Extinction coefficient inversion method, system and device and storage medium
By synchronizing observation data with micro-pulse lidar and Raman lidar and inversion using RBF neural network model, the accuracy deviation problem caused by the dependence assumption of the traditional extinction coefficient inversion method is solved, and higher accuracy and reliable extinction coefficient inversion are achieved.
Patent Information
- Application Number
- CN202411856615.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-17
- Publication Date
- 2025-03-18
AI Technical Summary
The traditional extinction coefficient inversion method relies on hypothetical atmospheric model and hypothetical reference boundary points, resulting in uncertainty and accuracy deviation of inversion results, especially when meteorological conditions are unstable or environments are complex, the inversion accuracy is poor.
By synchronous observation of micro-pulse lidar and high-energy Raman lidar, preprocess the data, and invert it using RBF neural network model, the assumption of extinction coefficient of atmospheric boundary points is reduced and the limitations of the vertical direction of traditional technologies are broken.
It improves the accuracy and reliability of extinction coefficient inversion, reduces errors, and can effectively eliminate signal interference in the presence of occlusion, and maintains high-precision detection of aerosol in front of occlusion.
Smart Images
Figure CN119644360A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of atmospheric detection, and in particular to an extinction coefficient inversion method, device, equipment and storage medium. Background Art
[0002] Micro-Pulse Lidar (MPL) is an important tool currently widely used in aerosol monitoring and atmospheric environment research, especially in measuring aerosol extinction coefficient. Micro-Pulse Lidar is a backscattering Lidar that measures the physical properties of aerosols in the atmosphere by emitting low-energy, high-repetition-rate laser pulses into the atmosphere and receiving atmospheric echoes.
[0003] Extinction coefficient inversion is one of the key technologies. The extinction coefficient inversion method refers to a technical means of using the data observed by the lidar system, processing it through specific mathematical models and algorithms, and thus calculating the extinction coefficient of light propagation in the atmosphere.
[0004] At present, the traditional extinction coefficient inversion method is based on the assumed atmospheric model and the assumed reference boundary point. It needs to assume a certain extinction-to-backscattering ratio and depends on parameters such as the atmospheric calibration height. These assumptions often lead to uncertainty and accuracy deviation in the inversion results, especially when the meteorological conditions are unstable or the environment is complex (such as when there are obstructions, clouds or strong aerosol concentrations), the inversion accuracy is poor. Summary of the invention
[0005] In order to provide a more accurate extinction coefficient inversion result, the present application provides an extinction coefficient inversion method, device, equipment and storage medium.
[0006] In a first aspect, the present application provides an extinction coefficient inversion method, which adopts the following technical solution:
[0007] Using the micro-pulse laser radar and the high-energy Raman laser radar to synchronously observe and collect training data, the training data includes micro-pulse laser radar data and Raman laser radar data;
[0008] Preprocessing the micro-pulse laser radar data to obtain a preprocessed echo signal;
[0009] Preprocessing the Raman laser radar data and using a Raman laser radar inversion algorithm to obtain extinction coefficient data corresponding to the preprocessed echo signal;
[0010] Using the pre-processed echo signal and the extinction coefficient data to train the RBF neural network to obtain an RBF neural network model, wherein the RBF neural network model can obtain the extinction coefficient data in real-time inversion according to the pre-processed echo signal;
[0011] When the actual echo signal collected by the micro-pulse laser radar is received, the actual echo signal is preprocessed to obtain the target echo signal, and the RBF neural network model is used to predict and invert the target echo signal to obtain the extinction coefficient.
[0012] Through the above technical solution, the inversion algorithm of Raman lidar does not need to make assumptions about the extinction coefficient of the atmospheric boundary points. The RBF neural network model is trained using micro-pulse lidar data and Raman lidar data, breaking through the vertical limitations of traditional technology. The inversion using the RBF neural network model improves the inversion accuracy and reduces errors.
[0013] In a specific implementation scheme, the synchronous observation and collection of training data using the micro-pulse laser radar and the high-energy Raman laser radar includes:
[0014] The micro-pulse laser radar and the high-energy Raman laser radar are controlled to synchronously observe and collect training data in the same direction.
[0015] Through the above technical solution, the micro-pulse laser radar and the high-energy Raman laser radar are controlled to collect data in the same direction at the same time, ensuring that the detection objects are the same and improving the reliability of the RBF neural network model.
