Collision object identification method, storage medium, controller and vehicle

Abstract

The invention relates to a collision object identification method, a storage medium, a controller and a vehicle, and the method comprises the steps: dividing a collected elastic wave data stream into a plurality of signal sections according to a sliding window, extracting the time domain features and frequency domain features of each signal section, splicing the time domain features and frequency domain features of the plurality of signal sections, and carrying out the recognition of a collision object through the splicing of the time domain features and frequency domain features of the plurality of signal sections; the method comprises the following steps: obtaining a time-varying feature of frequency domain and time domain features, namely a time sequence fusion feature vector, finally inputting the time sequence fusion feature vector into a learning model, and outputting a classification probability of a corresponding material. Real touch signals and interference signals can be effectively distinguished through time domain features, even if the frequency domains of the real touch signals and the interference signals are similar, the misjudgment rate can be reduced through dynamic characteristic differences, more comprehensive signal characterization can be provided through deep fusion of the time domain features and frequency domain features, and the signals can be analyzed in a segmented mode through sliding window segmentation processing without waiting for complete data streams, so that the accuracy of signal processing is improved. And the processing delay is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of elastic wave identification, and in particular to a collision object identification method, a storage medium, a controller and a vehicle. BACKGROUND

[0002] In the rapid development of current intelligent sensing and material identification technology, touch material identification methods based on elastic wave signals are increasingly becoming a research hotspot. This technology uses the elastic wave signals generated by objects when they are touched, and identifies different materials by analyzing the unique characteristics of these signals. However, most current methods mainly focus on extracting features in the frequency domain, such as frequency distribution, fundamental frequency components, etc., which indeed provide important basis for identification.

[0003] However, when there are interference sources in the working environment of the elastic wave sensor, such as continuous vibration of the contact surface of the elastic wave sensor itself, coupling effects of environmental noise, etc., these non-touch signals may also produce similar feature patterns in the frequency domain as the specified material. The existence of such "pseudo-features" greatly increases the difficulty of subsequent identification algorithms, leading to a rise in misidentification rate. SUMMARY

[0004] The collision object identification method provided by the embodiments of the present application can effectively distinguish touch signals and non-touch signals, and reduce the misidentification rate, at least partially solving the above technical problems.

[0005] To achieve the above purpose, according to the first aspect of the present application, a collision object identification method is provided, comprising:

[0006] forming a data stream based on the elastic wave signals generated by touching;

[0007] dividing the data stream into multiple signal sections;

[0008] extracting time domain features and frequency domain features for each signal section respectively;

[0009] concatenating the time domain features and the frequency domain features of the multiple signal sections to obtain a time sequence fusion feature vector;

[0010] outputting the material classification of the touched material based on the time sequence fusion feature vector.

[0011] Optionally, the step of forming a data stream based on the elastic wave signals generated by touching comprises:

[0012] monitoring the peak-to-peak value of the signal peak in real time based on the elastic wave signals through the elastic wave sensor;

[0013] when the peak-to-peak value of the signal peak exceeds a preset peak value, merging the pre-trigger cache data and the post-trigger fixed length data into a data stream.

[0014] Optionally, the preset peak value is no more than 300 mV.

[0015] Optionally, when the peak-peak value of the signal peak exceeds the preset peak value, the pre-trigger buffered data and the post-trigger fixed-length data are merged into a data stream, and the method further comprises: when the peak-peak value of the signal peak exceeds the preset peak value, the waveform rise time of the elastic wave signal is less than 10 ms, and the waveform fall time of the elastic wave signal is less than 20 ms, the pre-trigger buffered data and the post-trigger fixed-length data are merged into a data stream.

[0016] Optionally, the step of dividing the data stream into a plurality of signal sections comprises: dividing the data stream into a plurality of signal sections according to sliding windows.

[0017] Optionally, the step of dividing the data stream into a plurality of signal sections according to sliding windows comprises:

[0018] determining the length of the sliding window according to the signal sampling rate and the number of data points, wherein the lengths of the plurality of sliding windows are equal;

[0019] dividing the data stream into a plurality of signal sections without overlap according to the length of the sliding window.

[0020] Optionally, the length of the sliding window satisfies the following formula:

[0021] L = M / F s ;

[0022] wherein L is the length of the sliding window, M is the number of data points used in fast Fourier transform, and F s is the signal sampling rate.

[0023] Optionally, the time domain features include at least one of the root mean square value, the peak-peak value, the variance, the rectified mean value, the root amplitude, the waveform skewness, and the kurtosis of the signal.

