A highway vehicle detection and identification method based on distributed optical fiber sensing

Abstract

The present application relates to the technical field of vehicle detection and identification, and discloses a method for detecting and identifying vehicles on highways based on distributed optical fiber sensing. The method includes: when a vehicle is traveling on a detection section, a distributed optical fiber sensor collects sensing signals, generates first data after preprocessing; divides all extended feature information into multiple threshold categories based on a multi-level binarization technology, and generates a feature data table based on type labels after marking the threshold categories of the extended feature information; generates a driving data table based on type labels by summarizing all third data; fuses the feature data table and the driving data table to generate a comprehensive data table, and identifies the foundation type of the detection section; constructs a detection and identification model, inputs the extended feature information of the vehicle to be monitored into the detection and identification model, outputs the vehicle information of the vehicle to be monitored, and completes the detection and identification of the vehicle to be monitored. The present application achieves high precision, high efficiency and high reliability in the detection and identification of highway vehicles.

Description

Technical Field

[0001] The present application relates to the technical field of vehicle detection and identification, and in particular to a method for detecting and identifying vehicles on a highway based on distributed optical fiber sensing. Background Art

[0002] In modern traffic management, accurate detection and identification of highway vehicles is crucial. Traditional vehicle detection and identification methods mainly include electromagnetic induction coils, video surveillance, and microwave radar.

[0003] Among them, the electromagnetic induction coil needs to be constructed and installed on the road, which causes damage to the road surface, and is easily affected by the environment and vehicle wear and tear, with high maintenance costs and limited service life. The detection effect of video surveillance technology will be greatly reduced under severe weather conditions, such as fog and heavy rain, and the video data processing volume is large, which requires high computing resources. Microwave radar is prone to signal interference during multi-lane detection, and the recognition accuracy of complex vehicle types and driving conditions needs to be improved. In addition, most of these traditional methods can only provide local detection information, making it difficult to achieve continuous monitoring of the entire road section. For long-distance and complex traffic scenes such as highways, it is impossible to fully and timely grasp the operating status of vehicles. Therefore, with the continuous increase in traffic volume on highways, higher requirements are placed on the accuracy, real-time performance and reliability of vehicle detection and recognition.

[0004] Similar prior art includes a Chinese patent application with publication number CN107256635A, which discloses a vehicle identification method based on distributed fiber optic sensing in intelligent transportation. This method utilizes existing communication optical fibers on the roadside or lays optical fibers on the roadside, connects the optical fiber vibration detection sensor unit to the roadside optical fiber, forms a distributed fiber optic sensing system, and uses the system to collect vibration information of vehicles moving near the optical fiber. The collected vibration signal is subjected to time domain analysis, frequency domain analysis, and time-frequency combined analysis to extract features that can characterize vehicle vibration and form a feature vector. Through the method of pattern recognition, the extracted feature vector is used to identify traffic statistics such as the type, location, and number of vibration targets. This invention overcomes the blind spots of traditional vehicle detection and improves the recognition rate and reliability of vehicle detection in intelligent transportation. There is also a Chinese patent application with publication number CN115050189A, which provides a vehicle identification and lane positioning method and system based on distributed fiber optic sensing. By laying two sensor optical fibers on both sides of the road, and simultaneously receiving the Rayleigh scattered light signals in the two sensor optical fibers containing vehicle vibration signals, combined with the attenuation law of the vehicle vibration signal propagating along the ground and calling GPU parallel acceleration calculation for real-time data processing and demodulation, it can overcome the shortcomings of traditional highway cameras such as high cost, small coverage area, and susceptibility to weather factors such as night and fog, and accurately obtain information such as vehicle location, driving speed, lane and traffic flow, and realize long-distance, all-weather real-time monitoring of highway vehicles.

[0005] The shortcomings of the existing technology are that the efficiency of vehicle detection is low and the vehicle vibration information is not classified and analyzed, which reduces the diversity of vehicle detection and reduces the monitoring efficiency of highway vehicles. Summary of the invention

[0006] The present application provides a method for detecting and identifying vehicles on a highway based on distributed optical fiber sensing, which is used to improve the efficiency and accuracy of detecting and identifying vehicles on a highway based on distributed optical fiber sensing.

[0007] In a first aspect, the present application provides a method for detecting and identifying highway vehicles based on distributed optical fiber sensing, comprising:

[0008] A plurality of distributed optical fiber sensors are laid on the detection section, and type labels of all vehicles are marked. When the vehicle travels on the detection section, the distributed optical fiber sensors collect sensing signals, and the sensing signals are pre-processed to generate first data;

[0009] The first data is optimized and filtered to generate second data, extended feature information is extracted based on the spatial resolution of the second data, all the extended feature information is divided into multiple threshold categories based on a multi-level binarization technology, and the extended feature information is marked with the threshold category and then summarized based on the type label to generate a feature data table;

[0010] Acquire driving data of the vehicle, generate third data after filtering the driving data, and generate a driving data table by aggregating all the third data based on the type label;

[0011] Performing data fusion on the feature data table and the driving data table to generate a comprehensive data table, and identifying the foundation type of the detection section in the comprehensive data table based on the threshold category;

[0012] A detection and identification model is constructed based on the comprehensive data table, and the extended feature information of the vehicle to be monitored is input into the detection and identification model. The detection and identification model outputs the vehicle information of the vehicle to be monitored based on the foundation type and the type label, and the detection and identification of the vehicle to be monitored is completed based on the vehicle information.

