delay of event alarms at different positions of the optical fiber. The received data is parsed in batches using a multi-threaded parallel method, and the collected data information contained in each received data is parsed according to the protocol of the collection device, including data collection time, data collection location, and collected optical fiber vibration signal data, such as the vibration energy value collected by the optical fiber sensor, and is batch cached in the system in chronological order.

[0065] Furthermore, in one embodiment, when the vibration signal is decomposed by wavelet packet decomposition method, a multi-threaded parallel processing method is used to obtain the signal components.

[0066] In this embodiment, when the cached vibration signal reaches a batch processing data volume, it is subjected to wavelet packet decomposition. In order to effectively extract the characteristics of the event and consider the real-time alarm of the event, the batch data volume is generally one tenth of the sampling rate (unit Hz) of the sensor device to collect data.

[0067] Then, the vibration energy values ​​of each position point in a batch of vibration signals at different times form a vibration signal time series for each position point. Using a multi-threaded parallel method, the wavelet packet decomposition algorithm is used to decompose them to obtain the above signal components, which can improve data processing efficiency.

[0068] Furthermore, in one embodiment, the total number of the above-mentioned frequency bands obtained by decomposition is proportional to the number of types of medium and high frequency noises in the environment where the overhead optical fiber is located.

[0069] In this embodiment, the vibration energy values ​​of each position point at different times in a batch of vibration signals are used to form a vibration signal time series of each position point. The multi-threaded parallel method is used to decompose the data into n layers through the wavelet packet decomposition algorithm to obtain 2 n The number of decomposition layers n can be adjusted in special scenarios. For example, in a particularly harsh and complex environment, when the frequency bands of the event to be concerned are similar to those of the background noise, resulting in the frequency band decomposition being too coarse and unable to effectively filter out the noise similar to the event, the number of decomposition layers can be increased. For the decomposed signal components, weights are dynamically assigned from low to high frequency bands. The weights are in the interval [0,1], starting from 0 and ending at 1. The intermediate weights are randomly assigned decimals from 0 to 1 and gradually increase. That is, each dynamic weight ensures that the weight of the lowest frequency band is 0, the weight of the highest frequency band is 1, and the weights of the intermediate frequency bands are randomly assigned decimals between 0 and 1, such as 0.1, 0.2, 0.5, etc., but increase according to the frequency band.

[0070] The fiber optic sensor data feature map is extracted through wavelet packet decomposition and reconstruction, without the need for manual threshold setting and other operations, thus avoiding the impact of the threshold changes in different scenarios on the recognition accuracy and increasing the generalization of event recognition.

[0071] Further, in one embodiment, if Figure 3 As shown, the above step S3 specifically includes the following steps:

[0072] Step S31 : obtaining a new time-vibration energy value sequence by extracting the maximum value of the vibration energy value of the sampling point in a preset time period from each signal component.

[0073] Step S32: perform inverse transformation according to the new time-vibration energy value sequence to obtain a reconstructed signal.

[0074] In this embodiment, assuming that the original vibration signal contains the vibration energy values ​​of sampling point A, sampling point B, and sampling point C at time T1, time T2, and time T3, step S31 is equivalent to extracting the maximum vibration energy value of sampling point A, sampling point B, and sampling point C at time T1-T3, so as to characterize the vibration characteristics at the position point with the maximum mutation value of each sampling point in the specified time period. After a batch of data is processed, a characteristic value of each position point is obtained to form the optical fiber vibration data of a new time point, that is, the above-mentioned reconstructed signal.

[0075] Further, in one embodiment, if Figure 4 As shown, step S4 specifically includes the following steps:

[0076] Step S41: extracting the original vibration energy value at each sampling point from the reconstructed signal.

[0077] Step S42: normalize the original vibration energy value to obtain a final vibration energy value.

[0078] Step S43: Generate a heat map including time, sampling point location, and final vibration energy value information according to the final vibration energy value.

[0079] In this embodiment, multiple batches of data are processed in the above manner, and the multiple batches of data constitute multiple time period data of multiple position points of the optical fiber. The above reconstructed signals are globally normalized, and the values ​​are mapped to the interval of 0 to 1. The drawing toolkit matlibplot is used to map the normalized data to a two-dimensional RGB image, that is, a time-sampling point-energy heat map. The highlighted red area in the heat map indicates the area where the event occurs, and different events have different area and color characteristics.