[0016] In a specific implementation scheme, the preprocessing of the micro-pulse laser radar data to obtain a preprocessed echo signal includes:
[0017] Deducting background noise from the micro-pulse laser radar data to obtain a first pre-processed echo signal;
[0018] removing high frequency noise from the first preprocessed echo signal to obtain a second preprocessed echo signal;
[0019] Screening out the drastically changed echo signals whose change amplitude is greater than a normal range in the second preprocessed echo signals;
[0020] The drastically changing echo signal in the second preprocessed echo signal is removed to obtain a preprocessed echo signal.
[0021] Through the above technical solution, the preprocessing of the Raman laser radar data and obtaining the extinction coefficient data corresponding to the preprocessed echo signal using the Raman laser radar inversion algorithm includes:
[0022] Deducting the background noise of the Raman laser radar data to obtain a first Raman laser radar signal;
[0023] Removing high-frequency noise from the first Raman laser radar signal to obtain a Raman laser radar signal;
[0024] The Raman laser radar signal is processed by a Raman laser radar inversion algorithm to obtain extinction coefficient data corresponding to the preprocessed echo signal.
[0025] Through the above technical solution, the RBF neural network training results are made more accurate by preprocessing the micro-pulse lidar data and Raman lidar data.
[0026] In a specific implementation scheme, the using the pre-processed echo signal and the extinction coefficient data to train the RBF neural network to obtain the RBF neural network model comprises:
[0027] Using the pre-processed echo signal as input data of the RBF neural network and using the extinction coefficient data as a label of the RBF neural network;
[0028] The RBF neural network model is obtained by training the RBF neural network according to the input data and the label.
[0029] Through the above technical solution, the inversion algorithm of the Raman lidar does not need to make assumptions about the extinction coefficient of the atmospheric boundary points. The high-precision extinction coefficient data provided by the Raman radar is used as a training label to improve the accuracy and reliability of the inversion model.
[0030] In a specific implementation manner, after the RBF neural network is trained using the pre-processed echo signal and the extinction coefficient data to obtain the RBF neural network model, the method further includes:
[0031] Utilizing the micro-pulse laser radar and the high-energy Raman laser radar to synchronously observe, collect and process test data to obtain a test echo signal and test extinction coefficient data;
[0032] Importing the test echo signal into the RBF neural network model to obtain target extinction coefficient data;
[0033] Comparing the target extinction coefficient data with the test extinction coefficient data to obtain an error, and determining whether the error is within an allowable range;
[0034] If the error is not within the allowable range, it is determined that the RBF neural network model training has failed;
[0035] If the error is within the allowable range, it is determined that the RBF neural network model is successfully trained, and the use of the RBF neural network model to predict and invert the target echo signal to obtain the extinction coefficient includes:
[0036] The successfully trained RBF neural network model is used to predict and invert the target echo signal to obtain the extinction coefficient.
[0037] Through the above technical solution, the preprocessed micro-pulse lidar echo signal is input into the RBF neural network model to obtain the target extinction coefficient data, which is compared with the extinction coefficient data obtained by Raman lidar data processing to verify the RBF neural network model, so that the inversion accuracy of the RBF neural network model is guaranteed in actual use.
[0038] In a specific implementation scheme, the preprocessing of the actual echo signal to obtain the target echo signal includes:
[0039] Determine whether there is a drastic change echo signal with a change amplitude greater than a normal range in the actual echo signal;
[0040] If the actual echo signal contains the drastically changing echo signal, the actual echo signal is subjected to background subtraction and noise reduction processing, and then the drastically changing echo signal is processed by a linear interpolation method to obtain a target echo signal;
[0041] Otherwise, the target echo signal is obtained after background subtraction and noise reduction processing on the actual echo signal.
[0042] Through the above technical solution, these changing areas are smoothed by a linear interpolation method, thereby eliminating discontinuities in the signal and avoiding the oscillation of the extinction coefficient during the inversion process.
