Laser ranging method based on negative feedback adjustment phase type laser ranging system

Abstract

The invention discloses a laser ranging method based on a negative feedback adjustment phase type laser ranging system, and particularly relates to the technical field of laser ranging. The method comprises the following steps: acquiring to-be-predicted original measurement data by using a phase type laser ranging system, pre-processing the to-be-predicted original measurement data in the same way as that in the step S1, then loading the to-be-predicted original measurement data into a phase difference prediction neural network model for forward propagation once to obtain a phase difference value, and then calculating based on the phase difference value to obtain an estimated distance; and then, correcting the estimated distance by using a calibration error obtained by a negative feedback adjustment phase type laser ranging system to obtain a final measurement distance. The final measurement distance obtained by the method provided by the invention can have better precision.

Description

Technical Field

[0001] The invention relates to the technical field of laser ranging, and in particular to a laser ranging method based on a negative feedback regulation phase laser ranging system. Background Art

[0002] The phase-type laser ranging method is a common technology for distance measurement based on the phase difference of electromagnetic waves. Its core principle is to use the phase difference to determine the propagation time difference between the transmitted wave and the reflected wave, thereby inferring the distance between the object and the measuring device. In the prior art, the digital phase method has limited performance in processing nonlinear signals, especially in high-noise environments, and often requires additional processing steps, such as noise filtering and signal enhancement, to improve measurement accuracy. Therefore, researchers began to apply neural networks to laser ranging. However, the existing neural network-based laser ranging methods are generally not accurate when performing laser ranging in noisy environments. For this reason, the present application proposes a laser ranging method with better measurement accuracy based on a negative feedback-regulated phase-type laser ranging system. Summary of the invention

[0003] Based on the above problems, the present invention provides a laser ranging method based on a negative feedback regulation phase laser ranging system.

[0004] The present invention is achieved through the following technical solutions: A laser ranging method based on a negative feedback regulated phase laser ranging system comprises the following steps: S1. Use a negative feedback-regulated phase laser ranging system to obtain original measurement data, perform preprocessing, obtain phase difference samples, and then obtain a training set and a test set based on the phase difference samples; S2. Construct a phase difference prediction neural network for predicting phase difference; the phase difference prediction neural network includes an input layer, a first convolution layer, a first normalization layer, a first pooling layer, a second convolution layer, a second normalization layer, a first LeakyRelu layer, a second pooling layer, a third convolution layer, a third normalization layer, a second LeakyRelu layer, a first random inactivation layer, a flattening layer, a first fully connected layer, a fourth normalization layer, a third LeakyRelu layer, a second random inactivation layer, a second fully connected layer, a fifth normalization layer, a fourth LeakyRelu layer, a third random inactivation layer and an output layer connected in sequence; S3. Under the guidance of the existing mean square error loss function, the phase difference prediction neural network is trained using the training set to obtain a phase difference prediction neural network model; S4, using the phase laser ranging system to obtain the original measurement data to be predicted, and then preprocessing the original measurement data to be predicted in the same way as the preprocessing of the original measurement data in step S1, and then loading it into the phase difference prediction neural network model obtained in step S3 and forward propagating it once to obtain the phase difference value, and then calculating the estimated distance based on the phase difference value; then, using negative feedback to adjust the calibration error obtained by the phase laser ranging system to correct the estimated distance, and obtain the final measured distance.

[0005] Preferably, in step S1, the negative feedback regulated phase laser ranging system includes a control unit, the control unit is connected to the modulation unit, the modulation unit is respectively connected to the laser driving unit and the APD driving circuit; the laser driving unit is connected to the laser diode, the laser diode is respectively connected to the collimating lens and the optical fiber delay line; the APD driving circuit is respectively connected to the first APD detector, the second APD detector and the third APD detector, the optical fiber delay line is connected to the first APD detector, the present application also includes a converging lens, the converging lens is connected to the semi-transparent and semi-reflective mirror, the semi-transparent and semi-reflective mirror are respectively connected to the second APD detector and the third APD detector, the output ends of the first APD detector and the second APD detector are both connected to the signal conditioning unit, the output end of the third APD detector is connected to the signal processing unit, the output end of the signal processing unit is connected to the input end of the APD driving circuit; the output end of the signal conditioning unit is connected to the analog-to-digital conversion unit, and the analog-to-digital conversion unit is bidirectionally connected to the control unit.

[0006] Preferably, in step S2, the number of neurons in the first convolutional layer is 24, 64 or 128.

[0007] Preferably, step S3 specifically includes the following steps: inputting the one-dimensional sample data of the training set into the phase difference prediction neural network, and after forward propagating along the phase difference prediction neural network once, the phase difference prediction neural network outputs the mean square error calculated by the mean square error loss function, completing an epoch training process; then, back propagating under the guidance of the mean square error, and updating the weights of the phase difference prediction neural network, after iterating 2000 epochs, the training process of the phase difference prediction neural network can be completed; wherein, the phase difference prediction neural network parameters in the epoch training process with the smallest mean square error output during the 2000 epoch training process are used as the final phase difference prediction neural network model parameters to obtain the phase difference prediction neural network model.