[0013] In combination with the first aspect, preprocessing the sensing signal to generate first data includes:

[0014] Acquire the position information of the distributed optical fiber sensor, convert the sensing signal into a characteristic parameter based on a characteristic type, and acquire the frequency range of the detection section from the characteristic parameter based on the position information;

[0015] Based on the feature type, the feature parameters are subjected to standardized vibration amplitude enhancement processing to generate enhanced data, the duration of the maximum amplitude in any of the enhanced data is obtained, the enhanced data whose duration is greater than a first preset value is eliminated, and the remaining enhanced data is set as the first data.

[0016] In combination with the first aspect, extracting extended feature information based on the spatial resolution of the second data includes:

[0017] The first data is filtered based on the frequency range to generate the second data, features are extracted from the second data based on the spatial resolution to generate vibration information and propagation features, and the vibration information and the propagation features are set as the extended feature information.

[0018] In combination with the first aspect, the driving data is screened to generate third data, including:

[0019] The driving data includes driving speed, driving lane and vibration data that change with time information, and the driving data of vehicles with the same type of tags are set as combined driving data;

[0020] Extract the first sub-data when the vehicle passes the distributed optical fiber sensor from the combined driving data, extract the second sub-data with the same driving speed and the same driving lane from the first sub-data, calculate the deviation of any one of the vibration data in the second sub-data, delete the second sub-data with the deviation greater than the second preset value, and set the remaining second sub-data in all the combined driving data as the third data.

[0021] Combined with the first aspect, perform data fusion on the feature data table and the driving data table, including:

[0022] Establish a traffic flow simulation model based on the driving data table, and check whether there is a missing detection in the feature data table based on the traffic flow simulation model. If not, perform label fusion on the feature data table and the driving data table based on the table labels. If there is, supplement the missing features in the feature data table and perform label fusion to complete the data fusion.

[0023] Combined with the first aspect, check whether there is a missing detection in the feature data table based on the traffic flow simulation model, including:

[0024] Respectively obtain the lane information corresponding to any lane based on the driving data table. The lane information includes vehicle quantity distribution, vehicle interval, and vehicle type. Perform overlapping occlusion modeling on all the lane information based on the driving direction and driving time to generate the traffic flow simulation model;

[0025] The traffic flow simulation model counts the overlapping times and passing times of vehicles at any one of the distributed optical fiber sensors, and counts the detection times of any one of the distributed optical fiber sensors based on the feature data table;

[0026] Calculate the detection probability of any one of the distributed optical fiber sensors based on the first formula. The first formula is: , where P is the detection probability, m is the detection times, M1 is the overlapping times, and M2 is the passing times;

[0027] If the detection probability is less than the third preset value, it is determined that there is a missing detection in the feature data table. Otherwise, it is determined that there is no missing detection in the feature data table.

[0028] Combined with the first aspect, generate the comprehensive data table, including:

[0029] Extract suspected missing vehicles from the driving data table based on the time information, obtain the actual vehicle information of the suspected missing vehicles, obtain the extended feature information corresponding to the same actual vehicle information in the feature data table, and set it as the missing feature, and write the missing feature into the feature data table;

[0030] Based on the time information, the data information corresponding to any table label in the feature data table and the driving data table is aggregated to generate a combined data table, the detection error between the extended feature and the actual vehicle information is calculated based on the combined data table, and the detection error is written into the combined data table to generate the comprehensive data table.

[0031] In combination with the first aspect, building a detection and recognition model based on the comprehensive data table includes:

[0032] The third sub-data whose detection error is less than or equal to the fourth preset value are extracted from the comprehensive data table, all the third sub-data are normalized to generate a standard sample set, and the standard sample set is subjected to deep learning denoising based on ResUnet to generate a model sample set.

[0033] In combination with the first aspect, building a detection and recognition model based on the comprehensive data table further includes:

[0034] A training set is extracted from the model sample set to train the YOLO model, the YOLO model outputs a recognition result of the verification set, a first accuracy rate of the recognition result is verified based on actual vehicle information in the comprehensive data table, parameters of the YOLO model are adjusted based on the first accuracy rate, and the adjusted YOLO model is set as the detection and recognition model.

[0035] In combination with the first aspect, the detection and recognition model outputs the vehicle information of the vehicle to be monitored based on the foundation type and the type label, including:

[0036] The detection and recognition model extracts spectrum features based on the extended feature information of the vehicle to be monitored, outputs the recognition result of the vehicle to be monitored based on the spectrum features, and summarizes all the recognition results based on the foundation type and the type label to generate the vehicle information.

[0037] In the technical solution provided by the present application, firstly, a plurality of distributed optical fiber sensors are laid on the detection section to form a sensor array, which can collect the vibration signals generated during the vehicle driving process in real time, has high sensitivity and spatial resolution, and can continuously monitor the entire lane. Then, based on the characteristic information such as vibration frequency, a multi-level binarization technology is used to divide the extended characteristic information into multiple threshold categories, which improves the recognition ability of different vibration frequency characteristics and can more accurately identify different types of vehicles. In addition, by screening and summarizing the driving data, a driving data table is generated, and the characteristic data table is fused with the driving data table, the effective fusion of multi-source data is realized, which makes up for the limitations of a single sensor data and improves the integrity and accuracy of the comprehensive data table. Finally, the ResUnet model is used to perform deep learning denoising on the standard sample set to generate a model sample set, and the YOLO model is used to train the model sample set, and the model parameters are adjusted based on the actual vehicle information of the comprehensive data table to generate a detection and recognition model, which can automatically learn the characteristic patterns of each vehicle and improve the recognition accuracy and robustness. The detection and recognition model extracts spectral features based on the extended feature information of the vehicle to be monitored, and conducts a comprehensive analysis based on the foundation type and type label, ultimately generating complete vehicle information including vehicle type, speed, driving lane, etc., further improving the accuracy and reliability of the recognition results. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0039] Figure 1 A schematic diagram of an embodiment of a highway vehicle detection and identification method based on distributed optical fiber sensing in an embodiment of the present application;