[0043] In a second aspect, the present application provides an extinction coefficient inversion system, which adopts the following technical solution: the system comprises:
[0044] A training data collection module, used for synchronously observing and collecting training data using the micro-pulse laser radar and the high-energy Raman laser radar, wherein the training data includes micro-pulse laser radar data and Raman laser radar data;
[0045] A preprocessing echo signal module, used for preprocessing the micro-pulse laser radar data to obtain a preprocessing echo signal;
[0046] An extinction coefficient data module, used for preprocessing the Raman laser radar data and obtaining extinction coefficient data corresponding to the preprocessed echo signal using a Raman laser radar inversion algorithm;
[0047] An RBF neural network model training module is used to train the RBF neural network using the preprocessed echo signal and the extinction coefficient data to obtain an RBF neural network model, wherein the RBF neural network model can obtain the extinction coefficient data by real-time inversion according to the preprocessed echo signal;
[0048] The extinction coefficient actual inversion module is used to pre-process the actual echo signal to obtain the target echo signal when the actual echo signal collected by the micro-pulse laser radar is received, and use the RBF neural network model to predict and invert the target echo signal to obtain the extinction coefficient.
[0049] In a third aspect, the present application provides a computer device, which adopts the following technical solution: it includes a memory and a processor, and the memory stores a computer program that can be loaded by the processor and execute an extinction coefficient inversion method such as the above-mentioned one.
[0050] In a fourth aspect, the present application provides a computer-readable storage medium, which adopts the following technical solution: storing a computer program that can be loaded by a processor and execute the above-mentioned extinction coefficient inversion method.
[0051] In summary, this application has the following beneficial technical effects:
[0052] (1) The RBF neural network model is trained using micro-pulse lidar data and Raman lidar data. Through the training of the RBF neural network, the RBF neural network can break through the vertical limitations of traditional technology and automatically learn the complex nonlinear relationship between the echo signal and the extinction coefficient, thereby achieving high-precision aerosol extinction coefficient inversion. Even in the presence of obstructions, it can effectively eliminate interference in the signal and maintain high-precision detection of aerosols in front of the obstructions.
[0053] (2) The present invention improves the accuracy and reliability of the inversion model by synchronously collecting data from micro-pulse laser radar and high-energy Raman laser radar and using the high-precision extinction coefficient data provided by Raman radar as training labels.
[0054] (3) When drastic changes occur in the micro-pulse lidar echo signal (usually caused by clouds or obstructions), the present invention uses a linear interpolation method to smooth these change areas, thereby eliminating discontinuities in the signal and avoiding oscillations in the extinction coefficient during the inversion process. This application ensures that the inversion results of the extinction coefficient remain stable and accurate in the presence of clouds or obstructions. BRIEF DESCRIPTION OF THE DRAWINGS
[0055] Figure 1 It is a flow chart of an extinction coefficient inversion method in an embodiment of the present application.
[0056] Figure 2 Schematic diagram of a synchronous control system in an embodiment of the present application.
[0057] Figure 3 A structural block diagram of an extinction coefficient inversion method in an embodiment of the present application.
[0058] Figure numerals: 301, training data collection module; 302, pre-processing echo signal module; 303, extinction coefficient data module; 304, RBF neural network model training module; 305, extinction coefficient actual inversion module. DETAILED DESCRIPTION
[0059] The following is combined with Figure 1-Figure 3 This application is described in further detail.
[0060] The embodiments of the present application disclose an extinction coefficient inversion method, apparatus, device and storage medium, which are used to provide more accurate extinction coefficient inversion results.
[0061] Extinction coefficient inversion is one of the key technologies. The extinction coefficient inversion method refers to a technical means of using the data observed by the lidar system, processing it through specific mathematical models and algorithms, and thus calculating the extinction coefficient of light propagation in the atmosphere.
[0062] Micro-Pulse Lidar (MPL) is an important tool currently widely used in aerosol monitoring and atmospheric environment research, especially in measuring aerosol extinction coefficient. Micro-Pulse Lidar is a backscattering Lidar that measures the physical properties of aerosols in the atmosphere by emitting low-energy, high-repetition-rate laser pulses into the atmosphere and receiving atmospheric echoes.
[0063] The inversion of extinction coefficient under the existing technology has the following major defects:
[0064] First, many traditional methods (such as the Fernald method and the Klett method) rely on assumed extinction to backscatter ratios (AEBRs), which may not hold true under different meteorological conditions. Due to the dynamic changes in the atmosphere, these assumed values usually do not accurately reflect the actual situation, resulting in large inversion errors.
[0065] Second, in practical applications, the inversion accuracy of micro-pulse lidar is affected by obstructions (such as buildings, trees, etc.) and clouds. Since traditional inversion methods cannot handle the missing echo signals caused by obstructions, the inversion results are inaccurate and fail in environments with obstructions or clouds.