[0024] Optionally, the step of extracting the frequency domain features comprises:

[0025] calculating a frequency spectrum;

[0026] dividing the frequency spectrum into a low frequency band, a medium frequency band, and a high frequency band;

[0027] obtaining at least one feature of the peak frequency, the centroid frequency, the frequency spectrum standard deviation, the root mean square frequency, the frequency variance, and the frequency spectrum entropy of each frequency band.

[0028] Optionally, the step of splicing the time domain features and the frequency domain features of the plurality of signal sections to obtain a time sequence fusion feature vector comprises:

[0029] normalizing the time domain features and the frequency domain features;

[0030] concatenating the normalized features in time sequence into a time sequence fusion feature vector;

[0031] performing principal component analysis on the time sequence fusion feature vector, and reducing dimension of the time sequence fusion feature vector to a preset dimension.

[0032] Optionally, the step of outputting the material classification of the touched material based on the time sequence fusion feature vector comprises:

[0033] determining a learning model for classification;

[0034] inputting the time sequence fusion feature vector into the learning model, and calculating a probability distribution of each material category;

[0035] determining a final material type according to a maximum probability.

[0036] According to a second aspect of the present application, a computer readable storage medium is provided, which stores a computer program, and the computer program is executed by a processor to implement the steps of the method described above.

[0037] According to a third aspect of the present application, a controller is provided, which stores a computer program, and the computer program is executed by a processor to implement the steps of the method described above.

[0038] According to a fourth aspect of the present application, a vehicle is provided, which comprises the controller described above.

[0039] According to a fifth aspect of the present application, a computer program product is provided, which comprises a computer program or instructions, and the computer program or instructions are executed by a processor to implement the steps of the method described above.

[0040] In summary, in the embodiments of the present application, the collected elastic wave data stream is divided into multiple signal sections, and the time domain features and the frequency domain features are extracted for each signal section. The time domain features and the frequency domain features of the multiple signal sections are concatenated to obtain a feature that changes with time, i.e., a time sequence fusion feature vector. Finally, the classification probability of the corresponding material is output according to the time sequence fusion feature vector. The time domain features can effectively distinguish between real touch signals and interference signals. Even if the frequency domains of the two are similar, the misjudgment rate can be reduced through the difference in dynamic characteristics. The deep fusion of time and frequency domain features provides more comprehensive signal representation, such as the joint discrimination of high-frequency energy and sharp waveform of metal knocking, which is superior to single frequency domain method. And through the segmented processing, the signal can be analyzed segmentally, without waiting for the complete data stream, which is beneficial to reduce the processing delay.

[0041] Other features and advantages of the present application will be described in the following detailed description of the embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some of the embodiments of the present application, and all other drawings obtained by those skilled in the art without creative effort based on these drawings are within the protection scope of the present application.

[0043] In order to more completely understand the present application and its beneficial effects, the following will be described in conjunction with the drawings, wherein the same reference numerals in the following description represent the same parts.

[0044] Figure 1 is a step flow chart of a collision object recognition method provided in an exemplary embodiment of the present disclosure;

[0045] Figure 2 is a flow chart of a collision object recognition method provided in an exemplary embodiment of the present disclosure;

[0046] Figure 3 is a frequency spectrum diagram of the combination of time domain features and frequency domain features in a collision object recognition method provided in an exemplary embodiment of the present disclosure;

[0047] Figure 4 is a frequency spectrum diagram of frequency domain features in a collision object recognition method provided in an exemplary embodiment of the present disclosure;

[0048] Figure 5 is an architecture diagram of a controller provided in an exemplary embodiment of the present disclosure;

[0049] Figure 6 is an architecture diagram of a vehicle provided in an exemplary embodiment of the present disclosure. DETAILED DESCRIPTION

[0050] The technical solutions in the embodiments of the present application will be described clearly and completely in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the present application.

[0051] Based on the problems mentioned in the foregoing background art, in the related art, when there is an interference source in the working environment of the elastic wave sensor, such as continuous vibration of the elastic wave sensor contact surface itself, coupling effect of environmental noise, etc., these non-touch signals may also produce similar characteristic patterns in the frequency domain as the specified material, resulting in an increase in misidentification rate.

[0052] To this end, the present application provides a collision object identification method, please refer to Figure 1 and Figure 2 The collision object identification method provided by the embodiments of the present application comprises:

[0053] S1, forming a data stream based on the elastic wave signal generated by the touch;

[0054] S2, dividing the data stream into multiple signal sections;

[0055] S3, extracting time domain features and frequency domain features for each signal section, respectively;

[0056] S4, splicing the time domain features and the frequency domain features of the multiple signal sections to obtain a time sequence fusion feature vector;

[0057] S5, outputting the material classification of the touched material based on the time sequence fusion feature vector.

