Pulse laser ranging accuracy compensation method, system, computer and storage medium

Abstract

The present invention discloses a pulse laser ranging accuracy compensation method, system, computer and storage medium, the method comprising the following steps: building a laser ranging test environment, and collecting test data in the test environment; using a deep neural network to construct a pulse width anomaly detector, inputting the test data into the anomaly detector, and obtaining an anomaly detection result; eliminating the abnormal data, and using a fitting compensation method to compensate the pulse width of the test data; by calibration, compensating the system error, and completing the accuracy correction. The pulse width anomaly detector based on the deep learning network of the present invention can combine the original data layer and feature layer information of the photoelectric sensor to make more accurate abnormal value corrections to the pulse width and relative forward displacement of the echo signal obtained by the rangefinder, and at the same time, by using the pulse width compensation method, it can characterize the echo intensity in the form of pulse width, and at the same time, it can suppress the front wandering error, realize the adaptive accuracy compensation function, and improve the ranging accuracy.

Description

Technical Field

[0001] The present invention relates to the field of laser ranging technology, and in particular to a pulse laser ranging accuracy compensation method, system, computer and storage medium. Background Art

[0002] There are three methods for laser ranging: triangulation, interferometry, and time of flight. The time of flight method is widely used in military fields such as aviation and aerospace because it requires few auxiliary equipment and has no working period restrictions. The time of flight measurement method can be divided into two methods: phase ranging and pulse ranging. Pulse laser ranging technology is widely used in the detection of non-cooperative targets due to its narrow pulse width and high peak power. Its basic principle is to use extremely narrow pulses with high peak power to illuminate non-cooperative targets in space, collect the time difference between the signal echo reflected or scattered by the target and the transmitted signal, and use TOF to calculate the distance of the target.

[0003] First, the advantages of pulse ranging are long range, simple structure, low power consumption, but the ranging accuracy is slightly poor. The factors that affect the ranging accuracy include: random walk error and system error.

[0004] Random walk errors generally follow Gaussian distribution, including time interval measurement errors, waveform anomaly errors, comparator rising edge jitter, moment identification jitter and other errors. Time interval measurement is mainly achieved by synchronous counters. At present, the accuracy of timing chips can reach 10ps, and the distance error introduced is 1.5mm, which can be ignored in long-distance measurement; waveform anomaly errors are affected by irregular targets, pulse peak jitter, external interference and other factors, and cannot be accurately corrected; comparator rising edge jitter is affected by power supply ripple, impedance mismatch and other factors, and can be achieved through LVPECL and LVTTL circuits. The accuracy of 20ps and the distance error introduced is 3mm, which can be ignored; moment identification errors are affected by noise superposition and can produce ns-level errors. The constant ratio timing method can suppress the timing error caused by the fluctuation of the laser pulse echo amplitude to a certain extent. However, the constant ratio timing method is sensitive to changes in pulse width, and its implementation circuit is extremely complex.

[0005] System errors include hardware fixed delay, clock drift error, echo optical power drift, etc. Fixed delay is a fixed deviation that does not change with the distance to the target, and is usually eliminated by calibration methods; the source of clock drift error is the error caused by temperature and power supply changes in the timing chip, which is usually reduced by temperature compensation and high-precision power supply systems; and the change in echo optical power will cause the pulse leading and trailing edges of the photodetector to widen in different directions, resulting in wandering errors. This wandering error is a major problem that limits the accuracy of pulse laser ranging in short-range detection.

[0006] From the mechanism analysis of the photodetector, for a target at a fixed position, as the non-cooperative target enters the detection field from outside to inside, the photogenerated carriers of the photodetector chip increase, overflow, and discharge in three processes, corresponding to the linear amplification area, near saturation, shallow saturation, and deep saturation of the preamplifier circuit; the waveform of the echo signal has three dynamic changes: pulse width widening when the amplitude increases, pulse width rear edge widening when the amplitude remains unchanged, and front and rear edges widening when the amplitude remains unchanged. This change leads to wandering drift error at the identification moment.