[0066] Third, the signal-to-noise ratio of micro-pulse lidar is usually low, especially in long-distance measurement or low-concentration aerosol conditions, the echo signal is very weak. Traditional inversion methods are easily interfered by noise in such low signal-to-noise ratio environments, resulting in deviations in measurement results and affecting the accuracy of the extinction coefficient.
[0067] Fourth, traditional inversion methods often rely on a specific detection range or maximum detection altitude, and cannot effectively adapt to applications at different altitudes and complex atmospheric conditions. These methods lack adaptability to environmental changes, limiting the application of micro-pulse lidar in complex climate conditions.
[0068] In summary, the traditional extinction coefficient inversion method is based on an assumed atmospheric model and assumed reference boundary points. It requires the assumption of a certain extinction-to-backscattering ratio and depends on parameters such as the atmospheric calibration height. These assumptions often lead to uncertainty and accuracy deviation in the inversion results, especially when the meteorological conditions are unstable or the environment is complex (such as when there are obstructions, clouds or strong aerosol concentrations), the inversion accuracy is poor.
[0069] Therefore, the present application proposes an extinction coefficient inversion method, which is used to provide a more accurate extinction coefficient inversion result.
[0070] like Figure 1 As shown, the method is based on an extinction coefficient inversion system including an RBF neural network, a micro-pulse laser radar and a high-energy Raman laser radar, and the method specifically includes the following steps:
[0071] S10, using micro-pulse lidar and high-energy Raman lidar to synchronously observe and collect training data, the training data being micro-pulse lidar data and Raman lidar data.
[0072] Specifically, the first step of the present invention is data collection, and the data comes from micro-pulse laser radar and high-energy Raman laser radar. In order to train through RBF neural network, training data needs to be collected first. In order to ensure the accuracy of the inversion model, data collection needs to cover different meteorological conditions (such as sunny days, cloudy days, different concentrations of aerosols, etc.). Two different radars need to collect data in the same direction at the same time. The training data includes several groups of data, namely micro-pulse laser radar data and the corresponding Raman laser radar data collected at the same time.
[0073] S20, preprocessing the micro-pulse laser radar data to obtain a preprocessed echo signal.
[0074] Specifically, the collected data needs to be preprocessed to ensure the data quality of the input neural network. The processing of micro-pulse lidar data specifically includes background subtraction, filtering, and removing interference signals caused by clouds or obstructions.
[0075] S30, pre-processing the Raman laser radar data and using the Raman laser radar inversion algorithm to obtain extinction coefficient data corresponding to the pre-processed echo signal.
[0076] Specifically, the processing of micro-pulse lidar data includes background subtraction and filtering. Since the Raman lidar inversion algorithm does not require assumptions about the extinction coefficient of the atmospheric boundary points, the inversion result has a smaller error. In order to use the inversion coefficient of the Raman lidar as the label data of the neural network, it is also necessary to obtain the extinction coefficient profile of the aerosol through the Raman lidar inversion algorithm.
[0077] S40, using the preprocessed echo signal and the extinction coefficient data to train the RBF neural network to obtain an RBF neural network model, and the RBF neural network model can obtain the extinction coefficient data in real time according to the preprocessed echo signal.
[0078] Specifically, after the data is prepared, the present invention uses the RBF neural network for training, and uses the pre-processed echo signal and the extinction coefficient data to train the RBF neural network. Under a large amount of training data, the complex mapping relationship between the echo signal and the extinction coefficient is learned, so that the trained RBF neural network model can invert the pre-processed echo signal in real time to obtain the extinction coefficient data.
[0079] S50, when the actual echo signal collected by the micro-pulse laser radar is received, the actual echo signal is preprocessed to obtain the target echo signal, and the target echo signal is predicted and inverted using the RBF neural network model to obtain the extinction coefficient.
[0080] Specifically, after the RBF neural network model training is completed, the system enters the practical application stage. The actual echo signal collected by the micro-pulse lidar in real time is processed by background subtraction and noise reduction, and then input into the trained neural network model to invert the extinction coefficient.
[0081] This application takes into account that the inversion algorithm of Raman lidar does not require assumptions about the extinction coefficient of atmospheric boundary points, and uses micro-pulse lidar data and Raman lidar data to train an RBF neural network model, breaking through the vertical limitations of traditional technology. The inversion using the RBF neural network model improves the inversion accuracy and reduces errors.