[0058] In the technical solutions provided by the present application, the collected elastic wave data stream is divided into multiple signal sections according to the sliding window, and the time domain features and the frequency domain features are extracted for each signal section, respectively. The time domain features and the frequency domain features of the multiple signal sections are spliced to obtain a feature that changes with time, i.e. a time sequence fusion feature vector. Finally, the time sequence fusion feature vector is input into a learning model to output the classification probability of the corresponding material. Through the time domain features, the real touch signal and the interference signal can be effectively distinguished. Even if the frequency domains of the two are similar, the misjudgment rate can also be reduced through the difference in dynamic characteristics. The deep fusion of time and frequency domain features provides more comprehensive signal representation, such as the joint discrimination of high-frequency energy and sharp waveform of metal knocking, which is better than the single frequency domain method. And through the sliding window segmentation processing, the signal can be analyzed segmentally, without the need to wait for the complete data stream, which is conducive to reducing the processing delay.

[0059] In some embodiments, the step of collecting the elastic wave signal generated by the touch to form a data stream comprises:

[0060] S11, collecting the elastic wave signal through the elastic wave sensor.

[0061] The elastic wave sensor can be a piezoelectric sensor such as a PVDF film or a capacitive acceleration sensor such as a MEMS accelerometer, installed on the contact surface of the object to be detected, such as a vehicle door or an industrial part. The sensor sampling rate Fs is set to 10 kHz to 50 kHz, covering the typical frequency range of the elastic wave signal 0-20 kHz. The sensor output signal is amplified by a preamplifier and passes through a band-pass filter to remove DC offset and high-frequency noise.

[0062] S12, real-time monitoring of the peak-to-peak value of the signal peak.

[0063] Exemplarily, the peak-to-peak value of the signal peak is calculated in real time with a sliding window, and the trigger peak-to-peak value is set to 300 mV, which can be adjusted to 200 mV-500 mV according to the application scenario, and only when the signal strength is sufficient (greater than or equal to 300 mV) is the processing triggered, avoiding invalid calculation.

[0064] S13, triggering signal buffering when the peak-to-peak value of the signal peak exceeds the preset peak value.

[0065] Exemplarily, the system pre-allocates a ring buffer to continuously store the sensor data stream. When the peak-to-peak value is greater than or equal to 300 mV, the trigger time T is recorded, and the data from 200 ms before T to 300 ms after T is extracted, with a total length of 500 ms. If multiple triggers occur in succession, such as 2 times within 10 ms, they are combined as a single event to avoid repeated processing. By retaining the complete signal before and after the trigger, including the touch initiation and decay process, information loss can be avoided, and only the valid signal segment is processed, reducing memory and computing power consumption.

[0066] S14, merging the pre-trigger buffered data and the fixed-length post-trigger data into a data stream.

[0067] Exemplarily, the pre-trigger buffered data and the post-trigger data are spliced in chronological order to form a 500 ms continuous data stream. Through timestamp calibration, the data before and after the trigger is seamlessly connected, avoiding phase jumps, and can cover the complete life cycle of the touch signal, such as the pre-trigger silent state and the post-trigger vibration decay. And by using a fixed-length data stream, it is beneficial for subsequent standardized processing and model inference.

[0068] Further, the preset peak value is not more than 300 mV. In other embodiments, the preset peak value can be determined according to actual conditions. The preset peak value mainly serves as a calculation threshold to avoid excessive invalid calculations.

[0069] Further, the condition for triggering signal buffering also includes that the waveform rise time of the elastic wave signal is less than 10 ms and the fall time is less than 20 ms.

[0070] In some embodiments, the step of dividing the data stream into multiple signal sections in a sliding window comprises:

[0071] S21, determining the length of the sliding window according to the signal sampling rate and the number of data points, and the lengths of the multiple sliding windows are equal.

[0072] The signal sampling rate refers to the number of signal data collected per second, and the unit is hertz (Hz). The number of data points refers to the number of data points used in Fast Fourier Transform (FFT), such as 1024-point FFT, which means the signal is divided into 1024 frequency components. The signal sampling rate and the number of data points jointly determine the length of the sliding window, and the length of the sliding window is calculated as:

[0073] L = M / F s ;

[0074] Where L is the length of the sliding window, M is the number of data points used in Fast Fourier Transform, and F s is the signal sampling rate.

[0075] For example, the sampling rate F s = 10 kHz, the number of data points M = 1024, and the length of the sliding window L = 1024 / 10000 ≈ 0.1 seconds = 100 ms.