[0007] At present, there are many methods to solve the wandering error of the echo signal caused by the change of echo light intensity, such as constant ratio timing, dual threshold, full sampling, etc. Among them, the constant ratio timing method has a complex circuit and is greatly affected by the fluctuation of pulse width; the self-adaptation of high and low thresholds in dual threshold compensation is a difficult point; full sampling can fully process the waveform information, but the high-speed ADC sampling circuit is complex and costly, and is suitable for high-precision long-distance target detection.

[0008] Secondly, during the ranging process, affected by waveform jitter and clock drift errors, the measured data will produce abnormal fluctuations, resulting in singular values ​​or no data output. Currently, most projects use threshold detection to propose abnormal values, but the threshold setting is not flexible and difficult to adapt to a variety of ranging modes. Summary of the invention

[0009] In view of the deficiencies of the above-mentioned prior art, the technical problem to be solved by the present invention is: to provide a pulse laser ranging accuracy compensation method, system, computer and storage medium, which can perform more accurate outlier correction on the pulse width and relative advance amount of the echo signal obtained by the rangefinder, and solve the problem of the front and rear edge wandering error of the echo signal caused by the change of the echo light intensity.

[0010] In order to solve the above technical problems, the present invention designs a multi-layer constrained information pulse width anomaly detector based on a Bi-LSTM deep learning network, which can combine the original data layer and feature layer information of the photoelectric sensor to make more accurate abnormal value corrections to the pulse width and relative forward displacement of the echo signal obtained by the rangefinder, and proposes a pulse width compensation method based on 1D-CNN. This method can characterize the echo intensity in the form of pulse width, suppress the front wandering error, realize the adaptive precision compensation function, and improve the ranging accuracy.

[0011] A technical solution adopted by the present invention is to provide a pulse laser ranging accuracy compensation method, comprising the following steps:

[0012] S1: Build a laser ranging test environment and collect test data in the test environment;

[0013] S2: Use a deep neural network to build a pulse width anomaly detector, input the test data into the anomaly detector, and obtain the anomaly detection result;

[0014] S3: Eliminate abnormal data and use fitting compensation method to compensate the pulse width of test data;

[0015] S4: Compensate for system errors through calibration.

[0016] Furthermore, after step S4, the method further includes the following steps:

[0017] S5: Randomly select a distance measurement point to verify the feasibility of pulse width compensation and accuracy correction.

[0018] Furthermore, step S1 includes the following sub-steps:

[0019] S11: placing the target plate at a preset distance point, and adjusting the pitch platform so that the detection field of view of the rangefinder is higher than the target plate;

[0020] S12: Use a PD detector to collect the light emission time of the Er glass laser as the start signal, and record it after passing through the TDC circuit. The rising time is stt1 and the falling time is stt2;

[0021] S13: Adjust the pitch stage to move the detection field of the rangefinder downward until the detection field of view completely sweeps over the target plate. The echo signal collected by the APD detector is used as the stop signal and recorded after passing through the TDC circuit. The rising time is stp 1i , the falling time is stp 2i .

[0022] Furthermore, in step S2, the following sub-steps are included:

[0023] S21: Construct a Bi-LSTM neural network pulse width anomaly detector, whose forward propagation is:

[0024]

[0025] Among them, i k 、f k and k Respectively represent input, forgetting and output units; X k is the input vector of Bi-LSTM at time k; h k represents the hidden state vector; σ represents the sigmoid activation function, and tanh represents the tanh activation function; Represents the candidate vector, which is combined with the output vector as the memory state at time k; C k Indicates the current cell state; W i , W f , W o and W cRepresents the weight matrices of input gate, forget gate, output gate and cell state respectively; b i , b f , b o and b c Represent the bias terms of input gate, forget gate, output gate and cell state respectively;

[0026] S22: The back propagation of the Bi-LSTM neural network pulse width anomaly detector uses a mean square error function as a loss function, which can be specifically expressed as:

[0027]

[0028] Among them, Loss represents the MSE loss function; n represents the number of samples; x i represents the true value of the i-th sample; y i Represents the prediction value of the detector for the i-th sample.