[0080] Further, in one embodiment, if Figure 5 As shown, step S5 specifically includes the following steps:

[0081] Step S51: perform edge detection on the heat map to obtain relevant information of a preset color area, wherein the relevant information includes time information, location information, and color information.

[0082] Step S52: Perform event identification and classification based on relevant information to obtain corresponding sensor events and event categories.

[0083] In this embodiment, the time interval and event area of ​​the location event can be accurately identified through edge detection.

[0084] Furthermore, in one embodiment, the Sobel operator is used to perform the above-mentioned edge detection on the heat map to obtain the above-mentioned related information.

[0085] In this embodiment, an edge detection algorithm based on the Sobel operator is used to identify the prominent red area in the heat map, that is, to identify the event area. The precise time interval and location area where the event occurred are identified through the detected area to narrow the event feature area.

[0086] The Sobel operator weights the pixel position to make edge positioning more accurate than the Prewitt operator. The Sobel operator takes into account the influence of the pixel position, so in the case of more noise, its edge detection effect is usually better than the Prewitt operator.

[0087] The first-level noise reduction aims to remove the influence of natural noise in low-frequency bands such as wind and ice, retain the target event features in high-frequency bands such as mechanical intrusion, and retain and reduce the weight of the intermediate frequency band features, laying the foundation for the second-level recognition, removing the signal background noise energy value, reducing the signal energy value of possible secondary events, and extracting the signal energy value of the main event of concern in the signal. The second-level recognition aims to unify the signal energy value scale and use edge detection to accurately identify the time interval and location interval of the event.

[0088] Furthermore, in one embodiment, the ResNet18 network is used to perform event recognition and classification on the above-mentioned related information to obtain corresponding sensor events and event categories.

[0089] In this embodiment, the identified event areas are classified based on the lightweight ResNet18, and the identified areas and classification results are combined to intuitively display the event occurrence in the form of a dynamic waterfall chart, so as to detect events and issue alarms in real time.

[0090] Compared with general deep convolutional networks, the ResNet18 network has fewer network layers and parameters, occupies less resources, has a faster recognition speed, and only classifies areas where events have been identified, which increases classification efficiency and reduces resource waste in identifying areas where no events have occurred.

[0091] The three-level classification is designed to quickly categorize and dynamically display incidents in real time.

[0092] In a specific embodiment, based on the lightweight ResNet18 network, it is necessary to pre-train and save the model weights so as to load the model to predict the classification in the classification stage. The detailed steps are as follows:

[0093] A training data set is created based on the heat map, which is uniformly processed into images with a length and width of 256 pixels. The event categories to which the feature maps belong are manually labeled and divided into training and test sets.

[0094] Build a ResNet18 network, load pre-trained weights, and input training and test sets for training and testing.

[0095] After training is completed, the optimal model weights are saved.

[0096] In the classification stage, the optimal model weights are loaded to predict and classify the above related information.

[0097] In a second aspect, an embodiment of the present application also provides an overhead optical fiber sensor event identification system.

[0098] In one embodiment, referring to Figure 6 , Figure 6 This is a functional module diagram of an embodiment of the overhead optical fiber sensor event recognition system of the present application. Figure 6 As shown, the overhead optical fiber sensor event identification system includes:

[0099] The data acquisition module 1 uses an optical fiber sensor arranged at a preset sampling point to acquire a vibration signal at a corresponding position.

[0100] Data processing module 2 is used to decompose the vibration signal by wavelet packet decomposition method to obtain signal components of multiple different frequency bands, and perform inverse transformation after weighting each signal component to obtain a reconstructed signal. The weight value increases with the increase of the frequency band, the weight value corresponding to the minimum frequency band is 0, and the weight value corresponding to the maximum frequency band is 1. The vibration energy value at each sampling point is extracted from the reconstructed signal, and the vibration energy value is used as the thermal value to generate a thermal map.

[0101] The data analysis module 3 is used to perform image recognition and classification on the thermal map to extract sensor events and classify them to obtain event categories.