[0040] Figure 2 This is a schematic diagram of the laying of distributed optical fiber sensing data in an embodiment of the present application;

[0041] Figure 3 This is a schematic diagram of the process of missing detection test in the embodiment of the present application;

[0042] Figure 4 A schematic diagram of the framework for constructing a detection and recognition model in an embodiment of the present application. DETAILED DESCRIPTION

[0043] The embodiment of the present application provides a method for detecting and identifying highway vehicles based on distributed optical fiber sensing. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0044] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiment of the present application, an embodiment of a highway vehicle detection and identification method based on distributed optical fiber sensing includes:

[0045] Step S101: multiple distributed optical fiber sensors are laid on the detection section, and type labels of all vehicles are marked. When the vehicle is traveling on the detection section, the distributed optical fiber sensors collect sensing signals, and the sensing signals are pre-processed to generate first data.

[0046] It is understandable that the execution subject of the present application can be a highway vehicle detection and identification device based on distributed optical fiber sensing, or a terminal or a server, which is not limited here. The present application embodiment is described by taking the server as the execution subject as an example.

[0047] Specifically, the detection section refers to a highway that includes bridges, tunnels, and normal roadbeds. Distributed Acoustic Sensing (DAS) is a technology that uses optical fiber sensors to detect acoustic and vibration signals. By sending laser pulses to the optical fiber and measuring the phase change of the reflected light, DAS can detect tiny deformations along the optical fiber and infer acoustic events in the environment, such as Figure 2The diagram is a schematic diagram of the laying of distributed optical fiber sensing. The type label refers to the model type label of the vehicle, for example, small cars (such as sedans, SUVs), medium-sized cars (such as vans, pickup trucks), large cars (such as trucks, buses), special vehicles (such as motorcycles, trailers) and other type labels, so there are multiple corresponding vehicles. The sensing signal refers to the optical signal received when any vehicle passes through the distributed optical fiber sensor. Preprocessing refers to the use of filtering technology (such as low-pass filtering, high-pass filtering or band-pass filtering) to remove high-frequency noise and low-frequency interference in the sensing signal. Wavelet transform and other methods can be used for more refined noise removal, and the filtered signal is amplified to improve the signal-to-noise ratio of the signal. The preprocessed analog signal is converted into a digital signal to generate the first data.

[0048] Step S102: Optimize and filter the first data to generate second data, extract extended feature information based on the spatial resolution of the second data, divide all extended feature information into multiple threshold categories based on a multi-level binarization technique, mark the threshold categories of the extended feature information, and generate a feature data table based on the type label summary.

[0049] Specifically, filtering can adopt an adaptive filtering algorithm, dynamically adjust the filtering parameters according to the signal characteristics, filter according to the frequency, and generate the second data. Spatial resolution refers to the different signal channels contained in the second data. Extended feature information includes but is not limited to vibration amplitude, vibration frequency, vibration duration, and vibration waveform characteristics. Vibration amplitude refers to the vibration intensity caused by vehicle driving, reflecting information such as vehicle weight and speed. Vibration frequency refers to the vibration frequency caused by vehicle driving, reflecting information such as vehicle type and driving status. Vibration duration refers to the length of time the vehicle passes through the sensor area, reflecting information such as vehicle length and driving speed. Vibration waveform characteristics refer to the waveform characteristics of the vibration signal, such as the number of peaks, waveform shape, etc., which can be used to distinguish different types of vehicles. Multi-level binarization technology refers to dividing feature information such as vibration frequency into multiple levels according to a preset threshold range to generate multiple threshold categories, for example, low frequency range (0-10Hz), medium frequency range (10-30Hz), and high frequency range (above 30Hz). The threshold category can distinguish the vibration frequency corresponding to the same vehicle when driving on different foundation types. The feature data table refers to a list generated by summarizing and combining all extended feature information according to the type label. The information contained in the feature data table includes but is not limited to sensor ID, vehicle type label, vibration amplitude, vibration frequency (threshold category), vibration duration, vibration waveform characteristic parameters, and other related feature information.

[0050] Step S103: Acquire the driving data of the vehicle, generate third data after filtering the driving data, and generate a driving data table by aggregating all the third data based on the type label.

[0051] Specifically, driving data refers to various data information of a vehicle driving on a detection section, including but not limited to data information such as speed, lane, direction, and vibration intensity, and also includes identifying the vehicle's license plate number and obtaining vehicle registration information (such as vehicle type, color, etc.). Screening refers to eliminating incomplete, incomplete, and erroneous data information in the driving data. For example, if the speed data of vehicle A1 in the detection section is incomplete, the driving data of vehicle A1 will be eliminated. For another example, when the driving lanes and driving speeds corresponding to the driving data of vehicles A1 and A2 with the same type of labels are the same, but the vibration data corresponding to the vibration intensity are quite different, there may be erroneous data information between the two vehicles A1 and A2, and screening is required. The eliminated vehicles can be marked as abnormal vehicles, for example, abnormal parking, breakdown, overtaking, and other events occur in abnormal vehicles. The filtered driving data is set as the third data. All third data are aggregated based on type labels (sensor location, vehicle type label, etc.) to generate a driving data table, where the driving data table contains vehicle ID, vehicle type label, vehicle speed, driving direction, lane information, vibration data, and other relevant driving information.

[0052] Step S104: perform data fusion on the feature data table and the driving data table to generate a comprehensive data table, and identify the foundation type of the detection section in the comprehensive data table based on the threshold category.