[0029] Furthermore, in step S2, the following sub-steps are also included:

[0030] S23: Modeling the input vector of the Bi-LSTM neural network pulse width anomaly detector according to the test data, including:

[0031] Reference echo pulse width value: BW PD =stt2-stt1;

[0032] Signal echo pulse width: BW APDi =stp 2i -stp 1i ;

[0033] Front position change:

[0034] Pulse width variance:

[0035] Variance of frontier position change:

[0036] Direct distance value: R APDi =(stp 1i -stt1)*1000 / 150;

[0037] Interquartile range of direct distance:

[0038] Among them, BW APD (j) represents the pulse width of the jth signal echo; tt APDi Indicates the offset of the front edge of the i-th signal echo; i and j are serial numbers, representing the number of samples under different optical powers; represents the first quartile of the direct distance value; represents the third quartile of direct distance values;

[0039] S24: The Bi-LSTM neural network pulse width anomaly detector outputs an anomaly detection result based on the input vector.

[0040] Furthermore, in step S3, the following sub-steps are included:

[0041] S31: Construct a 1D-CNN network. The specific structure can be expressed as:

[0042]

[0043] Among them, x i represents the input features of the 1D-CNN network; k is represents the kernel weight; f() represents the RelU activation function; b s Indicates bias; O s Represents the output features of the 1D-CNN network; N represents the number of input features.

[0044] S32: Model the input vector and output vector of the 1D-CNN network. The input vector can be expressed as BW APDi =stp 2i -stp 1i ;

[0045] The output vector can be expressed as

[0046] Furthermore, in step S4, the system error is compensated, specifically including the following formula:

[0047] R=γ*f(i)+R z

[0048] Among them, γ represents the compensation weight, f(i) represents the measured distance value, R z Represents the true value of the measurement.

[0049] In order to solve the above technical problems, the second technical solution adopted by the present invention is: to provide a pulse laser ranging accuracy compensation system, comprising:

[0050] Data acquisition module, used to build a laser ranging test environment and collect test data in the test environment;

[0051] An anomaly detection module is used to construct a pulse width anomaly detector using a deep neural network, input test data into the anomaly detector, and obtain an anomaly detection result;

[0052] The pulse width compensation module is used to eliminate abnormal data and compensate the pulse width of the test data using a fitting compensation method;

[0053] The error compensation module is used to compensate for the system error through calibration to complete the accuracy correction.

[0054] To solve the above technical problems, the third technical solution adopted by the present invention is: a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the above-mentioned accuracy compensation methods.

[0055] In order to solve the above technical problems, the fourth technical solution adopted by the present invention is: a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the accuracy compensation method described in any one of the above items is implemented.

[0056] The pulse laser ranging accuracy compensation method, system, computer and medium of the present invention have at least the following beneficial effects: 1. The present invention uses a multi-layer constraint information pulse width anomaly detector based on a Bi-LSTM deep learning network, so that the pulse width and relative forward displacement of the echo signal obtained by the rangefinder can be combined with the original data layer and feature layer information of the photoelectric sensor to make more accurate anomaly corrections; 2. The present invention solves the waveform front drift error caused by the echo intensity change through a pulse width compensation method based on 1D-CNN, and at the same time avoids the waveform front caused by the amplitude change in the linear amplification area to improve the ranging accuracy; 3. Compared with the existing segmented pulse width compensation method, the method of the present invention improves the robustness of the ranging accuracy at the segmentation point and the full pulse width, and is suitable for ranging scenarios with different target types, optical mechanical energies and ranging ranges. 4. This method characterizes different target intensity information in the form of pulse width, reduces the ranging blind area, and expands the application range of the ranging system. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0058] Figure 1 It is a flow chart of an implementation method of the pulse laser ranging accuracy compensation method of the present invention.