[0008] Preferably, in step S4, the estimated distance is calculated based on the phase difference value by: The estimated distance is calculated, where D is the estimated distance, is the phase difference, The value is (0,2π), It is calculated by c×λ / 2, where c is the propagation speed of light and λ refers to the modulation wavelength of the modulation unit.

[0009] Preferably, in step S1, a training set and a test set are obtained based on phase difference samples, which specifically include the following steps: 3000 phase difference samples are collected, a part of the 3000 phase difference samples contain integer-cycle signals, and the remaining part contains non-integer-cycle signals, and the phase difference samples containing integer-cycle signals and the phase difference samples containing non-integer-cycle signals are randomly divided in a ratio of 8:2 to obtain an original training set and an original test set; then, 10% of the phase difference samples containing integer-cycle signals and the phase difference samples containing non-integer-cycle signals in the original training set are selected and 10% noise is added to them, and then, the remaining phase difference samples containing integer-cycle signals and the phase difference samples containing non-integer-cycle signals in the original training set are selected and 10% noise is added to them, and then, 10% of the phase difference samples of the periodic signal are selected and 20% noise is added to them. The phase difference samples without noise added and the phase difference samples with noise added in the above original training set together constitute the training set for training; then, 10% of the phase difference samples containing integer periodic signals and the phase difference samples containing non-integer periodic signals in the original test set are selected and 10% noise is added to them. Then, 10% of the remaining phase difference samples containing integer periodic signals and the phase difference samples containing non-integer periodic signals in the original test set are selected and 20% noise is added to them. The phase difference samples without noise added and the phase difference samples with noise added in the above original test set together constitute the test set for testing.

[0010] Preferably, the preprocessing method in step S4 is the same as the preprocessing method in step S1.

[0011] Compared with the prior art, the beneficial technical effects of this application are: The negative feedback regulation phase-type laser ranging system constructed in the present application utilizes the third APD detector to perform negative feedback regulation on the output voltage signal of the APD driving circuit to control the voltage signal converted from the current signal output by the first APD detector and the second APD detector to be a complete sinusoidal voltage signal, so that the negative feedback regulation phase-type laser ranging system in the present application can be applied to more test environments; moreover, since the phase information in the complete sinusoidal voltage signal is accurate, the original measurement data obtained by the negative feedback regulation phase-type laser ranging system in the present application has relatively accurate phase information, which facilitates the subsequent phase difference prediction network to more accurately predict the phase difference; In addition, the present application can also use the difference between the distance of the optical fiber delay line measured by the negative feedback adjustment phase laser ranging system and the distance of the optical fiber delay line itself as a calibration error, so as to further improve the accuracy of the measured distance obtained by the present application; in addition, the present application also constructs a phase difference prediction neural network. The use of the phase difference prediction neural network can not only effectively extract the phase information of the phase difference sample, but also effectively reduce the influence of local noise on the phase information in the feature data, and can also denoise the feature data. By testing the phase difference prediction neural network constructed by the present application, it is found that the phase difference prediction neural network model obtained by the present application tests the phase difference samples without adding noise, and the error of predicting the phase difference is small and the accuracy is high. Moreover, when the phase difference prediction neural network model obtained by the present application tests the phase difference samples with 10% noise added and the phase difference samples with 20% noise added, the error of predicting the phase difference is small and the accuracy is high. In addition, the present application also uses a phase-type laser ranging system to obtain the original measurement data to be predicted, and then pre-processes the original measurement data to be predicted, and then loads it into the phase difference prediction neural network model obtained in step S3 and propagates forward once to obtain the phase difference value; then, the estimated distance is corrected using the calibration error to obtain the final measured distance; through testing and comparison, it is found that the method described in the present application, the method for calculating the phase difference according to the discrete Fourier transform, and the full-phase Fourier error representation method for calculating the phase difference according to the full-phase Fourier transform are tested for the same test samples with a phase difference of 4π / 30 to 56π / 30; the absolute error value of the phase difference predicted by the method described in Example 1 of the present application accounts for a maximum of 2.34% of the true phase difference value, while the absolute error of the phase difference calculated by the method for calculating the phase difference according to the discrete Fourier transform accounts for a maximum of 28.51% of the true phase difference value, and the full-phase Fourier error representation method for calculating the phase difference according to the full-phase Fourier transform calculates the absolute error of the phase difference to the true phase difference value. The ratio of the maximum value is 5.65%. Obviously, the absolute error value of the phase difference obtained by the method described in the present application has a smaller ratio to the true phase difference value, which shows that the present application predicts the phase difference and uses a formula to calculate the estimated distance based on the phase difference, and then uses the calibration error to correct the estimated distance, so that the final measured distance can have better accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a schematic diagram of the structure of the negative feedback regulation phase laser ranging system; Figure 2 A schematic diagram of the network structure of the phase difference prediction neural network in this application; Figure 3 The test results of this application on the phase difference samples in the test set without adding noise; Figure 4A histogram of the absolute error and average error of the phase difference obtained by the method described in the present application, the method for calculating the phase difference according to the discrete Fourier transform, and the method for calculating the phase difference according to the full-phase Fourier transform is used to intuitively represent the histogram. DETAILED DESCRIPTION