[0076] For example, the number of data points is selected according to the signal frequency domain resolution requirement, and the typical value is 2048 or 4096. In a low-frequency scenario, such as industrial vibration monitoring, M = 2048 (frequency resolution Δf ≈ 9.77 Hz, signal sampling rate Fs = 20 kHz); in a high-frequency scenario, such as metal knocking detection, M = 4096 (Δf ≈ 4.88 Hz, Fs = 20 kHz). In an embedded device (such as STM32H7), the M value is dynamically selected according to the memory and computing power (such as reduced to 1024 when the computing power is insufficient). The signal sampling rate F s needs to meet the Nyquist theorem. For example, in the automotive scenario, F s = 10 kHz to cover the 0-5 kHz frequency band; in the industrial scenario, F s = 50 kHz to cover the 0-25 kHz high-frequency impact signal. Through the coordinated adjustment of the number of data points and the signal sampling rate, the frequency domain resolution and real-time performance are balanced, the number of data points is reduced in low-power devices to reduce the amount of calculation. In addition, by ensuring that the length of the sliding window is an integer multiple of the sampling period, it is beneficial to avoid truncation error.

[0077] S22, according to the length of the sliding window, the data stream is divided into multiple signal sections in a non-overlapping manner.

[0078] Exemplarily, the continuous data stream is divided into multiple signal sections according to the length L of the sliding window, and adjacent sections are not overlapped. If L=100 ms, the data stream is divided in the order of 0-100 ms, 100 ms-200 ms, 200 ms-300 ms, and so on. As soon as the data of the length of a sliding window is filled, the feature extraction and classification process is triggered. By making multiple signal sections not overlapped, sectioning reduces data redundancy, improves real-time performance, and reduces processing delay. In addition, since the signal sections are not overlapped, repeated calculation of the same data can be avoided, which is beneficial to saving computing power.

[0079] In some embodiments, referring to Figure 3 , the time domain features include at least one of the root mean square value, the peak-to-peak value, the variance, the rectified average value, the root amplitude, the waveform skewness, and the kurtosis of the signal. By adding the extraction of the time domain features in the elastic wave identification method, the time domain features can be used to analyze the law of the change of the signal with time, thereby eliminating part of the signal interference. For example, the contact surface vibration caused by the blowing of the wind or environmental noise, etc. This kind of non-touch signal may produce a similar feature pattern in the frequency domain as the specified material, which may lead to the misidentification of this kind of "pseudo feature" as the specified material by using only the frequency domain feature analysis. By using the time domain features in combination with the frequency domain features, such misidentification can be excluded, because the non-touch signal caused by environmental noise, etc. is generally continuous, while the generation of the touch signal will form a mutation in the waveform, thereby easily identifying the "pseudo feature".

[0080] The root mean square value, the peak-to-peak value, the variance, the rectified average value, the root amplitude, the waveform skewness, and the kurtosis of the signal will be introduced below:

[0081] The calculation formula of the root mean square value is:

[0082] Exemplarily, the square values of the signal are accumulated point by point in the sliding window, and the square root is calculated at the end of the window, which is used to represent the average energy intensity of the signal and filter low-energy noise such as the signal of the wind blowing sensor.

[0083] The calculation formula of the peak-to-peak value is: pp = max(x i )-min(x i ), i=1, 2, 3,..., N.

[0084] Exemplarily, the maximum value and the minimum value are dynamically updated in the sliding window, and the difference is calculated at the end of the window, which is used to judge the dynamic range of the signal and trigger the effective touch event, such as the peak-to-peak value≥300 mV.

[0085] The calculation formula of the variance is:

[0086] Exemplarily, the mean value is calculated first in the sliding window The square deviation is accumulated again for quantifying the signal fluctuation, distinguishing stationary noise like low variance from dynamic touch signal like high variance.

[0087] The calculation formula of the rectified mean value is as follows:

[0088] Exemplarily, the absolute value of the signal is accumulated in the sliding window, and the average is taken at the end of the window, which is used to represent the average oscillation amplitude of the signal and is sensitive to low-amplitude noise, such as the signal of a wind-blow sensor.

[0089] The calculation formula of the square root amplitude is as follows:

[0090] Exemplarily, the average value of the square root of the signal is calculated first in the sliding window, and then the output is squared, which is used to enhance the sensitivity to transient changes of the signal, such as weak vibrations in the initial stage of knocking.

[0091] The calculation formula of the waveform skewness is as follows:

[0092] It is used to describe the symmetry of the waveform. For example, the waveform of a metal knocking signal is right-skewed, and the waveform of a plastic knocking signal is symmetrical.