[0059] Figure 2 Schematic diagram of hardware construction for laser ranging test of the present invention.

[0060] Figure 3 This is a schematic diagram of the environment construction for the laser ranging test of the present invention.

[0061] Figure 4 for Figure 1 Sub-flow chart of step S1 in .

[0062] Figure 5 for Figure 1 Sub-flow chart of step S2 in .

[0063] Figure 6 Schematic diagram of the change process of the leading and trailing edge drift error of the waveform.

[0064] Figure 7 for Figure 1 Sub-flow chart of step S3 in .

[0065] Figure 8 It is a structural block diagram of an implementation manner of a ranging accuracy compensation system of the present invention. DETAILED DESCRIPTION

[0066] The present invention will be further described below in conjunction with the accompanying drawings.

[0067] See also Figure 1 , is a flow chart of an implementation method of the pulse laser ranging accuracy compensation method of the present invention. This implementation method may specifically include the following steps:

[0068] S1: Build a laser ranging test environment and collect test data in the built test environment.

[0069] Specifically, in this embodiment, a specific hardware system, such as Figure 2 As shown, it may include an Er glass laser with a driving circuit, a transmitting antenna, a receiving lens, an InGaAs APD detector with a preamplifier, an InGaAs PD detector, a high-speed digital comparator, a signal processing unit, a high-precision TDC circuit, a power supply system and a communication module. Compared with the conventional mrad transmitting optical antenna, the present invention sets a transmitting full angle of 2°, which is convenient for dynamically capturing the process of a long-distance non-cooperative target entering the field of view. During operation, a part of the Er glass laser pulse signal (1535nm) enters the PD detector and is converted into an analog electrical signal, which is converted into a digital signal start through a high-speed digital conversion circuit; a part hits the non-cooperative target after beam expansion, and the signal is converged to the APD detector through the receiving lens and converted into an analog electrical signal, and then converted into a digital signal stop through a high-speed digital conversion circuit; the two digital signals measure the time interval and pulse width through a high-precision timing circuit, and are transmitted to the FPGA for data compensation and correction, and the distance calculation result is output through the communication module.

[0070] In addition, the acquisition of raw data is a key step in the implementation of the present invention. Figure 3The test environment shown in the figure includes the main modules such as the rangefinder, total station, tripod with pitch table, target plate, host computer, etc. The measurement accuracy of the total station is ±2mm, which is used as the true value of the measurement; the transmitting field angle of the rangefinder is 2°, the reflectivity of the target plate is r=0.2, the diameter is 1m, the pitch table step is 20′, and the host computer reads and stores the raw data.

[0071] See also Figure 4 , this step S1 may include the following sub-steps:

[0072] S11: placing the target plate at a preset distance point, and adjusting the pitch platform so that the detection field of view of the rangefinder is higher than the target plate.

[0073] Specifically, in this embodiment, the target plate is placed at a certain distance point to facilitate the acquisition of full pulse width data within the energy range, and the pitch platform is adjusted so that the detection field of view of the rangefinder is higher than the target.

[0074] S12: Use a PD detector to collect the light emission time of the Er glass laser as the start signal, and record it after passing through the TDC circuit. The rising time is stt1 and the falling time is stt2.

[0075] Specifically, an InGaAs PD detector is used to collect the 1535nm laser light emission time as a start signal, and the rising time stt1 and the falling time stt2 are recorded after passing through a TDC.

[0076] S13: Adjust the pitch stage to move the detection field of the rangefinder downward until the detection field of view completely sweeps over the target plate. The echo signal collected by the APD detector is used as the stop signal and recorded after passing through the TDC circuit. The rising time is stp 1i , the falling time is stp 2i .

[0077] Specifically, the pitch stage is adjusted to move the detection field of the rangefinder downward to the target plate. The displacement of the pitch stage is 20'. The InGaAsAPD detector collects the signal echo signal as the stop signal and records it after passing through the TDC. The rising time stp 1i , falling time stp 2i (i=1,2,3,…n), the detection field is completely hit on the target board.