[0053] Specifically, the feature data table contains the detection data information of the vehicle by the distributed optical fiber sensor, and the driving data table contains the actual data information of the vehicle. By fusing the feature data table with the driving data table, the detection data information of the distributed optical fiber sensor can be verified, and each detection data information can be labeled. For example, the fusion method can be based on the sensor ID and the vehicle type label for association matching, or the vibration feature information can be integrated with the driving information to form a complete vehicle data description. The fused list is set as a comprehensive data table. Different types of foundation materials will produce different vibration frequency responses. By analyzing the vibration frequency distribution characteristics and combining the known material vibration characteristics, the foundation type can be identified, where the foundation type includes but is not limited to bridges, tunnels and normal roadbeds.

[0054] Step S105: construct a detection and recognition model based on the comprehensive data table, input the extended feature information of the vehicle to be monitored into the detection and recognition model, the detection and recognition model outputs the vehicle information of the vehicle to be monitored based on the foundation type and type label, and completes the detection and recognition of the vehicle to be monitored based on the vehicle information.

[0055] Specifically, the comprehensive data table contains complete and relatively accurate sample data. The model type corresponding to the detection and recognition model can be constructed using a machine learning algorithm (such as a support vector machine, a neural network, a decision tree, etc.), wherein the model is trained using a labeled comprehensive data table. During the training process, the model learns the characteristic patterns of different types of vehicles and the influence of different foundation types on vibration signals. The model is verified using a test data set to evaluate its detection and recognition accuracy. The specific construction process will be described in detail later. After the extended feature information of the vehicle to be monitored is input into the detection and recognition model, the model outputs the following vehicle information based on the input vibration feature information, combined with the foundation type and type label, including but not limited to vehicle type, vehicle speed, driving direction, lane information, and other relevant vehicle information (such as vehicle length, weight, etc.). The vehicle detection and recognition results can complete the detection and recognition of the vehicle to be monitored based on the output vehicle information, which can be used for applications such as traffic flow statistics, vehicle classification statistics, and vehicle speed monitoring. It can be integrated with the traffic management system to realize real-time monitoring and management of highway traffic conditions. For example, it is used to identify events such as slow-moving, speeding, abnormal parking, breaking down, and overtaking.

[0056] In a specific embodiment, generating first data after preprocessing the sensing signal includes:

[0057] (1) Obtain the location information of the distributed optical fiber sensor, convert the sensing signal into a characteristic parameter based on the characteristic type, and obtain the frequency range of the detection section from the characteristic parameter based on the location information.

[0058] (2) Based on the feature type, the feature parameters are subjected to standardized vibration amplitude enhancement processing to generate enhanced data, the duration of the maximum amplitude in any enhanced data is obtained, the enhanced data with a duration greater than a first preset value is eliminated, and the remaining enhanced data is set as the first data.

[0059] Specifically, the location information of the laid distributed optical fiber sensor is obtained, including but not limited to the geographic location coordinates, installation lane information, relative location information, etc., wherein the geographic location coordinates refer to the longitude and latitude coordinates or the distance information relative to the reference object of the detection section, the installation lane information refers to the lane number where the sensor is located (such as lane 1, lane 2, etc.), and the relative location information refers to the distance of the sensor relative to the center line of the lane or the shoulder. The feature type refers to the feature category contained in the sensing signal, including but not limited to vibration frequency, width and duration. The feature parameter refers to the numerical information corresponding to each feature type. For example, the vibration frequency is obtained by converting the time domain signal into the frequency domain signal through methods such as fast Fourier transform (FFT), and the vibration frequency component of the signal is obtained, and the main vibration frequency component and its corresponding amplitude are extracted. The vibration width refers to the duration width of the vibration signal, which reflects information such as the vehicle's driving speed, and can be achieved by calculating the width or duration of the signal envelope. The vibration duration refers to the length of time that the vehicle passes through the sensor area, which can be calculated by detecting the time period when the signal amplitude exceeds a certain threshold. According to the location information of the sensor, the corresponding detection section is determined, and the frequency range of the section is obtained from the characteristic parameters, where the frequency range refers to the vibration frequency range in the characteristic parameters received by the distributed optical fiber sensor when vehicles with different types of tags travel on detection sections of different types of foundations.

[0060] The characteristic parameters are normalized to eliminate the amplitude differences that may exist between different sensors. The normalization method can adopt methods such as zero mean normalization (Z-score normalization) or minimum-maximum normalization (Min-Maxnormalization), which can make the data of different sensors comparable and facilitate the subsequent vibration amplitude enhancement processing and data analysis. The vibration amplitude enhancement processing is performed on the standardized characteristic parameters. The vibration amplitude enhancement processing method can adopt an adaptive gain control (AGC) algorithm to dynamically adjust the gain parameters according to the signal amplitude, or adopt methods such as wavelet transform to perform multi-scale decomposition and reconstruction of the signal to enhance the effective components in the signal, which can improve the signal-to-noise ratio of the signal and highlight the characteristics of the vehicle vibration signal. The duration of the maximum amplitude in any enhanced data is obtained by calculating the duration corresponding to the maximum amplitude point in the enhanced signal, that is, the time period when the signal amplitude exceeds a certain threshold. A first preset value (for example, 10 seconds) is set as the screening threshold. Signals with a duration that is too long may be the vibration of a large vehicle (for example, a truck or a construction vehicle) covering the vibration generated by a small vehicle (for example, a bus). Therefore, the enhanced data with a duration greater than the first preset value is eliminated. At the same time, another first sub-preset value can be set for screening enhanced data. Signals with a duration that is too short may be caused by noise or transient interference and do not have vehicle vibration characteristics. Enhanced data with a duration that is less than or equal to the first sub-preset value can be eliminated, wherein the first sub-preset value is much smaller than the first preset value. The screened enhanced data is set as the first data (pre-processed data) for subsequent data processing and analysis.