[0082] In one embodiment, in order to ensure the accuracy of the inversion model, the step of synchronously observing and collecting training data using a micro-pulse laser radar and a high-energy Raman laser radar can be specifically performed as follows:
[0083] Control the micro-pulse lidar and high-energy Raman lidar to synchronously observe and collect training data in the same direction. Specifically, combined with Figure 2The micro-pulse laser radar and high-energy Raman laser radar work in coordination through a synchronous control system to ensure that the two radars collect data at the same time. The synchronous control system also controls the laser emission direction of the two laser radars to ensure that the detection directions of the two devices are consistent. Therefore, the synchronous control system ensures the time consistency of the data, so that the radar echo signal can correspond to the actual extinction coefficient data one by one, and ensures that the detection objects of the two laser radars are the same.
[0084] The micro-pulse lidar and high-energy Raman lidar are controlled by a synchronous control system to collect data in the same direction at the same time, ensuring that the detection objects are the same and improving the reliability of the RBF neural network model.
[0085] In one embodiment, in order to ensure the quality of data input into the neural network, the step of preprocessing the micro-pulse laser radar data to obtain the preprocessed echo signal can be specifically performed as follows:
[0086] First, the background noise of the micro-pulse lidar data is subtracted to obtain the first preprocessed echo signal. Specifically, when the micro-pulse lidar collects data, it will be affected by ambient light, detector noise, etc. and generate background noise. The first step of preprocessing is to subtract the background noise of the micro-pulse lidar data, remove the background noise in the echo signal, and improve the signal-to-noise ratio of the signal, so as to more accurately extract the target information.
[0087] Then, the high-frequency noise of the first preprocessed echo signal is removed to obtain the second preprocessed echo signal. Specifically, a low-pass filter is used to remove the high-frequency noise of the first preprocessed echo signal to ensure the smoothness of the echo signal and reduce the interference of signal noise.
[0088] Next, the drastically changing echo signals whose change amplitude is greater than the normal range are screened out in the second preprocessed echo signal, and the drastically changing echo signals in the second preprocessed echo signal are removed to obtain the preprocessed echo signal. Specifically, a normal range is set according to the characteristics of the echo signal. Clouds or obstructions can cause drastic changes in the echo signal, which is manifested in the echo signal as a change amplitude greater than the normal range. The drastically changing echo signals are eliminated to ensure the data quality of the input neural network.
[0089] In one embodiment, in order to ensure the quality of data input to the neural network, the step of preprocessing the Raman lidar data and using the Raman lidar inversion algorithm to obtain the extinction coefficient data corresponding to the preprocessed echo signal can be specifically performed as follows:
[0090] First, the background noise of the Raman lidar data is deducted to obtain the first Raman lidar signal, and the high-frequency noise of the first Raman lidar signal is removed to obtain the Raman lidar signal. Specifically, the steps of deducting the background noise and the high-frequency noise are consistent with the above-mentioned method for processing micro-pulse lidar data.
[0091] Then, the Raman lidar signal is processed by the Raman lidar inversion algorithm to obtain the extinction coefficient data corresponding to the preprocessed echo signal. Specifically, the extinction coefficient profile of the aerosol, that is, the extinction coefficient data, is obtained through the Raman lidar inversion algorithm. The extinction coefficient here is also the data that should be obtained by inverting the micro-pulse lidar data at the same time.
[0092] By preprocessing the micro-pulse lidar data and Raman lidar data, the RBF neural network training results are made more accurate.
[0093] In one embodiment, in order to improve the accuracy and reliability of the inversion model, the step of training the RBF neural network using the preprocessed echo signal and the extinction coefficient data to obtain the RBF neural network model can be specifically performed as follows:
[0094] Firstly, the preprocessed echo signal is used as the input data of the RBF neural network, and the extinction coefficient data is used as the label of the RBF neural network. Specifically, the RBF neural network is used for training. The RBF neural network is particularly suitable for solving nonlinear relationship problems and can learn the complex mapping relationship between the echo signal and the extinction coefficient under a large amount of training data. In order to obtain the mapping relationship, the preprocessed echo signal is used as the input data of the RBF neural network, and the extinction coefficient data is used as the label of the RBF neural network.