[0093] The calculation formula of the kurtosis is as follows:

[0094] It is used to quantify the sharpness of the waveform. The kurtosis of a hard material (such as metal) knocking signal is high (sharp peak), and the kurtosis of a soft material (such as rubber) is low (flat).

[0095] It should be noted that in all the formulas involved in the above time domain features, N represents the length of the signal, x i represents the amplitude of the signal at time i.

[0096] It should also be noted that for the various features included in the above time domain features, in actual application, part of the features can be selected as time domain feature analysis, or all the features can be selected as time domain feature analysis, which can be selected according to actual situation.

[0097] In some embodiments, referring to Figure 4 , the extraction step of the frequency domain feature includes:

[0098] calculating the frequency spectrum;

[0099] dividing the low frequency band, the medium frequency band and the high frequency band according to the frequency spectrum;

[0100] obtaining at least one feature of the peak frequency, the centroid frequency, the frequency spectrum standard deviation, the root mean square frequency, the frequency variance and the frequency spectrum entropy of each frequency band.

[0101] Exemplarily, a Hanning window is applied to each signal section, and then a data point number calculation is performed to obtain a complex spectrum, calculate an amplitude spectrum and a power spectrum density. The Hanning window reduces the truncation effect and is beneficial to reduce the risk of spectrum leakage. A low frequency band can be defined as 0-200Hz, such as the rubber slow response. A medium frequency band can be defined as 200Hz-2000Hz, representing the fundamental frequency and harmonics of common materials, such as metal tapping. A high frequency band can be defined as 2000Hz-10000Hz, detecting transient impact and surface characteristics, such as glass breaking. Based on the material tapping experimental data statistics, such as the high frequency component of metal accounting for more than 60%, and the low and medium frequency of plastic accounting for more than 70%. When performing material analysis, the high frequency band corresponding to metal, the medium frequency band corresponding to plastic, and the low frequency band filtering engine vibration can be used to reduce the misjudgment rate.

[0102] For the peak frequency, centroid frequency, spectral standard deviation, root mean square frequency, frequency variance and spectral entropy contained in the frequency domain features, they will be introduced one by one as follows:

[0103] The calculation formula of the peak frequency is y1=index(max(p)), which is used to represent the frequency at the highest amplitude.

[0104] The calculation formula of the centroid frequency is: which is used to represent the frequency at the center of gravity of the entire spectrum. The centroid frequency of hard materials is higher, such as the centroid frequency of metal is higher than that of rubber.

[0105] The calculation formula of the spectral entropy is:

[0106] which is used to represent the uncertainty of the spectrum, reflecting the amount of spectral information, or the degree of chaos of energy distribution in the frequency band. The spectral entropy value is higher in the noise scene.

[0107] The calculation formula of the spectral standard deviation is: which is used to describe the fluctuation of the spectral energy change.

[0108] The calculation formula of the root mean square frequency is: which is used to describe the shift of the main frequency band in the spectrum.

[0109] The calculation formula of the frequency variance is: which is used to reflect the energy distribution of the spectrum.

[0110] wherein f i represents the frequency at the spectrum i, N represents the data point number, P i represents the absolute value of the spectral amplitude.

[0111] In some embodiments, the step of splicing the time domain features and the frequency domain features of the plurality of signal sections to obtain a time sequence fusion feature vector comprises:

[0112] S41, normalize the time domain features and the frequency domain features.

[0113] The formula for normalization is prior art, which will not be described here. It should be noted that the normalization of the time domain features and the frequency domain features is performed respectively to eliminate the dimensional difference. Exemplarily, the peak-to-peak value range in the time domain features can be normalized from 0-1000 mV to [-1, 1], and the spectral entropy in the frequency domain features can be normalized from 0-1 to [0, 1], avoiding the dominance of a large range of features in model training and improving the stability of classification.

[0114] S42, concatenate the normalized features in time sequence into a time sequence fusion feature vector.

[0115] The normalized time domain features and the frequency domain features of each signal measure are concatenated in time sequence. Exemplarily, if there are 10 measures, the time domain features are 7-dimensional, and the frequency domain features are 5-dimensional, then the total feature dimension = 10 x (7+5) = 120 dimensions. The strict time sequence arrangement of each measure is ensured by the hardware timestamp, avoiding phase misplacement. A ring buffer is used to store the time sequence feature vector, covering old data and reducing memory occupation. The change rule of the signal over time, such as the starting and decay process of the touch signal, is captured. The ring buffer is beneficial to reduce memory occupation.

[0116] S43, perform principal component analysis on the time sequence fusion feature vector, and reduce the dimension of the time sequence fusion feature vector to a preset dimension.