[0078] S2: Use a deep neural network to build a pulse width anomaly detector, input the test data into the anomaly detector, and obtain the anomaly detection results.

[0079] Specifically, in the ranging scenario, affected by waveform jitter and clock drift errors, the measured data produces abnormal fluctuations or no data output. To address this problem, a multi-layer constrained information pulse width anomaly detector based on a Bi-LSTM deep learning network is designed in this embodiment. The detector can combine the original data layer and feature layer information of the photoelectric sensor to make more accurate abnormal value corrections to the pulse width and relative forward displacement of the echo signal obtained by the rangefinder.

[0080] See also Figure 5 , this step S2 may include the following sub-steps:

[0081] S21: Construct a Bi-LSTM neural network pulse width anomaly detector.

[0082] Specifically, in this implementation, the forward propagation of the Bi-LSTM neural network is:

[0083]

[0084] Among them, i k 、f k and k Respectively represent input, forgetting and output units; X k is the input vector of Bi-LSTM at time k; h k represents the hidden state vector; σ represents the sigmoid activation function, and tanh represents the tanh activation function; Represents the candidate vector, which is combined with the output vector as the memory state at time k; C k Indicates the current cell state; W i , W f , W o and W c Represents the weight matrices of input gate, forget gate, output gate and cell state respectively; b i , b f , b o and b c They represent the bias terms of the input gate, forget gate, output gate and cell state respectively.

[0085] S22: The back propagation of the Bi-LSTM neural network pulse width anomaly detector uses the mean square error function as the loss function.

[0086] Specifically, in this implementation, the above loss function can be specifically expressed as:

[0087]

[0088] Among them, Loss represents the MSE loss function; n represents the number of samples; x i represents the true value of the i-th sample; yi Represents the prediction value of the detector for the i-th sample.

[0089] In some implementations, this step S2 may further include the following steps:

[0090] S23: Modeling the input vector of the Bi-LSTM neural network pulse width anomaly detector according to the test data.

[0091] Specifically, in this embodiment, the input vector X of the Bi-LSTM neural network pulse width anomaly detector is k , modeled as, the following raw data and characteristic parameters:

[0092] 1. Start1, start2 and reference echo pulse width value BW of PD photoelectric sensor PD :

[0093] BW PD =stt2-stt1;

[0094] 2. Stop1, stop2 and signal echo pulse width value BW of APD photoelectric sensor APD :

[0095] BW APDi =stp 2i -stp 1i ;

[0096] 3. Change in the front position of the APD photoelectric sensor tt APDi :

[0097]

[0098] 4. Pulse width variance BE output by APD photoelectric sensor APD 2 :

[0099]

[0100] 5. APD photoelectric sensor front position change variance tt APD 2 :

[0101]

[0102] 6. Direct distance value R output by APD photoelectric sensor APDi :

[0103] R APDi =(stp 1i -stt1)t1000 / 150;

[0104] 7. Interquartile range IQR of the direct distance value output by the APD photoelectric sensor RAPD :

[0105]

[0106] Among them, BW APD (j) represents the pulse width of the jth signal echo; tt APDi Indicates the offset of the front edge of the i-th signal echo; i and j are serial numbers, representing the number of samples under different optical powers; represents the first quartile of the direct distance value; Represents the third quartile of direct distance values.

[0107] S24: The Bi-LSTM neural network pulse width anomaly detector outputs an anomaly detection result based on the input vector.

[0108] Specifically, in this embodiment, during the training phase, the input vector of the Bi-LSTM is modeled as a combination of feature values ​​extracted from two photoelectric sensors: the output vector is classified into two results by softmax, represented by 1 and 0. 0 represents no abnormality and 1 represents abnormality. For abnormal values, the corresponding BW APDi =(BW APD(i-1) +BW APD(i+1) ) / 2 and tt APDi =(tt APD(i-1) +tt APD(i+1) ) / 2. In the prediction stage, the modeled vector is directly used as the input vector, and the trained Bi-LSTM network is used to autonomously predict and output the anomaly detection result.