[0095] Then, the RBF neural network model is obtained by training the RBF neural network according to the input data and labels. Specifically, the RBF neural network adopts the radial basis function as the activation function and uses a high-performance GPU processor. Through a large number of training samples, the neural network optimizes the weights and biases through the back propagation algorithm, gradually approaches the actual extinction coefficient inversion result, and obtains the RBF neural network model.
[0096] The inversion algorithm of Raman lidar does not require any assumptions about the extinction coefficient of the atmospheric boundary points. The high-precision extinction coefficient data provided by Raman radar is used as training labels to improve the accuracy and reliability of the inversion model.
[0097] In one embodiment, after the RBF neural network is trained using the preprocessed echo signal and the extinction coefficient data to obtain the RBF neural network model, the following steps may be performed:
[0098] Firstly, the test data is collected and processed by synchronous observation of the micro-pulse laser radar and the high-energy Raman laser radar to obtain the test echo signal and the test extinction coefficient data. Specifically, the synchronous control system controls the micro-pulse laser radar and the high-energy Raman laser radar to collect the test data and perform pre-processing to obtain the test echo signal of the micro-pulse laser radar and the test extinction coefficient data of the high-energy Raman laser radar.
[0099] Then, the test echo signal is introduced into the RBF neural network model to obtain the target extinction coefficient data. Specifically, the test echo signal is introduced into the RBF neural network model for inversion to obtain the target extinction coefficient data under the RBF neural network model.
[0100] Next, the target extinction coefficient data and the test extinction coefficient data are compared to obtain the error, and it is determined whether the error is within the allowable range. Specifically, the error of the extinction coefficient of the predicted value output by the RBF neural network model and the test data is compared, and a threshold (representing the prediction accuracy of the model) is set to determine whether the error is less than this threshold.
[0101] If the error is not within the allowable range, it is determined that the RBF neural network model training has failed. Specifically, if there is a large error between the target extinction coefficient data and the extinction coefficient data obtained by inversion using the RBF neural network model and exceeds the threshold, it means that the RBF neural network model training has failed.
[0102] If the error is within the allowable range, it is determined that the RBF neural network model is successfully trained, and using the RBF neural network model to predict and invert the target echo signal to obtain the extinction coefficient includes: using the successfully trained RBF neural network model to predict and invert the target echo signal to obtain the extinction coefficient. Specifically, if the target extinction coefficient data obtained by inverting the RBF neural network model and the extinction coefficient data are within the allowable error range, the RBF neural network model is successfully trained. In actual use, the successfully trained RBF neural network model is used to invert the target echo signal to obtain the extinction coefficient.
[0103] The target extinction coefficient data is obtained by inputting the preprocessed micro-pulse lidar echo signal into the RBF neural network model, and then compared with the extinction coefficient data obtained by Raman lidar data processing to verify the RBF neural network model, so that the inversion accuracy of the RBF neural network model is guaranteed when it is actually used.
[0104] In one embodiment, the step of preprocessing the actual echo signal to obtain the target echo signal may be specifically performed as follows:
[0105] It is determined whether there is a drastic change echo signal with a change amplitude greater than a normal range in the actual echo signal. Specifically, whether there is a drastic change echo signal will affect the data preprocessing method.
[0106] If there are drastically changing echo signals in the actual echo signals, the actual echo signals are subjected to background subtraction and noise reduction processing, and then the drastically changing echo signals are processed by the linear interpolation method to obtain the target echo signals. Specifically, if there are drastically changing echo signals with a change amplitude greater than the normal range, after the conventional background subtraction and noise reduction preprocessing, the drastically changing echo signals are also required to be processed by the linear interpolation method to obtain the target echo signals, and then input into the model for inversion.
[0107] On the contrary, the target echo signal is obtained after background subtraction and noise reduction processing of the actual echo signal. Specifically, if there is no drastic change echo signal with a change amplitude greater than the normal range, conventional background subtraction and noise reduction preprocessing are performed and input into the model for inversion.
[0108] The present invention adopts a linear interpolation method to perform smoothing processing on these change regions, thereby eliminating discontinuities in the signal and avoiding the oscillation phenomenon of the extinction coefficient during the inversion process.