[0117] Exemplarily, in the offline stage, a principal component analysis (PCA) model is trained using a labeled data set, and the principal components corresponding to 90% of the energy are retained. In the online stage, the pre-trained PCA parameters are loaded, and the feature vector is projected in real time. The preset dimension is 50-100 dimensions, and the specific value is dynamically adjusted according to the hardware computing power. For example, high-performance devices (such as GPU): reduce the dimension to 100 dimensions; embedded devices (such as STM32H7): reduce the dimension to 50 dimensions.

[0118] In some embodiments, the step of inputting the time sequence fusion feature vector into a learning model and outputting a classification probability of the corresponding material includes:

[0119] S51, determine a learning model for classification.

[0120] The machine learning model can include a support vector machine (SVM), a random forest, a Bayesian model; deep learning models include: long short-term memory model (LSTM), recurrent neural network (RNN), deep residual network (ResNet), etc. The main function of these models is to map the input vector to the classification probability of the preset touch material category.

[0121] S52, input the time sequence fusion feature vector into the learning model, and calculate the probability distribution of each material category.

[0122] Exemplarily, 100,000 labeled samples are collected, which can cover eight representative materials such as knuckles, nails, finger pads, palms, wooden rods, plastic rods, glass rods and metal rods. These materials can be divided into two categories, human hand materials and non-human hand materials. These materials have different densities and hardnesses, and will produce elastic wave signals with unique characteristics when knocking on the door. Specifically, human hand materials are part of the human body, and the signals produced by knuckles are relatively weak, but have unique biomechanical characteristics. Non-human hand materials have higher density and hardness, and the signals produced when knocking may contain more high-frequency components and have stronger impact. Environmental noise is also collected, including car music, rain, opening and closing doors, violent shaking, and large vehicles passing by. These environmental noises are also recorded in the data set as non-knuckle knocking. The cross-entropy loss function is used for classification tasks, and the mean square error is used for regression tasks such as material hardness prediction.

[0123] S53, determining the final material type according to the maximum probability.

[0124] Exemplarily, the material category corresponding to the maximum probability is selected as the output result. A confidence threshold is set, and if the highest probability is less than the threshold, it is determined as "unknown material" and the review mechanism is triggered. In the vehicle scene, if the identification is malicious knocking, such as metal tools, real-time sound and light alarm is triggered.

[0125] In summary, the technical scheme provided by the present application is to divide the collected elastic wave data stream into multiple signal sections according to the sliding window, extract time domain features and frequency domain features for each signal section, splice the time domain features and frequency domain features of multiple signal sections, obtain a time-varying feature of frequency domain and time domain, i.e. time sequence fusion feature vector, and finally input the time sequence fusion feature vector into the learning model to output the classification probability of the corresponding material.

[0126] Compared with the related art, the real-time data stream elastic wave signal classification method has higher timeliness and less algorithm calculation amount. Taking an algorithm with a complexity of O(n^2) as an example, after the signal is split into m segments, the complexity of m times of running the algorithm is m x O((n / m)^2), and when m > 1, m x O((n / m)^2) > O(n^2). Taking an algorithm with a complexity of O(n) as an example, after the signal is split into m segments, the complexity of m times of running the algorithm is m x O((n / m)), and in fact m x O((n / m)) = O(n). Therefore, the algorithm complexity required by the method for splitting the real-time signal into multiple segments is not lower than the complexity of the method without splitting, but the calculation amount of each small segment is greatly reduced, and the calculation is performed in segments, without the need to wait for the loading of the complete signal, so that the processing delay is lower.

[0127] In addition, compared with the related art which focuses on the extraction and calculation of frequency domain features, the technical solution provided in the present application combines time domain features and frequency domain features, the time domain features are used as auxiliary identification features, and the frequency domain features are stacked in the time dimension. By inputting the feature that the frequency spectrum information of the signal changes over time into a subsequent discriminant model, the present application can effectively distinguish non-touch signals with similar frequency spectra, and the stability of the system is higher.