[0109] S3: Eliminate abnormal data and use the fitting compensation method to compensate the pulse width of the test data.

[0110] Specifically, the pulse width compensation algorithm is another important part of the present invention. In this embodiment, after eliminating abnormal data, the fitting degree of pulse width and leading edge position change determines the key factor of ranging accuracy. Through preliminary data analysis, as the echo power increases, the leading and trailing edge drift error change process of the waveform is Figure 6 As shown in the waveforms in . Among them, waveform 1 to waveform 2 are the linear amplification area of ​​the photodetector. In this process, the waveform amplitude increases, the pulse width is slightly widened, and the front and rear edges have a slight walking trend; waveform 2 to waveform 3 are the shallow saturation state of the photodetector. In this process, the waveform amplitude remains unchanged, the front edge is slightly widened, and the rear edge has a significant trend of wandering with the optical power; waveform 3 to waveform 4 are the photodetector entering the deep saturation state. In this process, the waveform amplitude remains unchanged, the front edge moves forward more, and the rear edge wanders relatively less. This process has a nonlinear change trend, and linear compensation will bring a large error.

[0111] It is worth mentioning that in pulse width compensation, conventional methods generally adopt the segmented fitting method. Figure 6 As an example, the waveform in Figure 6 The waveform in is piecewise fitted, which can be specifically expressed as:

[0112] When the pulse width BW APDi ≤66ns, leading edge position compensation Δtt APDi The fit can be expressed as:

[0113] Δtt APDi =0.0048BW APDi 2 -0.5086BW APDi +26.323

[0114] When pulse width 66ns<BW APDi ≤500ns, front position compensation Δtt APDi The fit can be expressed as:

[0115] Δtt APDi =-0.0151BW APDi +12.92

[0116] When the pulse width BW APDi >500ns, front position compensation Δtt APDi The fit can be expressed as:

[0117] Δtt APDi =-9×10- 6 BW APDi 5 +0.0227BW APDi 4 -23.478BW APDi 3 -3×10- 6 BW APDi +3×10 8

[0118] See also Figure 7 , this step S3 may include the following sub-steps:

[0119] S31: Build a 1D-CNN network.

[0120] Specifically, this implementation simulates a non-cooperative target entering the detection field of view of the rangefinder from far to near, so that the echo signal of the InGaAsAPD detector changes from a small enhancement. A regression analysis algorithm based on 1D-CNN is designed to better fit the trend of the offset of the front edge of the echo signal changing with the pulse width.

[0121] The input vector is obtained by modeling the characteristic output waveform of the InGaAsAPD photoelectric sensor collected by TDC, and the movement of the front position is calculated as the output vector of the 1D-CNN network. The specific 1D-CNN network can be expressed as:

[0122]

[0123] Among them, x i represents the input features of the 1D-CNN network; k is represents the kernel weight; f() represents the RelU activation function; b s Indicates bias; O s Represents the output features of the 1D-CNN network.

[0124] S32: Modeling the input vector and output vector of the 1D-CNN network.

[0125] Specifically, for the above 1D-CNN network, its input vector and output vector are modeled as follows: the input vector can be expressed as BW APDi =stp 2i -stp 1i ;

[0126] The output vector can be expressed as

[0127] S4: Compensate for system errors through calibration to complete accuracy correction.

[0128] Specifically, the distance measurement error caused by the fixed delay of the circuit system is called a system error. This error does not change with the measured distance and is usually compensated by calibration. This method is used for compensation in this embodiment.

[0129] In some implementations, the compensation for the system error in step S4 specifically includes the following formula:

[0130] R=γ*f(i)+R z

[0131] Among them, γ represents the compensation weight, f(i) represents the measured distance value, R z Represents the true value of the measurement.

[0132] In order to verify the feasibility of the above-mentioned pulse width compensation method and precision correction, in this embodiment, the following steps are also included:

[0133] S5: Randomly select a distance measurement point to verify the feasibility of pulse width compensation and accuracy correction.