[0102] In this embodiment, there is no need to manually set the threshold. The method of reconstructing the vibration signal components by assigning dynamic weights through wavelet packet decomposition has better adaptability to complex overhead optical fiber scenes, and the accuracy of optical fiber sensor event recognition and classification in various environments is higher.

[0103] Using the three-level detection idea, through the first-level noise reduction, second-level recognition, and third-level classification collaboration, although the system resource occupancy and computational complexity of theoretical methods data acquisition and analysis, wavelet packet reconstruction, edge detection, and convolutional neural network recognition increase step by step, after the three-level detection collaborative system, the data complexity decreases step by step, the event characteristics become more obvious step by step, and the amount of data required for calculation decreases step by step, so the overall system efficiency is greatly improved and the resource occupancy rate is reduced.

[0104] In a specific embodiment, a data collection and storage directory of the overhead optical fiber sensor is set or the optical fiber sensor device is connected via a Socket to obtain data.

[0105] When the system of the present invention is started, the first-level noise reduction part of the system starts to work, real-time monitoring directory or receiving data collected by the optical fiber sensor through Socket connection, and real-time reading of the original data of the sensor.

[0106] The system reads the raw data according to the distributed parsing protocol, parses out the data collection time, the value of each collection point, the collection frequency and other information, and caches the parsed data into the system in chronological order according to batches, with a batch size of one tenth of the collection frequency.

[0107] For batch cache data, the wavelet packet decomposition algorithm is used to decompose the data of each position point into 3 layers in parallel through the thread pool, and the components of 8 frequency bands are obtained. The frequency bands are sorted according to the frequency. The weight of the lowest frequency band is assigned to 0, the highest frequency band is assigned to 1, and the weight of the middle frequency band is assigned to a random decimal between 0 and 1, and it increases in the order of the frequency band. After multiplying the corresponding weight and the frequency band, the original signal is reconstructed using the inverse transform method of wavelet packet decomposition. The maximum mutation energy value of each position point in each batch is taken as the energy value of this position at this batch time.

[0108] The secondary identification part of the system starts working. After multiple batches of data processing, multiple batches of data are obtained to form multiple time point data of multiple positions of the optical fiber, that is, a two-dimensional matrix of time-position point-energy value. The above vibration data is globally normalized and the value is mapped to the range of 0 to 1. The drawing toolkit matlibplot is used to map the normalized data to a two-dimensional RGB image, that is, a time-sampling point-energy heat map. The highlighted red area in the heat map indicates the area where the event occurred. The Sobel operator-based edge detection algorithm is used to detect the highlighted red area in the heat map, and remember the time interval and position area where the event occurred.

[0109] The three-level classification part of the system starts working, and the ResNet18 network parallelly classifies the event area heat map part, recording the event category, occurrence area and occurrence time interval.

[0110] Based on the recorded event information, the event occurrence is intuitively displayed in the form of a waterfall chart, and event alarms are issued.

[0111] Among them, the functional implementation of each module in the above-mentioned overhead optical fiber sensor event identification system corresponds to the various steps in the above-mentioned overhead optical fiber sensor event identification method embodiment, and its functions and implementation processes will not be repeated here one by one.

[0112] It should be noted that the serial numbers of the above-mentioned embodiments of the present application are only for description and do not represent the advantages or disadvantages of the embodiments.

[0113] The terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally includes other steps or units inherent to these processes, methods, products or devices. The terms "first", "second" and "third" are used to distinguish different objects, etc., and do not represent a sequence, nor do they limit "first", "second" and "third" to different types.

[0114] In the description of the embodiments of the present application, "exemplary", "for example" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary", "for example" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary", "for example" or "for example" is intended to present related concepts in a specific way.

[0115] In the description of the embodiments of the present application, unless otherwise specified, “ / ” means or, for example, A / B can mean A or B; the “and / or” in the text is merely a description of the association relationship of associated objects, indicating that three relationships may exist, for example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the description of the embodiments of the present application, “multiple” refers to two or more than two.