[0109] Based on the above method, the present application also discloses an extinction coefficient inversion system. Figure 3 The system includes the following modules:
[0110] A training data collection module 301 is used to collect training data by synchronously observing a micro-pulse laser radar and a high-energy Raman laser radar, wherein the training data includes micro-pulse laser radar data and Raman laser radar data;
[0111] The preprocessing echo signal module 302 is used to preprocess the micro-pulse laser radar data to obtain a preprocessed echo signal;
[0112] An extinction coefficient data module 303 is used to pre-process the Raman laser radar data and obtain the extinction coefficient data corresponding to the pre-processed echo signal using the Raman laser radar inversion algorithm;
[0113] RBF neural network model training module 304, used to train the RBF neural network using the pre-processed echo signal and the extinction coefficient data to obtain the RBF neural network model, and the RBF neural network model can obtain the extinction coefficient data in real time according to the pre-processed echo signal;
[0114] The extinction coefficient actual inversion module 305 is used to pre-process the actual echo signal to obtain the target echo signal when the actual echo signal collected by the micro-pulse laser radar is received, and use the RBF neural network model to predict and invert the target echo signal to obtain the extinction coefficient.
[0115] In one embodiment, the test data collection module 301 is specifically used to control the micro-pulse laser radar and the high-energy Raman laser radar to synchronously observe and collect training data in the same direction.
[0116] In one embodiment, the preprocessing echo signal module 302 is specifically used to deduct the background noise of the micro-pulse lidar data to obtain a first preprocessing echo signal; remove the high-frequency noise of the first preprocessing echo signal to obtain a second preprocessing echo signal; screen out the drastically changing echo signals in the second preprocessing echo signal whose change amplitude is greater than the normal range; and remove the drastically changing echo signals in the second preprocessing echo signal to obtain a preprocessing echo signal.
[0117] In one embodiment, the extinction coefficient data module 303 is specifically used to subtract the background noise of the Raman lidar data to obtain a first Raman lidar signal; remove the high-frequency noise of the first Raman lidar signal to obtain a Raman lidar signal; process the Raman lidar signal through a Raman lidar inversion algorithm to obtain extinction coefficient data corresponding to the preprocessed echo signal.
[0118] In one embodiment, the RBF neural network model training module 304 is specifically used to use the preprocessed echo signal as the input data of the RBF neural network and the extinction coefficient data as the label of the RBF neural network; and obtain the RBF neural network model by training the RBF neural network according to the input data and the label.
[0119] In one embodiment, the extinction coefficient actual inversion module 305 is specifically used to determine whether there is a drastically changing echo signal with a change amplitude greater than a normal range in the actual echo signal; if there is a drastically changing echo signal in the actual echo signal, the actual echo signal is subjected to background subtraction and noise reduction processing, and then the drastically changing echo signal is processed by a linear interpolation method to obtain a target echo signal; otherwise, the actual echo signal is subjected to background subtraction and noise reduction processing to obtain a target echo signal.
[0120] The embodiment of the present application also discloses a computer device.
[0121] Specifically, the computer device includes a memory and a processor, and the memory stores a computer program that can be loaded by the processor and execute the above-mentioned extinction coefficient inversion method.
[0122] The embodiment of the present application also discloses a computer-readable storage medium.
[0123] Specifically, the computer-readable storage medium stores a computer program that can be loaded by a processor and execute the extinction coefficient inversion method as described above. The computer-readable storage medium includes, for example: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.
[0124] This specific embodiment is merely an explanation of the present invention and is not a limitation of the present invention. After reading this specification, those skilled in the art may make non-creative modifications to the present embodiment as needed. However, as long as they are within the scope of the claims of the present invention, they are protected by the patent law.
Claims
1. A method for inverting an extinction coefficient, characterized in that: The method is based on an extinction coefficient inversion system comprising an RBF neural network, a micro-pulse laser radar and a high-energy Raman laser radar, and the method comprises: Using the micro-pulse laser radar and the high-energy Raman laser radar to synchronously observe and collect training data, the training data includes micro-pulse laser radar data and Raman laser radar data; Preprocessing the micro-pulse laser radar data to obtain a preprocessed echo signal; Preprocessing the Raman laser radar data and using a Raman laser radar inversion algorithm to obtain extinction coefficient data corresponding to the preprocessed echo signal; Using the pre-processed echo signal and the extinction coefficient data to train the RBF neural network to obtain an RBF neural network model, wherein the RBF neural network model can obtain the extinction coefficient data in real-time inversion according to the pre-processed echo signal; When the actual echo signal collected by the micro-pulse laser radar is received, the actual echo signal is preprocessed to obtain the target echo signal, and the RBF neural network model is used to predict and invert the target echo signal to obtain the extinction coefficient.