[0061] In a specific embodiment, extracting the extended feature information based on the spatial resolution of the second data includes:

[0062] The first data is filtered based on the frequency range to generate the second data, features are extracted from the second data based on the spatial resolution to generate vibration information and propagation features, and the vibration information and propagation features are set as extended feature information.

[0063] Specifically, the purpose of filtering is to remove noise and interference signals outside the frequency range and retain signal components related to vehicle vibration. The filter type can be a bandpass filter, which allows signals within a specific frequency range to pass and suppresses other frequency components. Vibration information includes but is not limited to vibration amplitude, vibration frequency, vibration duration, etc., where vibration amplitude is to extract parameters such as the maximum amplitude and average amplitude of the signal, vibration frequency is to extract the main vibration frequency components and their amplitudes, and vibration duration is to extract parameters such as signal duration. Propagation characteristics include but are not limited to propagation speed, propagation direction, spatial distribution characteristics, etc. Among them, the propagation speed is to calculate the propagation speed of the vibration signal based on the time difference between the signals received by different sensors in the sensor array. The propagation direction is to determine the propagation direction of the vibration signal by analyzing the phase difference of the signals received by different sensors. The spatial distribution characteristics are to analyze the spatial distribution patterns of vibration signals on different sensors, such as the attenuation law and spatial correlation of the signal.

[0064] In a specific embodiment, the third data is generated after the driving data is screened, including:

[0065] (1) The driving data includes the driving speed, driving lane and vibration data that change with time. The driving data of vehicles with the same type of labels are set as combined driving data.

[0066] (2) extracting the first sub-data when the vehicle passes the distributed optical fiber sensor from the combined driving data, extracting the second sub-data with the same driving speed and the same driving lane from the first sub-data, calculating the deviation of any vibration data in the second sub-data, deleting the second sub-data with a deviation greater than a second preset value, and setting the remaining second sub-data in all the combined driving data as the third data.

[0067] Specifically, the driving speed refers to the speed value of the vehicle at different time points during the driving process. The driving lane refers to the lane number of the vehicle at different time points. The vibration data refers to the tire vibration signal data of the vehicle at different time points. The driving speed, driving lane and vibration data of vehicles with the same type of labels are combined in chronological order to form combined driving data.

[0068] The first sub-data refers to a subset of the combined driving data, wherein the first sub-data is the driving data of any vehicle passing through each distributed optical fiber sensor, and the driving data fragment of the vehicle passing through the distributed optical fiber sensor can be identified. The same driving speed and driving lane are screened out from the first sub-data to generate the second sub-data, and the difference in vibration data between different vehicles with the same type of label can be analyzed, wherein the second sub-data is a subset of the first sub-data. The deviation can be defined as the absolute difference between the vibration data and the mean of all vibration data. The second preset value is set as the deviation threshold to determine the degree of abnormality of the vibration data. If the deviation is greater than the second preset value, it means that the vehicle corresponding to the second sub-data is an abnormal vehicle, which can be deleted or marked as an event for the abnormal vehicle, for example, marking the abnormal vehicle as a vehicle overload event. Similarly, according to setting the same variable of different driving data as the second sub-data, various event types can be marked in turn, for example, the second sub-data with the same vibration data and the same driving lane is extracted from the first sub-data, and the abnormal vehicle of the speeding event can be marked by the deviation of the driving speed. The remaining first sub-data in all combined driving data are set as the third data (screened driving data).

[0069] In a specific embodiment, data fusion is performed on the feature data table and the driving data table, including:

[0070] A traffic flow simulation model is established based on the driving data table. Based on the traffic flow simulation model, it is checked whether there is a missing detection in the feature data table. If not, the feature data table and the driving data table are fused with labels based on the table labels. If so, the missing features of the feature data table are supplemented and label fusion is performed to complete data fusion.

[0071] Specifically, the traffic flow simulation model refers to a model used to describe the driving records of each vehicle on the detection section. Missing detection refers to the omission of vehicle detection by the distributed optical fiber sensor. Therefore, the traffic flow simulation model is constructed using the driving data table to simulate the movement trajectory and distribution of vehicles on the highway. The traffic flow simulation model is used to evaluate the integrity of the feature data table and identify whether there are areas or time periods where vehicles are not detected, for example, Figure 3 Schematic diagram of the missing detection test process.

[0072] If there is no missing detection in the feature data table, label fusion is directly performed. Label fusion refers to the association and matching of the feature data table and the driving data table based on table labels (such as sensor ID, vehicle type label, etc.). The relevant data items in the two tables are merged to generate a comprehensive data table.

[0073] In a specific embodiment, checking whether there is a missing feature in the feature data table based on the traffic simulation model includes:

[0074] (1) Based on the driving data table, the lane information corresponding to each lane is obtained. The lane information includes the number of vehicles, vehicle intervals, and vehicle types. Based on the driving direction and driving time, the overlap occlusion model of all lane information is modeled to generate a traffic simulation model.

[0075] (2) The traffic flow simulation model counts the number of overlaps and passing times of vehicles at any distributed optical fiber sensor, and counts the number of detections of any distributed optical fiber sensor based on the feature data table.

[0076] (3) The detection probability of any distributed optical fiber sensor is calculated based on the first formula. The first formula is: , where P is the detection probability, m is the number of detections, M1 is the number of overlaps, and M2 is the number of passes.

[0077] (4) If the detection probability is less than the third preset value, it is determined that there is a missing detection in the feature data table; otherwise, it is determined that there is no missing detection in the feature data table.