[0128] In order to more intuitively reflect the advantages of the technical solution provided in the present application compared with the technical solution provided in the related art, the following experiment is performed:

[0129] In this experiment, the performance indicators of the four groups of models on the test set are determined, including precision, recall, F1 score and overall accuracy (OA). Among them, the precision is the proportion of samples that are actually of a certain material among the samples predicted by the model to be of that material, and the higher the precision, the fewer the misjudgments; the recall is the proportion of samples that are actually of a certain material among the samples that are correctly predicted by the model, and the higher the recall, the fewer the missed detections; the F1 score is the harmonic mean of the precision and the recall, which comprehensively measures the classification ability of the model, and the highest is 1 and the lowest is 0; the overall accuracy is the proportion of correctly classified samples among all samples. The specific experimental results are as follows:

[0130]

[0131] In the recognition method only for frequency domain analysis, the frequency domain features and the integrated learning algorithm (Boosting) are used, the precision is 89.62%, the recall is 92.17%, and the F1 score is 90.87%. The frequency domain and the support vector machine (SVM) are used, the precision is 91.76%, the recall is 91.10%, and the F1 score is 91.42%.

[0132] It can be seen that only relying on frequency domain features is easy to be affected by interference signals with similar spectrum, such as wind vibration and metal knocking.

[0133] In the recognition method of fusion of time domain features and frequency domain features, the precision rate is improved to 94.72%, the recall rate is 95.73%, and the F1 score is 95.22% by using time domain features, frequency domain features and integrated learning algorithm. By using time domain features, frequency domain features and support vector machine (SVM), the precision rate is 95.77%, the recall rate is 96.80%, and the F1 score is 96.28%, and the comprehensive performance is optimal.

[0134] It can be seen that the time domain feature captures dynamic differences and filters interference signals with similar spectrum; the frequency domain feature enhances the material discrimination ability with time superposition.

[0135] It can be seen from the comparison of integrated learning algorithm (Boosting) and support vector machine (SVM) that the integrated learning algorithm is suitable for small and medium-sized data, but is sensitive to noise, and the support vector machine constructs a hyperplane in a high-dimensional feature space and has stronger modeling capability for time series data after feature fusion.

[0136] Figure 5 is a block diagram of a controller 300 according to an example embodiment. As shown, the controller 300 can include a processor 301, a memory 302. The controller 300 can also include one or more of a multimedia component 303, an input / output (I / O) component 304, and a communication component 305. Figure 5

[0137] ​The processor 301 is configured to control overall operations of the controller 300 to complete all or part of the steps in the above method. The memory 302 is configured to store various types of data to support the operations of the controller 300, which can include, for example, instructions for any application or method operating on the controller 300, and application-related data, such as contact data, transmitted and received messages, pictures, audio, video, and the like. The memory 302 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The multimedia component 303 can include a screen and an audio component. The screen can be, for example, a touch screen, and the audio component is configured to output and / or input audio signals. For example, the audio component can include a microphone configured to receive external audio signals. The received audio signals can be further stored in the memory 302 or transmitted through the communication component 305. The audio component also includes at least one speaker configured to output audio signals. The I / O component 304 provides an interface between the processor 301 and other interface modules, which can be a keyboard, a mouse, a button, and the like. The buttons can be virtual buttons or physical buttons. The communication component 305 is configured to perform wired or wireless communication between the controller 300 and other devices. Wireless communication, such as WiFi, Bluetooth, near field communication (NFC), 2G, 3G, 4G, MB-IOT, eMTC, or other 5G, and the like, or a combination of one or more of them, is not limited herein. Therefore, the corresponding communication component 305 can include a WiFi module, a Bluetooth module, an NFC module, and the like.

[0138] In an example embodiment, the controller 300 can be implemented by one or more Application Specific Integrated Circuits (ASICs), Digital Signal Processors (DSPs), Digital Signal Processing Devices (DSPDs), Programmable Logic Devices (PLDs), Field Programmable Gate Arrays (FPGAs), controllers, micro-controllers, microprocessors, or other electronic components, for executing the above-described methods.

[0139] In another example embodiment, a computer program product including a computer program or instructions is also provided, which, when executed by a processor, implements the steps of the above-described methods. For example, the computer program product can be the above-described memory 302 including the computer program or instructions, which can be executed by the processor 301 of the controller 300 to complete each step included in the above-described methods.

[0140] Figure 6 is a block diagram of a vehicle provided in an embodiment of the present application, as shown in the figure, the vehicle 400 includes the above-described controller 300. Figure 6

[0141] The embodiments of the present application also provide a computer readable storage medium, wherein the computer readable storage medium stores a computer program for electronic data exchange, and the computer program causes a computer to execute some or all steps of any one of the collision object identification methods described in the above method embodiments.

[0142] It should be noted that, for the above-described method embodiments, in order to simply describe, they are all expressed as a combination of a series of actions, but those skilled in the art should know that the present application is not limited to the order of the actions described, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should know that the embodiments described in the specification all belong to preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0143] In the above-described embodiments, the description of each embodiment is focused on, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0144] ​In several embodiments provided in the present application, it should be understood that the disclosed apparatus can be implemented in other manners. For example, the division of the apparatus embodiments described above is merely illustrative, and the division of the units can be changed according to actual needs. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections can be indirect couplings or communication connections through some interfaces, devices or units, and can be in electrical, mechanical or other forms.