[0116] In some processes described in the embodiments of the present application, multiple operations or steps that appear in a specific order are included, but it should be understood that these operations or steps may not be executed in the order in which they appear in the embodiments of the present application or in parallel, and the sequence number of the operation is only used to distinguish the different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed in sequence or in parallel, and these operations or steps may be combined.

[0117] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, disk, CD) as described above, and includes a number of instructions for a terminal device to execute the methods described in each embodiment of the present application.

[0118] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. An overhead optical fiber sensor event recognition method, characterized in that: The method comprises: Acquiring vibration signals at corresponding positions using optical fiber sensors arranged at preset sampling points; Decomposing the vibration signal by wavelet packet decomposition method to obtain a plurality of signal components of different frequency bands, weighting each signal component and then performing inverse transformation to obtain a reconstructed signal; the weight value increases with the increase of the frequency band, the weight value corresponding to the minimum frequency band is 0, and the weight value corresponding to the maximum frequency band is 1; Extracting the vibration energy value at each sampling point from the reconstructed signal, using the vibration energy value as a thermal value, and generating a thermal map; Image recognition and classification are performed on the heat map to extract sensor events and classify them into event categories.

2. The aerial optical fiber sensor event recognition method according to claim 1, characterized in that: The method of obtaining the vibration signal at the corresponding position by using the optical fiber sensor arranged at the preset sampling point specifically comprises the following steps: Real-time collection of sensor data collected by optical fiber sensors set at preset sampling points; The sensor data is analyzed by multi-threaded parallel processing to obtain the vibration signal; The vibration signals are stored in the order of acquisition time.

3. The aerial optical fiber sensor event recognition method according to claim 1, characterized in that: When the vibration signal is decomposed by the wavelet packet decomposition method, the signal components are obtained by using a multi-threaded parallel processing method.

4. The aerial optical fiber sensor event recognition method according to claim 1, characterized in that: The total number of the frequency bands obtained by decomposition is proportional to the number of types of medium and high frequency noises in the environment where the overhead optical fiber is located.

5. The aerial optical fiber sensor event recognition method according to claim 1, characterized in that: The weighting of each signal component and then performing inverse transformation to obtain a reconstructed signal specifically includes the following steps: A new time-vibration energy value sequence is obtained by extracting the maximum value of the vibration energy value of the sampling point in a preset time period from each signal component; An inverse transformation is performed according to the new time-vibration energy value sequence to obtain the reconstructed signal.

6. The aerial optical fiber sensor event recognition method according to claim 1, characterized in that: The generation of the heat map specifically includes the following steps: Extracting the original vibration energy value at each sampling point from the reconstructed signal; Normalize the original vibration energy value to obtain the final vibration energy value; According to the final vibration energy value, a heat map including time, sampling point location, and final vibration energy value information is generated.

7. The aerial optical fiber sensor event recognition method according to claim 1, characterized in that: The acquisition of the sensor events and event categories specifically includes the following steps: Performing edge detection on the heat map to obtain relevant information of a preset color area, wherein the relevant information includes time information, location information, and color information; Event identification and classification are performed according to the relevant information to obtain corresponding sensor events and event categories.

8. The aerial optical fiber sensor event recognition method according to claim 7, characterized in that: The Sobel operator is used to perform edge detection on the heat map to obtain the relevant information.

9. The aerial optical fiber sensor event recognition method according to claim 7, characterized in that: The ResNet18 network is used to perform event recognition and classification on the relevant information to obtain corresponding sensor events and event categories.

10. An overhead optical fiber sensor event recognition system, characterized in that: The system comprises: A data acquisition module, which uses an optical fiber sensor arranged at a preset sampling point to acquire a vibration signal at a corresponding position; A data processing module, which is used to decompose the vibration signal by wavelet packet decomposition method to obtain a plurality of signal components of different frequency bands, perform inverse transformation after weighting each signal component, and obtain a reconstructed signal; the weight value increases with the increase of the frequency band, the weight value corresponding to the minimum frequency band is 0, and the weight value corresponding to the maximum frequency band is 1; extract the vibration energy value at each sampling point from the reconstructed signal, use the vibration energy value as the thermal value, and generate a thermal map; The data analysis module is used to perform image recognition and classification on the thermal map to extract sensor events and classify them into event categories.

Images