[0078] Specifically, overlap occlusion modeling includes overlap modeling and occlusion modeling, where overlap modeling refers to analyzing the movement trajectories of vehicles on different lanes and identifying the overlapping areas or time periods that may occur during the driving process of vehicles. For example, on a multi-lane highway, vehicles may switch between different lanes, causing their trajectories to cross. Occlusion modeling refers to considering the occlusion effect between vehicles. For example, a large vehicle may block the vehicle behind it, causing the sensor to be unable to effectively detect the obscured vehicle. By analyzing factors such as vehicle size and driving speed, an occlusion model is established to evaluate the impact of occlusion on detection. Based on the above information, a traffic simulation model is constructed to simulate the movement and distribution of vehicles on the highway. For example, the traffic simulation model can be expressed as: traffic simulation model = f (lane information, driving direction, driving time, overlap modeling, occlusion modeling), where f is a mathematical function.

[0079] Using the traffic simulation model, count the number of overlaps (M1) and the number of passes (M2) of vehicles at any distributed fiber optic sensor. The number of overlaps (M1) refers to the number of times a vehicle overlaps with other vehicles at the sensor. The number of passes (M2) refers to the total number of times a vehicle passes the sensor. Count the number of detections (m) of any distributed fiber optic sensor from the feature data table, that is, the number of times the sensor detects a vehicle passing.

[0080] In the first formula, the detection probability of any distributed optical fiber sensor can be calculated. If the calculated detection probability P is less than the third preset value, it is determined that there is a missing detection in the feature data table. For example, if P<0.9, it means that the sensor's detection rate for the vehicle is lower than expected, and there may be missed detections. Otherwise, it is determined that there is no missing detection in the feature data table.

[0081] In a specific embodiment, generating a comprehensive data table includes:

[0082] (1) Based on the time information, the suspected missing vehicles are extracted from the driving data table, the actual vehicle information of the suspected missing vehicles is obtained, the extended feature information corresponding to the same actual vehicle information is obtained from the feature data table, and is set as the missing feature, and the missing feature is written into the feature data table.

[0083] (2) Based on the time information, the data information corresponding to any table label in the feature data table and the driving data table is aggregated to generate a combined data table, the detection error between the extended feature and the actual vehicle information is calculated based on the combined data table, and the detection error is written into the combined data table to generate a comprehensive data table.

[0084] Specifically, suspected missing vehicles refer to vehicles that are identified from the driving data table and have no corresponding detection records in the feature data table. The actual vehicle information refers to the data information of the suspected missing vehicle in the driving data table, including vehicle type (small car, medium-sized car, large car, etc.), license plate number, driving speed (the speed of the vehicle during the suspected missing time period), driving lane (the lane where the vehicle is located during the suspected missing time period), and other relevant information (driving data). Search for records matching the actual vehicle information in the feature data table. The matching basis includes vehicle type, license plate number (if any), driving speed range, driving lane, etc. The purpose of matching is to find the vehicle detection record that is most similar to the suspected missing vehicle to obtain its extended feature information. The matched extended feature information is set as the missing feature of the suspected missing vehicle, including vibration amplitude, vibration frequency, vibration duration, vibration waveform characteristics, and other related features. Write the missing feature information into the position corresponding to the suspected missing vehicle in the feature data table to complete the supplement of the feature data table.

[0085] Based on the time information, the data information of the corresponding table tags in the feature data table and the driving data table are summarized to generate a combined data table, where the time information refers to the timestamp. The combined data table includes table tags and summary information, where the table tags include sensor ID, vehicle type tags, etc., which are used to associate the data in the two tables. The summary content refers to integrating the extended feature information in the feature data table with the actual vehicle information in the driving data table to form a complete vehicle data description.

[0086] The detection error is used to evaluate the matching degree between the extended feature information and the actual vehicle information, including but not limited to speed error, lane error, and other related errors, where the speed error refers to the difference between the speed estimate in the extended feature information and the actual vehicle speed, and the lane error refers to the difference between the lane identified in the extended feature information and the actual lane. Each detection error indicator is calculated and recorded in the combined data table. After the detection error information is written into the combined data table, the final comprehensive data table is generated, which contains the following information: time information (the time point when the vehicle passes the sensor), actual vehicle information (vehicle type, license plate number, driving speed, driving lane, etc.), extended feature information (vibration amplitude, vibration frequency, vibration duration, etc.), detection error information (speed error, lane error, etc.).

[0087] In a specific embodiment, a detection and recognition model is constructed based on a comprehensive data table, including:

[0088] The third sub-data whose detection error is less than or equal to the fourth preset value are extracted from the comprehensive data table, all the third sub-data are normalized to generate a standard sample set, and the standard sample set is subjected to deep learning denoising based on ResUnet to generate a model sample set.

[0089] Specifically, high-quality data samples are screened out from the comprehensive data table based on the detection error for subsequent model training. A fourth preset value (for example, 5%) is set as the threshold of the detection error, which is used to screen out samples with smaller detection errors and higher data quality. Samples with detection errors less than or equal to the fourth preset value are extracted from the comprehensive data table to generate the third sub-data. For example, if the speed error of a sample is 3% and the lane error is 2%, then its total detection error is 5%, which meets the screening criteria. Data with different features are standardized through normalization processing so that they have the same scale, avoiding excessive impact of certain features on model training due to different dimensions. The normalized third sub-data is set as the standard sample set for subsequent deep learning denoising.

[0090] ResUnet is a deep learning model that combines the residual network (ResNet) and U-Net structure. It is suitable for tasks such as image segmentation and denoising. In this application, ResUnet is used to denoise the standard sample set, remove noise and outliers in the data, and generate a denoised model sample set for subsequent model training.