[0134] Specifically, in this implementation, a ranging point (such as 50.57m) can be randomly selected to verify the feasibility of pulse width compensation and precision correction. Within the ranging capability, ranging data under different echo energies are randomly collected within the dynamic range of 4m to 110m, and the average ranging accuracy of the 1D-CNN pulse width compensation and precision correction method is calculated.

[0135] See also Figure 8 , is a structural block diagram of an embodiment of the pulse laser ranging accuracy compensation system of the present invention. The pulse laser ranging accuracy compensation system of this embodiment is used to implement the pulse laser ranging accuracy compensation method described in the above embodiment. Specifically, the pulse laser ranging accuracy compensation system of this embodiment includes a data acquisition module 100, an abnormality detection module 200, a pulse width compensation module 300 and an error compensation module 400. Among them:

[0136] The data acquisition module 100 is used to build a laser ranging test environment and collect test data in the test environment; the anomaly detection module 200 is used to construct a pulse width anomaly detector using a deep neural network, input the test data into the anomaly detector, and obtain an anomaly detection result; the pulse width compensation module 300 is used to eliminate abnormal data and compensate for the pulse width of the test data using a fitting compensation method; the error compensation module 400 is used to compensate for the system error by calibration to complete precision correction.

[0137] In order to ensure the smoothness of the motion trajectory of the device, in this embodiment, a distance measurement verification module 500 is further provided, which is used to randomly select a distance measurement point to verify the feasibility of pulse width compensation and precision correction.

[0138] The specific principles of the data acquisition module 100, the anomaly detection module 200, the pulse width compensation module 300, the error compensation module 400 and the distance measurement verification module 500 can be found in the description of the corresponding method steps in the method embodiment above, so they will not be repeated here.

[0139] The present invention uses a multi-layer constrained information pulse width anomaly detector based on a Bi-LSTM deep learning network, so that the pulse width and relative forward displacement of the echo signal obtained by the rangefinder can be combined with the original data layer and feature layer information of the photoelectric sensor to perform more accurate abnormal value correction; through a pulse width compensation method based on 1D-CNN, the waveform front drift error caused by the echo intensity change is solved, and the waveform front caused by the amplitude change in the linear amplification area is avoided to improve the ranging accuracy; compared with the existing segmented pulse width compensation method, the method of the present invention improves the robustness of the ranging accuracy at the segmentation point and the full pulse width, and is suitable for ranging scenarios of different target types, optical mechanical energies and ranging ranges. Characterizing different target intensity information in the form of pulse width reduces the ranging blind area and expands the application range of the ranging system.

[0140] On the other hand, an embodiment of the present invention further provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any one of the above-mentioned accuracy compensation methods. The computer device comprises a processor, a memory, a network interface, a display screen, and an input device connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a multi-agent data analysis method based on a large language model is implemented. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a key, a trackball, or a touchpad provided on the housing of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0141] On the other hand, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the accuracy compensation method described in any one of the above items is implemented.

[0142] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0143] The above content only expresses the preferred embodiments of the present invention, and its description is relatively specific and detailed, but it cannot be understood as limiting the scope of the invention patent. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be based on the attached claims.

Claims

1. A pulse laser ranging accuracy compensation method, characterized in that: The following steps are involved: S1: Build a laser ranging test environment and collect test data in the built test environment; S2: Use a deep neural network to build a pulse width anomaly detector, input the test data into the anomaly detector, and obtain the anomaly detection result; S3: Eliminate abnormal data and use fitting compensation method to compensate the pulse width of test data; S4: Compensate for system errors through calibration to complete accuracy correction.

2. The pulse laser ranging accuracy compensation method according to claim 1, characterized in that: After step S4, the following steps are also included: S5: Randomly select a distance measurement point to verify the feasibility of pulse width compensation and accuracy correction.