[0013] Embodiment 1: A laser ranging method based on a negative feedback regulated phase laser ranging system comprises the following steps: S1. Use a negative feedback-regulated phase laser ranging system to obtain original measurement data, perform preprocessing, obtain phase difference samples, and then obtain a training set and a test set based on the phase difference samples; The negative feedback regulation phase laser ranging system includes a control unit, the control unit is connected to a modulation unit, the modulation unit is respectively connected to a laser driving unit and an APD driving circuit; the laser driving unit is connected to a laser diode, the laser diode is respectively connected to a collimating lens and an optical fiber delay line; the APD driving circuit is respectively connected to a first APD detector, a second APD detector and a third APD detector, the optical fiber delay line is connected to the first APD detector, the application also includes a converging lens, the converging lens is connected to a semi-transparent and semi-reflective mirror, the semi-transparent and semi-reflective mirror is respectively connected to the second APD detector and the third APD detector, the output ends of the first APD detector and the second APD detector are both connected to a signal conditioning unit, the output end of the third APD detector is connected to a signal processing unit, the output end of the signal processing unit is connected to an input end of the APD driving circuit; the output end of the signal conditioning unit is connected to an analog-to-digital conversion unit, the analog-to-digital conversion unit is bidirectionally connected to the control unit, and the control unit is bidirectionally connected to a computer.