2. The method according to claim 1, characterized in that: The synchronous observation and collection of training data by the micro-pulse laser radar and the high-energy Raman laser radar comprises: The micro-pulse laser radar and the high-energy Raman laser radar are controlled to synchronously observe and collect training data in the same direction.
3. The method according to claim 1, characterized in that: The preprocessing of the micro-pulse laser radar data to obtain a preprocessed echo signal comprises: Deducting background noise from the micro-pulse laser radar data to obtain a first pre-processed echo signal; removing high frequency noise from the first preprocessed echo signal to obtain a second preprocessed echo signal; Screening out the drastically changed echo signals whose change amplitude is greater than a normal range in the second preprocessed echo signals; The drastically changing echo signal in the second preprocessed echo signal is removed to obtain a preprocessed echo signal.
4. The method according to claim 1, characterized in that: The preprocessing of the Raman laser radar data and obtaining the extinction coefficient data corresponding to the preprocessed echo signal using the Raman laser radar inversion algorithm includes: Deducting the background noise of the Raman laser radar data to obtain a first Raman laser radar signal; Removing high-frequency noise from the first Raman laser radar signal to obtain a Raman laser radar signal; The Raman laser radar signal is processed by a Raman laser radar inversion algorithm to obtain extinction coefficient data corresponding to the preprocessed echo signal.
5. The method according to claim 1, characterized in that: The method of training the RBF neural network using the pre-processed echo signal and the extinction coefficient data to obtain the RBF neural network model comprises: Using the pre-processed echo signal as input data of the RBF neural network and using the extinction coefficient data as a label of the RBF neural network; The RBF neural network model is obtained by training the RBF neural network according to the input data and the label.
6. The method according to claim 1, characterized in that: After the RBF neural network is trained using the pre-processed echo signal and the extinction coefficient data to obtain the RBF neural network model, the method further includes: Utilizing the micro-pulse laser radar and the high-energy Raman laser radar to synchronously observe, collect and process test data to obtain a test echo signal and test extinction coefficient data; Importing the test echo signal into the RBF neural network model to obtain target extinction coefficient data; Comparing the target extinction coefficient data with the test extinction coefficient data to obtain an error, and determining whether the error is within an allowable range; If the error is not within the allowable range, it is determined that the RBF neural network model training has failed; If the error is within the allowable range, it is determined that the RBF neural network model is successfully trained, and the use of the RBF neural network model to predict and invert the target echo signal to obtain the extinction coefficient includes: The successfully trained RBF neural network model is used to predict and invert the target echo signal to obtain the extinction coefficient.
7. The method according to claim 1, characterized in that: The preprocessing of the actual echo signal to obtain the target echo signal comprises: Determine whether there is a drastic change echo signal with a change amplitude greater than a normal range in the actual echo signal; If the actual echo signal contains the drastically changing echo signal, the actual echo signal is subjected to background subtraction and noise reduction processing, and then the drastically changing echo signal is processed by a linear interpolation method to obtain a target echo signal; Otherwise, the target echo signal is obtained after background subtraction and noise reduction processing on the actual echo signal.
8. An extinction coefficient inversion system, characterized in that: The system comprises: A training data collection module (301) is used to collect training data by synchronously observing the micro-pulse laser radar and the high-energy Raman laser radar, wherein the training data includes micro-pulse laser radar data and Raman laser radar data; A preprocessing echo signal module (302), used for preprocessing the micro-pulse laser radar data to obtain a preprocessing echo signal; An extinction coefficient data module (303) is used to pre-process the Raman laser radar data and obtain extinction coefficient data corresponding to the pre-processed echo signal using a Raman laser radar inversion algorithm; An RBF neural network model training module (304) is used to train the RBF neural network using the pre-processed echo signal and the extinction coefficient data to obtain an RBF neural network model, wherein the RBF neural network model can obtain the extinction coefficient data by real-time inversion according to the pre-processed echo signal; The extinction coefficient actual inversion module (305) is used to pre-process the actual echo signal to obtain the target echo signal when the actual echo signal collected by the micro-pulse laser radar is received, and use the RBF neural network model to predict and invert the target echo signal to obtain the extinction coefficient.
9. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program which can be loaded by the processor and executes the method according to any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that: A computer program is stored which can be loaded by a processor and execute the method according to any one of claims 1 to 7.