[0091] In a specific embodiment, building a detection and recognition model based on the comprehensive data table also includes:

[0092] A training set is extracted from the model sample set to train the YOLO model. The YOLO model outputs the recognition result of the verification set. The first accuracy of the recognition result is verified based on the actual vehicle information in the comprehensive data table. The parameters of the YOLO model are adjusted based on the first accuracy, and the adjusted YOLO model is set as the detection and recognition model.

[0093] Specifically, a certain proportion of data is randomly extracted from the model sample set as a training set (for example, 70%), and the remaining data is used as a validation set for model evaluation. The YOLO (You Only Look Once) model is a real-time target detection model based on deep learning. In this application, the YOLO model is used to detect and identify vehicles. Input the training set data, including vibration feature information, etc., learn the characteristic pattern of the vehicle, and achieve accurate detection and identification of the vehicle. During the training process, the back propagation algorithm and optimization algorithm (such as Adam, SGD) are used for model training. The pre-trained model can be used for transfer learning to improve training efficiency and model performance. Input the validation set into the trained YOLO model to obtain the recognition result, compare the recognition result with the actual vehicle information in the comprehensive data table, and calculate the recognition accuracy (first accuracy). The recognition accuracy refers to the ratio of the number of correctly identified vehicles to the total number of vehicles, where the comparison basis includes but is not limited to information such as license plate number, vehicle type, and driving lane.

[0094] Adjust the parameters of the YOLO model according to the recognition accuracy to improve the model performance. The adjustment methods include learning rate adjustment, network structure adjustment, hyperparameter optimization, etc. Among them, learning rate adjustment means that the accuracy is low, and the learning rate can be appropriately reduced to prevent the model from falling into the local optimum. Network structure adjustment refers to increasing or decreasing the number of network layers, adjusting the number of convolution kernels, etc. Hyperparameter optimization refers to adjusting hyperparameters such as batch size and number of iterations. The adjustment goal is to maximize the recognition accuracy, and set the adjusted YOLO model as the final detection and recognition model for detecting and identifying vehicles on the highway, for example, Figure 4 Schematic diagram of the framework for building a detection and recognition model.

[0095] In a specific embodiment, the detection and recognition model outputs vehicle information of the vehicle to be monitored based on the foundation type and the type label, including:

[0096] The detection and recognition model extracts spectral features based on the extended feature information of the vehicle to be monitored, outputs the recognition result of the vehicle to be monitored based on the spectral features, summarizes all recognition results based on the foundation type and type label, and generates vehicle information.

[0097] Specifically, the extended feature information of the vehicle to be monitored is input into the constructed detection and recognition model, and the input vibration frequency information is subjected to spectrum analysis. For example, the time domain signal is converted into a frequency domain signal using the fast Fourier transform (FFT) to obtain a spectrum diagram that can reflect the vibration frequency components of the vehicle. Key spectrum features are extracted from the spectrum diagram, such as the main frequency component (the frequency component with the largest energy in the vibration signal), frequency distribution characteristics (the distribution range and distribution law of the frequency component), and harmonic components (harmonic frequency components existing in the vibration signal). These spectrum features can reflect the type, speed, load and other characteristics of the vehicle.

[0098] Input the spectral features into the trained detection and recognition model (such as the YOLO model), classify the input spectral features according to the learned feature patterns, and the model outputs preliminary recognition results, such as vehicle type (small car, medium car, large car, etc.), driving direction, lane information, etc. Since there are multiple distributed fiber optic sensors, the output recognition results of the monitored vehicle may also be multiple. The final recognition result is aggregated with the following information to generate complete vehicle information, including but not limited to time information (the time point when the vehicle passes the sensor), driving speed (which can be calculated based on the vibration duration and sensor spacing), driving lane (provided by the type label), driving direction (may be determined based on the fiber optic channel), vehicle type (provided by the recognition result), etc.

[0099] Through the cooperation of the above components, firstly, multiple distributed fiber optic sensors are laid on the detection section to form a sensor array, which can collect the vibration signals generated by the vehicle in real time. It has high sensitivity and spatial resolution and can continuously monitor the entire lane. Then, based on the characteristic information such as vibration frequency, the multi-level binarization technology is used to divide the extended characteristic information into multiple threshold categories, which improves the recognition ability of different vibration frequency characteristics and can more accurately identify different types of vehicles. In addition, by screening and summarizing the driving data, a driving data table is generated, and the characteristic data table is fused with the driving data table, the effective fusion of multi-source data is realized, which makes up for the limitations of a single sensor data and improves the integrity and accuracy of the comprehensive data table. Finally, the ResUnet model is used to perform deep learning denoising on the standard sample set to generate a model sample set, and the YOLO model is used to train the model sample set. The model parameters are adjusted based on the actual vehicle information of the comprehensive data table to generate a detection and recognition model, which can automatically learn the characteristic patterns of each vehicle and improve the recognition accuracy and robustness. The detection and recognition model extracts spectral features based on the extended feature information of the vehicle to be monitored, and conducts a comprehensive analysis based on the foundation type and type label, ultimately generating complete vehicle information including vehicle type, speed, driving lane, etc., further improving the accuracy and reliability of the recognition results.

[0100] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0101] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0102] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for detecting and identifying highway vehicles based on distributed optical fiber sensing, characterized in that: The highway vehicle detection and identification method based on distributed optical fiber sensing includes: A plurality of distributed optical fiber sensors are laid on the detection section, and type labels of all vehicles are marked. When the vehicle travels on the detection section, the distributed optical fiber sensors collect sensing signals, and the sensing signals are pre-processed to generate first data; The first data is optimized and filtered to generate second data, extended feature information is extracted based on the spatial resolution of the second data, all the extended feature information is divided into multiple threshold categories based on a multi-level binarization technology, and the extended feature information is marked with the threshold category and then summarized based on the type label to generate a feature data table; Acquire driving data of the vehicle, generate third data after filtering the driving data, and generate a driving data table by aggregating all the third data based on the type label; Performing data fusion on the feature data table and the driving data table to generate a comprehensive data table, and identifying the foundation type of the detection section in the comprehensive data table based on the threshold category; A detection and identification model is constructed based on the comprehensive data table, and the extended feature information of the vehicle to be monitored is input into the detection and identification model. The detection and identification model outputs the vehicle information of the vehicle to be monitored based on the foundation type and the type label, and the detection and identification of the vehicle to be monitored is completed based on the vehicle information.