[0014] The working principle of the phase-type laser ranging system based on negative feedback regulation in this application is: The computer is connected to the control unit via a serial port, and is used to send and receive signals to the control unit. The control unit adopts an STM controller, and is used to send a control signal to the modulation unit. The modulation unit transmits a sinusoidal voltage signal according to the received control signal. The laser driving unit receives the sinusoidal voltage signal sent by the modulation unit, and outputs a current signal for controlling the brightness change of the laser diode, so that the laser diode generates sinusoidal varying light. The sinusoidal varying light refers to the light intensity of the light emitted by the laser diode fluctuating according to the fluctuation of the sinusoidal wave. The laser diode generates sinusoidal varying light after receiving the driving signal, and outputs the sinusoidal varying light to the optical fiber delay line and the collimating lens. In the present application, the optical fiber delay line can be used as an internal optical path. Since the distance of the optical fiber delay line itself is known, when using a phase-type laser ranging system based on negative feedback regulation for measurement, the difference between the measured distance of the optical fiber delay line and the distance of the optical fiber delay line itself can be used as a calibration error. In this application, the phase difference prediction neural network model is used to obtain the phase difference value, and then based on the phase difference value, the formula The estimated distance is calculated, where D is the estimated distance, is the phase difference, The value is (0,2π), It is calculated by c×λ / 2, where c is the propagation speed of light and λ is the modulation wavelength of the modulation unit. Then, the estimated distance is corrected using the calibration error to obtain the final measured distance. After passing through the optical fiber delay line, the sinusoidal changing light is transmitted to the first APD detector; the collimating lens is used to collimate the sinusoidal changing light so that it is projected onto the target to be measured in the form of parallel light, thereby reducing the beam divergence of the sinusoidal changing light and improving the measurement accuracy; and after receiving the sinusoidal wave voltage signal sent by the modulation unit, the APD driving circuit unit performs a mixing operation and a high voltage bias drive on the sinusoidal wave voltage signal, and outputs a driving signal and a modulation signal to the first APD detector, the second APD detector and the third APD detector respectively; under the action of the driving signal output by the APD driving circuit, the first APD detector converts the sinusoidal changing light signal output by the optical fiber delay line into a current signal, mixes the current signal with the modulation signal, and then outputs the current signal to the signal conditioning unit; In the present application, after the sinusoidal variation light signal projected onto the target by the collimating lens is reflected by the target, the scattered light is converged by the converging lens to obtain converged light, and the converging lens transmits the converged light to the semi-transparent and semi-reflective mirror; the semi-transparent and semi-reflective mirror divides the converged light into two halves to obtain two halves of light, which are respectively transmitted to the second APD detector and the third APD detector, wherein the phase difference between the sinusoidal voltage signal emitted by the modulation unit and the halves of the light signal received by the second APD detector is (0,2π); the phase difference between each halves of light The light intensity is 1 / 2 of the light intensity of the converged light; under the action of the driving signal output by the APD driving circuit, the second APD detector converts the bisected light signal output by the semi-transparent and semi-reflective mirror into a current signal, mixes the current signal with the modulation signal, and then outputs the current signal to the signal conditioning unit; under the action of the driving signal output by the APD driving circuit, the third APD detector converts the bisected light signal output by the semi-transparent and semi-reflective mirror into a current signal, mixes the current signal with the modulation signal, and then outputs the current signal to the signal processing unit; The signal processing unit is used to convert the current signal output by the third APD detector into a voltage signal, and determine whether the voltage signal is in the range of 0 to 3.3V; when the voltage signal is in the range of 0 to 3.3V, the voltage signal converted from the current signal output by the third APD detector is a complete sinusoidal voltage signal, and the signal processing unit does not provide feedback to the APD drive circuit; when the voltage signal is greater than 3.3V, the voltage signal converted from the current signal output by the third APD detector is an incomplete sinusoidal voltage signal. Specifically, the sinusoidal waveform of the sinusoidal voltage signal shows a peak clipping state at the top of the sinusoidal wave. At this time, the signal processing unit provides feedback to the APD drive circuit, and the APD drive circuit reduces the drive voltage , so that the voltage signal converted from the current signal output by the first APD detector, the second APD detector and the third APD detector can be controlled to be a complete sinusoidal voltage signal; and when the voltage signal is less than 0V, the voltage signal converted from the current signal output by the third APD detector is also an incomplete sinusoidal voltage signal. Specifically, the sinusoidal waveform of the sinusoidal voltage signal shows a state of peak clipping at the bottom of the sinusoidal wave. At this time, the signal processing unit feeds back to the APD driving circuit, and the APD driving circuit increases the driving voltage, so that the voltage signal converted from the current signal output by the first APD detector, the second APD detector and the third APD detector can be controlled to be a complete sinusoidal voltage signal; The negative feedback regulation phase-type laser ranging system constructed in the present application utilizes the third APD detector to perform negative feedback regulation on the output voltage signal of the APD driving circuit to control the voltage signal converted from the current signal output by the first APD detector and the second APD detector to be a complete sinusoidal voltage signal, so that the negative feedback regulation phase-type laser ranging system in the present application can be applied to more test environments; moreover, since the phase information in the complete sinusoidal voltage signal is accurate, the original measurement data obtained by the negative feedback regulation phase-type laser ranging system in the present application has relatively accurate phase information, which facilitates the subsequent phase difference prediction network to more accurately predict the phase difference; The signal conditioning unit converts the two current signals output by the first APD detector and the second APD detector into analog voltage signals respectively, and outputs the two analog voltage signals to the analog-to-digital conversion unit; the analog-to-digital conversion unit is used to sample the two analog voltage signals output by the signal conditioning unit with 6.33 microseconds as a time unit, and the samples collected by the two analog voltage signals based on the same time unit are paired sampling samples. Every time a pair of sampling samples is obtained, the pair of sampling samples are transmitted to the control unit until all the two analog voltage signals are sampled according to the time unit; the control unit also stores the paired sampling samples output by the analog-to-digital conversion unit in two arrays, and then transmits the paired sampling sample data to the computer through the serial port, and then transmits the next pair of sampling sample data that has been saved again, and repeats the cycle until all paired sampling sample data are transmitted to the computer, and these data can be used as original measurement data; wherein, the number of bits of the above array is 1024, and when different sampling samples are stored in the array, they should be stored in chronological order.

[0015] In the present application, a negative feedback-regulated phase laser ranging system is used to obtain raw measurement data, and the raw measurement data is preprocessed, including the following steps: using the np.vstack() function to align and vertically stack the corresponding sampling sample data in the two arrays (the corresponding sampling sample data are the paired sampling sample data in the previous text) to obtain a two-dimensional matrix, which constitutes a phase difference sample.

[0016] In the present application, a training set and a test set are obtained based on phase difference samples, which specifically include the following steps: In the present application, 3000 phase difference samples are collected, a part of the 3000 phase difference samples contain integer-cycle signals, and the remaining part contains non-integer-cycle signals, and the phase difference samples containing integer-cycle signals and the phase difference samples containing non-integer-cycle signals are randomly divided in a ratio of 8:2 to obtain an original training set and an original test set; then, 10% of the phase difference samples containing integer-cycle signals and the phase difference samples containing non-integer-cycle signals in the original training set are selected and 10% noise is added to them respectively, and then, the remaining phase difference samples containing integer-cycle signals and the phase difference samples containing non-integer-cycle signals in the original training set are selected and the noise is added to them respectively. 10% of the phase difference samples of the periodic signal are selected and 20% noise is added to them. The phase difference samples without noise added and the phase difference samples with noise added in the above original training set together constitute the training set for training; then, 10% of the phase difference samples containing integer periodic signals and the phase difference samples containing non-integer periodic signals in the original test set are selected and 10% noise is added to them. Then, 10% of the remaining phase difference samples containing integer periodic signals and the phase difference samples containing non-integer periodic signals in the original test set are selected and 20% noise is added to them. The phase difference samples without noise added and the phase difference samples with noise added in the above original test set together constitute the test set for testing.