3. The pulse laser ranging accuracy compensation method according to claim 1, characterized in that: The S1 step includes the following sub-steps: S11: placing the target plate at a preset distance point, and adjusting the pitch platform so that the detection field of view of the rangefinder is higher than the target plate; S12: Use a PD detector to collect the light emission time of the Er glass laser as the start signal, and record it after passing through the TDC circuit. The rising time is stt1 and the falling time is stt2; S13: Adjust the pitch stage to move the detection field of the rangefinder downward until the detection field of view completely sweeps over the target plate. The echo signal collected by the APD detector is used as the stop signal and recorded after passing through the TDC circuit. The rising time is stp 1i , the falling time is stp 2i .

4. The pulse laser ranging accuracy compensation method according to claim 3, characterized in that: In step S2, the following sub-steps are included: S21: Construct a Bi-LSTM neural network pulse width anomaly detector, whose forward propagation is: Among them, i k 、f k and k Respectively represent input, forgetting and output units; X k is the input vector of Bi-LSTM at time k; h k represents the hidden state vector; σ represents the sigmoid activation function, and tanh represents the tanh activation function; Represents the candidate vector, which is combined with the output vector as the memory state at time k; C k Indicates the current cell state; W i , W f , W o and W c Represents the weight matrices of input gate, forget gate, output gate and cell state respectively; b i 、b f 、b o and b c Represent the bias terms of input gate, forget gate, output gate and cell state respectively; S22: The back propagation of the Bi-LSTM neural network pulse width anomaly detector uses a mean square error function as a loss function, which can be specifically expressed as: Among them, Loss represents the MSE loss function; n represents the number of samples; x i represents the true value of the i-th sample; y i Represents the prediction value of the detector for the i-th sample.

5. The pulse laser ranging accuracy compensation method according to claim 4, characterized in that: In step S2, the following sub-steps are also included: S23: Modeling the input vector of the Bi-LSTM neural network pulse width anomaly detector according to the test data, including: Reference echo pulse width value: BW PD =stt2-stt1; Signal echo pulse width: BW APDi =stp 2i -stp 1i ; Front position change: Pulse width variance: Variance of frontier position change: Direct distance value: R APDi =(stp 1i -stt1)*1000 / 150; Interquartile range of direct distance: Among them, BW APD (j) represents the pulse width of the jth signal echo; tt APDi Indicates the offset of the front edge of the i-th signal echo; i and j are serial numbers, representing the number of samples under different optical powers; represents the first quartile of the direct distance value; represents the third quartile of direct distance values; S24: The Bi-LSTM neural network pulse width anomaly detector outputs an anomaly detection result based on the input vector.

6. The pulse laser ranging accuracy compensation method according to claim 5, characterized in that: The S3 step includes the following sub-steps: S31: Construct a 1D-CNN network. The specific structure can be expressed as: Among them, x i represents the input features of the 1D-CNN network; k is represents the kernel weight; f(·) represents the RelU activation function; b s Indicates bias; O s Represents the output features of the 1D-CNN network; N represents the number of input features; S32: Model the input vector and output vector of the 1D-CNN network. The input vector can be expressed as BW APDi =stp 2i -stp 1i ; The output vector can be expressed as 7. The pulse laser ranging accuracy compensation method according to claim 1, characterized in that: In step S4, the system error is compensated, specifically including the following formula: R=γ*f(i)+R z Among them, γ represents the compensation weight, f(i) represents the measured distance value, R z Represents the true value of the measurement.

8. A pulse laser ranging accuracy compensation system, characterized in that: include: A data acquisition module is used to collect test data in the constructed test environment based on the constructed laser ranging test environment; An anomaly detection module is used to construct a pulse width anomaly detector using a deep neural network, input test data into the anomaly detector, and obtain an anomaly detection result; The pulse width compensation module is used to eliminate abnormal data and compensate the pulse width of the test data using a fitting compensation method; The error compensation module is used to compensate for the system error through calibration to complete the accuracy correction.

9. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the accuracy compensation method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the accuracy compensation method described in any one of claims 1 to 7 is implemented.

Images