2. The highway vehicle detection and identification method based on distributed optical fiber sensing according to claim 1 is characterized in that: Preprocessing the sensing signal to generate first data includes: Acquire the position information of the distributed optical fiber sensor, convert the sensing signal into a characteristic parameter based on a characteristic type, and acquire the frequency range of the detection section from the characteristic parameter based on the position information; Based on the feature type, the feature parameters are subjected to standardized vibration amplitude enhancement processing to generate enhanced data, the duration of the maximum amplitude in any of the enhanced data is obtained, the enhanced data whose duration is greater than a first preset value is eliminated, and the remaining enhanced data is set as the first data.

3. The highway vehicle detection and identification method based on distributed optical fiber sensing according to claim 2 is characterized in that: Extracting extended feature information based on the spatial resolution of the second data includes: The first data is filtered based on the frequency range to generate the second data, features are extracted from the second data based on the spatial resolution to generate vibration information and propagation features, and the vibration information and the propagation features are set as the extended feature information.

4. The highway vehicle detection and identification method based on distributed optical fiber sensing according to claim 1 is characterized in that: The driving data is screened to generate third data, including: The driving data includes driving speed, driving lane and vibration data that change with time information, and the driving data of vehicles with the same type of tags are set as combined driving data; The first sub-data when the vehicle passes through the distributed optical fiber sensor is extracted from the combined driving data, and the second sub-data with the same driving speed and the same driving lane is extracted from the first sub-data, the deviation of any vibration data in the second sub-data is calculated, and the second sub-data whose deviation is greater than a second preset value is deleted, and the remaining second sub-data in all the combined driving data are set as the third data.

5. The highway vehicle detection and identification method based on distributed optical fiber sensing according to claim 1 is characterized in that: Performing data fusion on the feature data table and the driving data table includes: A traffic flow simulation model is established based on the driving data table, and based on the traffic flow simulation model, it is checked whether there is a missing detection in the feature data table. If not, the feature data table and the driving data table are fused with labels based on the table labels. If so, the missing features of the feature data table are supplemented and label fusion is performed to complete data fusion.

6. The highway vehicle detection and identification method based on distributed optical fiber sensing according to claim 5 is characterized in that: Checking whether there is a missing detection in the feature data table based on the traffic flow simulation model includes: Based on the driving data table, lane information corresponding to any lane is obtained respectively, wherein the lane information includes vehicle quantity distribution, vehicle interval and vehicle type, and overlap occlusion modeling is performed on all lane information based on driving direction and driving time to generate the traffic flow simulation model; The traffic flow simulation model counts the number of overlaps and the number of passings of vehicles at any of the distributed optical fiber sensors, and counts the number of detections of any of the distributed optical fiber sensors based on the characteristic data table; The detection probability of any of the distributed optical fiber sensors is calculated based on a first formula, wherein the first formula is: , where P is the detection probability, m is the number of detections, M1 is the number of overlaps, and M2 is the number of passes; If the detection probability is less than a third preset value, it is determined that there is a missing detection in the feature data table; otherwise, it is determined that there is no missing detection in the feature data table.

7. The highway vehicle detection and identification method based on distributed optical fiber sensing according to claim 5 is characterized in that: Generate the comprehensive data table, including: Extracting a suspected missing vehicle from the driving data table based on time information, obtaining actual vehicle information of the suspected missing vehicle, obtaining extended feature information corresponding to the same actual vehicle information from the feature data table, setting it as the missing feature, and writing the missing feature into the feature data table; Based on the time information, the data information corresponding to any table label in the feature data table and the driving data table is aggregated to generate a combined data table, the detection error between the extended feature and the actual vehicle information is calculated based on the combined data table, and the detection error is written into the combined data table to generate the comprehensive data table.

8. The highway vehicle detection and identification method based on distributed optical fiber sensing according to claim 1 is characterized in that: Building a detection and recognition model based on the comprehensive data table includes: The third sub-data whose detection error is less than or equal to the fourth preset value are extracted from the comprehensive data table, all the third sub-data are normalized to generate a standard sample set, and the standard sample set is subjected to deep learning denoising based on ResUnet to generate a model sample set.

9. The method for detecting and identifying highway vehicles based on distributed optical fiber sensing according to claim 8, characterized in that: Building a detection and recognition model based on the comprehensive data table also includes: A training set is extracted from the model sample set to train the YOLO model, the YOLO model outputs a recognition result of the verification set, a first accuracy rate of the recognition result is verified based on actual vehicle information in the comprehensive data table, parameters of the YOLO model are adjusted based on the first accuracy rate, and the adjusted YOLO model is set as the detection and recognition model.

10. The highway vehicle detection and identification method based on distributed optical fiber sensing according to claim 1, characterized in that: The detection and recognition model outputs the vehicle information of the vehicle to be monitored based on the foundation type and the type label, including: The detection and recognition model extracts spectrum features based on the extended feature information of the vehicle to be monitored, outputs the recognition result of the vehicle to be monitored based on the spectrum features, and summarizes all the recognition results based on the foundation type and the type label to generate the vehicle information.

Images