[0017] S2. Constructing a phase difference prediction neural network: In this application, the design structure of the phase difference prediction neural network is as follows: Figure 2 As shown, the phase difference prediction neural network includes an input layer, a first convolution layer, a first standardization layer, a first pooling layer, a second convolution layer, a second standardization layer, a first LeakyRelu layer, a second pooling layer, a third convolution layer, a third standardization layer, a second LeakyRelu layer, a first random inactivation layer, a flattening layer, a first fully connected layer, a fourth standardization layer, a third LeakyRelu layer, a second random inactivation layer, a second fully connected layer, a fifth standardization layer, a fourth LeakyRelu layer, a third random inactivation layer and an output layer, which are connected in sequence; wherein, the number of neurons in the first convolution layer is 24, and the output layer is set as a fully connected layer.

[0018] The working principle of the phase difference prediction neural network in this application is as follows: In the present application, the input layer is used to input one-dimensional sample data in the training set, one-dimensional sample data in the test set, or one-dimensional sample data for prediction; the first convolution layer is used to perform a convolution operation on the phase difference samples output by the input layer, extract the phase information of the phase difference samples, and obtain feature data with primary features; the first normalization layer performs a normalization operation on the feature data to prevent overfitting of the phase difference prediction neural network during the training process; the first pooling layer downsamples the feature data output by the first normalization layer to reduce the influence of local noise on the phase information in the feature data; the second convolution layer performs a convolution operation on the feature data output by the first pooling layer to further extract the phase information in the feature data The second normalization layer performs normalization on the feature data output by the second convolution layer to further prevent the phase difference prediction neural network from overfitting during the training process; the first LeakyRelu layer denoises the feature data; the second pooling layer downsamples the feature data to effectively reduce the amount of subsequent calculations; the third convolution layer performs convolution on the feature data to further extract phase information to obtain feature data with high-level features; the third normalization layer performs normalization on the feature data to further prevent the phase difference prediction neural network from overfitting during the training process; the second LeakyRelu layer performs The first random dropout layer randomly discards some neurons during the feature data processing to prevent some large noise from affecting the phase information. The flattening layer is used to convert the feature data into a one-dimensional vector value to obtain one-dimensional feature data with rich phase information. The first fully connected layer performs dimensionality reduction on the one-dimensional feature data. The fourth normalization layer is used to normalize the feature data output by the first fully connected layer to further prevent the phase difference prediction neural network from overfitting during the training process and improve the stability of the feature data. The third LeakyRelu layer further denoises the feature data. The second random dropout layer randomly discards some neurons during the feature data processing to prevent some large noise from affecting the phase information. The flattening layer is used to convert the feature data into a one-dimensional vector value to obtain one-dimensional feature data with rich phase information. The first fully connected layer performs dimensionality reduction on the one-dimensional feature data. The fourth normalization layer is used to normalize the feature data output by the first fully connected layer to further prevent the phase difference prediction neural network from overfitting during the training process and improve the stability of the feature data. The first layer discards some neurons to further avoid the influence of some larger noise on the phase information; the second fully connected layer further reduces the dimension of the feature data; the fifth normalization layer performs normalization on the feature data to further prevent the overfitting of the phase difference prediction neural network during the training process and improve the stability of the feature data; the fourth LeakyRelu layer further denoises the feature data to make the feature data more convergent; the third random inactivation layer randomly discards some neurons during the feature data processing to further avoid the influence of some larger noise on the phase information; the output layer performs feature space reduction on the feature data and outputs the phase difference and mean square error.

[0019] In this application, the first convolutional layer has 24 neurons and a convolution kernel size of 5×5, the second convolutional layer has 128 neurons and a convolution kernel size of 5×5, and the third convolutional layer has 256 neurons and a convolution kernel size of 3×3.

[0020] S3. Under the guidance of the existing mean square error loss function, the phase difference prediction neural network is trained using the training set to obtain a phase difference prediction neural network model, which specifically includes the following steps: The one-dimensional sample data of the training set is input into the phase difference prediction neural network. After forward propagation along the phase difference prediction neural network once, the phase difference prediction neural network outputs the mean square error calculated by the mean square error loss function to complete an epoch training process; then, back propagation is performed under the guidance of the mean square error, and the weight of the phase difference prediction neural network is updated. After iterating 2000 epochs, the training process of the phase difference prediction neural network can be completed; wherein, the phase difference prediction neural network parameters in the epoch training process with the smallest mean square error output during the 2000 epoch training process are used as the final phase difference prediction neural network model parameters to obtain the phase difference prediction neural network model; wherein, the training process of the phase difference prediction neural network is based on the TensorFlow development platform, and, in the training process of the phase difference prediction neural network, the batch size is set to 32 and the initial learning rate is set to 0.0001.

[0021] S4, using the phase laser ranging system to obtain the original measurement data to be predicted, and then preprocessing the original measurement data to be predicted, the preprocessing method is the same as the preprocessing method of the original measurement data in step S1, and then loading it into the phase difference prediction neural network model obtained in step S3, forward propagating once, and then the phase difference value can be obtained, and the formula is used The estimated distance is calculated, where D is the estimated distance, is the phase difference, The value is (0,2π), It is calculated by c×λ / 2, where c is the propagation speed of light and λ refers to the modulation wavelength of the modulation unit; then, the estimated distance is corrected using the calibration error to obtain the final measured distance.

[0022] Embodiment 2: The difference between the embodiment and the first embodiment is that: the number of neurons in the first convolutional layer in the second embodiment is 12; Embodiment three: The difference between the third embodiment and the first embodiment is that: the number of neurons in the first convolutional layer in the third embodiment is 32; Embodiment 4: The difference between the fourth embodiment and the first embodiment is that: the number of neurons in the first convolutional layer in the fourth embodiment is 64; Embodiment five: The difference between the fifth embodiment and the first embodiment is that: the number of neurons in the first convolutional layer in the fifth embodiment is 128; Embodiment six: The difference between the sixth embodiment and the first embodiment is that the number of neurons in the first convolutional layer in the sixth embodiment is 256.

[0023] test: In order to test the effect of the phase difference prediction neural network model in the present application on predicting the phase difference, the present application adopts a test set for testing. It is found through testing that the phase difference samples without adding noise in the test set are tested, and the test results are as follows: Figure 3 As shown. Figure 3 It can be seen that when Examples 1 to 6 of the present application test the phase difference samples without adding noise in the test set, the CORR value obtained by the test is greater than or equal to 0.9999, which indicates that the phase difference prediction values ​​obtained by testing the phase difference samples without adding noise in the test set in Examples 1 to 6 of the present application are very close to the true value of the phase difference; wherein, the phase difference samples without adding noise refer to the phase difference samples in the test set except the phase difference samples with 10% noise added and the phase difference samples with 20% noise added.

[0024] In order to explore the effect of the phase difference prediction neural network model of the present application on predicting the phase difference of the noisy phase difference samples, the present application specifically tests the phase difference samples with 10% noise added and the phase difference samples with 20% noise added in the test set, and the test results are shown in Table 1 and Table 2; Table 1 10% noise evaluation index

[0025] In Table 1, MSE (Mean Squared Error) represents mean square error, RMSE (Root Mean Squared Error) represents root mean square error, MAE (Mean Absolute Error) represents mean absolute error, MRE (Mean Relative Error) represents mean relative error, and CORR (Correlation Coefficient) represents correlation coefficient. The smaller the error values ​​of mean square error, root mean square error, mean absolute error, and mean relative error, the more accurate the phase difference predicted by the phase difference prediction neural network model in the embodiment is, and the larger the correlation coefficient is, the closer the predicted phase difference value is to the true phase difference value.

[0026] Table 2 20% noise evaluation index

[0027] From the test results shown in Table 1 and Table 2, it can be seen that when the first convolution layer in the phase difference prediction neural network uses 24 neurons, that is, the phase difference prediction neural network model described in Example 1, when testing phase difference samples with 10% noise and phase difference samples with 20% noise, from the perspective of the mean square error, root mean square error, mean absolute error, mean relative error and CORR index test results, the phase difference prediction neural network model described in Example 1 can achieve better prediction effect, that is, the accuracy of predicting the phase difference is better.

[0028] In addition, the present application also uses a phase laser ranging system to obtain the original measurement data to be predicted, and then pre-processes the original measurement data to be predicted in the same way as the pre-processing of the original measurement data in step S1, and then loads it into the phase difference prediction neural network model obtained in step S3 and propagates forward once to obtain the phase difference value; then, the estimated distance is corrected using the calibration error to obtain the final measured distance; Since the embodiments of the present application are carried out at room temperature (25°C), the calibration error of the negative feedback-regulated phase-type laser ranging system constructed in the present application is small. Therefore, the present application focuses on the effect of predicting the phase difference value; in order to more intuitively show the advantages and disadvantages of the method described in the present application compared with the traditional spectral analysis method (for example: the method of calculating the phase difference based on discrete Fourier transform and the method of calculating the phase difference based on full-phase Fourier transform) in obtaining the absolute error of the phase difference (the absolute error refers to the absolute value of the difference between the predicted phase difference and the actual phase difference) and the average error (the average error refers to the average value of the absolute error of the phase difference), the present application uses a histogram to intuitively represent the absolute error and the average error of the phase difference, such as Figure 4 shown.

[0029] Figure 4 In the embodiment, the average error of embodiment one represents the average error of the phase difference predicted by embodiment one; the discrete Fourier transform average error represents the average error of the phase difference calculated according to the method of calculating the phase difference by discrete Fourier transform; the full-phase Fourier average error represents the average error of the phase difference calculated according to the method of calculating the phase difference by full-phase Fourier transform; the error of embodiment one represents the absolute error of the phase difference predicted by embodiment one; the discrete Fourier transform error represents the absolute error of the phase difference calculated according to the method of calculating the phase difference by discrete Fourier transform; the full-phase Fourier error represents the absolute error of the phase difference calculated according to the method of calculating the phase difference by full-phase Fourier transform.

[0030] Figure 4In the figure, the horizontal axis is the test sample, indicating that the test sample is divided into 30 parts according to the phase difference (0, 2π) in turn. Moreover, the present application uses a phase laser ranging system based on negative feedback regulation for measurement. During the measurement, the phase difference between the sinusoidal voltage signal emitted by the modulation unit and the bisected light signal received by the second APD detector is (0, 2π). Therefore, the original measurement data obtained by the present application does not include the case where the phase difference is 0, nor does it include the case where the phase difference is 2π. Therefore, Figure 4 The test results shown in the figure are obtained by testing the test samples with phase differences of 2π / 30, 4π / 30, 6π / 30...56π / 30, 58π / 30 respectively; for example, Figure 4 The horizontal coordinate 2π / 30 in the middle represents the result obtained by testing the test sample with a phase difference of 2π / 30; Figure 4 The test data obtained by testing the test samples with phase differences ranging from 2π / 30 to 58π / 30 are shown in Tables 3 and 4.

[0031] In Table 3 and Table 4, the DFT method refers to a method for calculating the phase difference based on the discrete Fourier transform, and the ApFFT method refers to a method for calculating the phase difference based on the full phase Fourier transform; taking the ratio of the absolute error value of the phase difference predicted by the method described in Example 1 to the actual phase difference value as 2.91% as an example, the calculation method is introduced as follows: the absolute value of the difference between the phase difference value 0.2033 predicted by the method described in Example 1 and the actual phase difference value 2π / 30 is calculated to obtain the absolute error value of the phase difference predicted by the method described in Example 1 as 0.0061, and then, the ratio of the absolute error value 0.0061 of the phase difference predicted by the method described in Example 1 to the actual phase difference value 2π / 30 is calculated and multiplied by 100% to obtain 2.91%; Table 3 shows Figure 4 The test data shown in the test samples with phase differences ranging from 2π / 30 to 34π / 30 are obtained by testing

[0032] Table 4 shows Figure 4 The test data shown in the test samples with phase differences ranging from 36π / 30 to 58π / 30 are obtained by testing

[0033] In the prior art, when actually measuring, for example, measuring the liquid level of a ladle or the liquid level of water, there is usually a relatively accurate measurement range. Specifically in the present application, the test results obtained by testing the test samples with phase differences of 2π / 30 and 58π / 30 in Example 1 of the present application are not accurate enough, while the test results obtained by testing the test samples with phase differences of 4π / 30 to 56π / 30 are better. That is to say, in the actual measurement of the present application, the phase difference is obtained based on the test samples with phase differences of 4π / 30 to 56π / 30; To this end, the present application focuses on comparing the test results of the method described in the present application, the method for calculating the phase difference according to the discrete Fourier transform, and the method for calculating the phase difference according to the full phase Fourier transform for the same test samples with a phase difference of 4π / 30 to 56π / 30; from Tables 3 and 4, when the present application predicts the phase difference based on the test samples with a phase difference of 4π / 30 to 56π / 30, the absolute error value of the phase difference predicted by the method described in Example 1 of the present application accounts for a maximum of 2.34% of the true phase difference value, while the absolute error of the phase difference calculated by the method for calculating the phase difference according to the discrete Fourier transform accounts for a maximum of 28.51% of the true phase difference value, and the full phase Fourier error represents that the absolute error of the phase difference calculated by the method for calculating the phase difference according to the full phase Fourier transform accounts for a maximum of 5.65% of the true phase difference value. Obviously, the absolute error value of the phase difference obtained by the method described in the present application has a smaller ratio to the true phase difference value, which shows that the present application predicts the phase difference and uses a formula to calculate the estimated distance based on the phase difference, and then uses the calibration error to correct the estimated distance, so that the final measured distance can have better accuracy.

Claims

1. A laser ranging method based on a negative feedback regulated phase laser ranging system, characterized in that: The following steps are involved: S1. Obtaining and preprocessing original measurement data using a phase laser ranging system to obtain phase difference samples, and obtaining a training set and a test set based on the phase difference samples; S2. Construct a phase difference prediction neural network for predicting phase difference; including an input layer, a first convolution layer, a first normalization layer, a first pooling layer, a second convolution layer, a second normalization layer, a first LeakyRelu layer, a second pooling layer, a third convolution layer, a third normalization layer, a second LeakyRelu layer, a first random inactivation layer, a flattening layer, a first fully connected layer, a fourth normalization layer, a third LeakyRelu layer, a second random inactivation layer, a second fully connected layer, a fifth normalization layer, a fourth LeakyRelu layer, a third random inactivation layer and an output layer connected in sequence; S3. Under the guidance of the existing mean square error loss function, the phase difference prediction neural network is trained using the training set to obtain a phase difference prediction neural network model; S4. Use the phase-type laser ranging system to obtain the original measurement data to be predicted and pre-process it, load it into the phase difference prediction neural network model, forward propagate it once, and obtain the phase difference value. The estimated distance is calculated based on the phase difference value; use negative feedback to adjust the calibration error obtained by the phase-type laser ranging system to correct the estimated distance and obtain the final measured distance.

2. The laser ranging method based on the negative feedback regulation phase laser ranging system according to claim 1, characterized in that: In step S1, the negative feedback regulated phase laser ranging system includes a control unit, the control unit is connected to the modulation unit, the modulation unit is respectively connected to the laser driving unit and the APD driving circuit; the laser driving unit is connected to the laser diode, the laser diode is respectively connected to the collimating lens and the optical fiber delay line; the APD driving circuit is respectively connected to the first APD detector, the second APD detector and the third APD detector, the optical fiber delay line is connected to the first APD detector, the present application also includes a converging lens, the converging lens is connected to the semi-transparent and semi-reflective mirror, the semi-transparent and semi-reflective mirror are respectively connected to the second APD detector and the third APD detector, the output ends of the first APD detector and the second APD detector are both connected to the signal conditioning unit, the output end of the third APD detector is connected to the signal processing unit, the output end of the signal processing unit is connected to the input end of the APD driving circuit; the output end of the signal conditioning unit is connected to the analog-to-digital conversion unit, and the analog-to-digital conversion unit is bidirectionally connected to the control unit.

3. The laser ranging method based on the negative feedback regulated phase laser ranging system according to claim 1, characterized in that: In step S2, the number of neurons in the first convolutional layer is 24, 64 or 128.

4. The laser ranging method based on the negative feedback regulated phase laser ranging system according to claim 1, characterized in that: Step S3 specifically includes the following steps: inputting the one-dimensional sample data of the training set into the phase difference prediction neural network, and after forward propagating along the phase difference prediction neural network once, the phase difference prediction neural network outputs the mean square error calculated by the mean square error loss function, completing an epoch training process; then, back propagating under the guidance of the mean square error, and updating the weights of the phase difference prediction neural network, after iterating 2000 epochs, the training process of the phase difference prediction neural network can be completed; wherein, the phase difference prediction neural network parameters in the epoch training process with the smallest mean square error output during the 2000 epoch training process are used as the final phase difference prediction neural network model parameters to obtain the phase difference prediction neural network model.

5. The laser ranging method based on the negative feedback regulated phase laser ranging system according to claim 1, characterized in that: In step S4, the estimated distance is calculated based on the phase difference value by: The estimated distance is calculated, where D is the estimated distance, is the phase difference, It is calculated by c×λ / 2, where c is the propagation speed of light and λ refers to the modulation wavelength of the modulation unit.

6. The laser ranging method based on the negative feedback regulated phase laser ranging system according to claim 1, characterized in that: In step S1, a training set and a test set are obtained based on the phase difference samples, which specifically include the following steps: 3000 phase difference samples are collected, a part of the 3000 phase difference samples contain integer-cycle signals, and the remaining part contain non-integer-cycle signals, and the phase difference samples containing integer-cycle signals and the phase difference samples containing non-integer-cycle signals are randomly divided in a ratio of 8:2 to obtain an original training set and an original test set; then, 10% of the phase difference samples containing integer-cycle signals and the phase difference samples containing non-integer-cycle signals in the original training set are selected and 10% noise is added to them respectively, and then, the remaining phase difference samples containing integer-cycle signals and the phase difference samples containing non-integer-cycle signals in the original training set are selected and 10% noise is added to them respectively. 10% of the phase difference samples of the signal are selected and 20% noise is added to them respectively. The phase difference samples without noise added and the phase difference samples after noise added in the above original training set together constitute the training set for training; then, 10% of the phase difference samples containing integer periodic signals and the phase difference samples containing non-integer periodic signals in the original test set are selected and 10% noise is added to them respectively. Then, 10% of the remaining phase difference samples containing integer periodic signals and the phase difference samples containing non-integer periodic signals in the original test set are selected and 20% noise is added to them respectively. The phase difference samples without noise added and the phase difference samples after noise added in the above original test set together constitute the test set for testing.

7. The laser ranging method based on the negative feedback regulated phase laser ranging system according to claim 1, characterized in that: The preprocessing method in step S4 is the same as the preprocessing method in step S1.

Images