[0145] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e., can be located in one place, or can be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiments.

[0146] In addition, the functional units in each embodiment of the application can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software program module.

[0147] The integrated unit, if realized in the form of a software program module and sold or used as an independent product, can be stored in a computer readable storage unit. Based on this understanding, the technical solutions of the present application essentially or the part of the prior art that contributes to the technical solutions or all or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage unit and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage unit includes: a U disk, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk, and various media that can store program codes.

[0148] A person of ordinary skill in the art can understand that all or part of the steps of the various methods in the above embodiments can be completed by a program instructing relevant hardware, and the program can be stored in a computer readable storage unit, which can include a flash disk, a read-only memory, a random access memory, a magnetic disk or an optical disk, etc.

[0149] The preferred embodiments of the present application are described in detail above with reference to the drawings, but the present application is not limited to the specific details of the above-described embodiments, and various simple modifications can be made to the technical solutions of the present application within the scope of the technical concept of the present application, and these simple modifications all belong to the protection scope of the present application.

[0150] In addition, it should be noted that each specific technical feature described in the above specific embodiments can be combined in any appropriate manner without contradiction, and in order to avoid unnecessary repetition, various possible combinations are not described again in the present application.

[0151] In addition, any combination of various different embodiments of the present application can also be made, as long as it does not deviate from the idea of the present application, and it should also be considered as disclosed in the present application.

[0152] In the description of the present application, the terms "first", "second" are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0153] In the above embodiments, the description of each embodiment is focused on, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0154] The embodiments, embodiments and related technical features of the present application can be combined, replaced with each other without conflict.

[0155] The above is only the preferred embodiment of the present application, and does not limit the present application in any form, but any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present application without departing from the technical solution content of the present application still belongs to the scope of the technical solution of the present application.

Claims

1. A method for identifying colliding objects, characterized in that, include: A data stream is formed based on the elastic wave signal generated by touch; The data stream is divided into multiple signal segments; For each of the aforementioned signal segments, extract time-domain features and frequency-domain features respectively; The time-domain features and frequency-domain features of multiple signal segments are concatenated to obtain a time-series fusion feature vector; Based on the temporal fusion feature vector, the material classification of the touched material is output.

2. The method according to claim 1, characterized in that, The step of forming a data stream based on the elastic wave signal generated by touch includes: Based on the elastic wave signal transmitted through the elastic wave sensor, the peak-to-peak value of the signal peak is monitored in real time. When the peak-to-peak value of the signal exceeds the preset peak value, the pre-trigger buffered data and the fixed-length post-trigger data are merged into a data stream.

3. The method according to claim 2, characterized in that, The preset peak value is no more than 300mV.

4. The method according to claim 2, characterized in that, The step of merging the pre-trigger buffered data and the post-trigger fixed-length data into a data stream when the peak-to-peak value of the signal peak exceeds the preset peak value also includes: merging the pre-trigger buffered data and the post-trigger fixed-length data into a data stream when the peak-to-peak value of the signal peak exceeds the preset peak value, the rise time of the elastic wave signal is less than 10ms, and the fall time is less than 20ms.

5. The method according to claim 1, characterized in that, The step of dividing the data stream into multiple signal segments includes: dividing the data stream into multiple signal segments according to a sliding window.

6. The method according to claim 5, characterized in that, The step of dividing the data stream into multiple signal segments according to the sliding window includes: The length of the sliding window is determined based on the signal sampling rate and the number of data points, wherein multiple sliding windows have the same length; Based on the length of the sliding window, the data stream is divided into multiple non-overlapping signal segments.

7. The method according to claim 6, characterized in that, The length of the sliding window satisfies the following formula: L=M / F s ; Where L is the length of the sliding window, M is the number of data points used in the Fast Fourier Transform; F s This represents the signal sampling rate.

8. The method according to claim 1, characterized in that, The time-domain features include at least one of the following: root mean square value, peak-to-peak value, variance, rectified average value, root square amplitude, waveform skewness, and kurtosis.

9. The method according to claim 1, characterized in that, The frequency domain feature extraction steps include: Calculate the spectrum; The spectrum is divided into low-frequency band, mid-frequency band, and high-frequency band. Obtain at least one of the following features for each frequency band: peak frequency, centroid frequency, spectral standard deviation, root mean square frequency, frequency variance, and spectral entropy.

10. The method according to claim 1, characterized in that, The step of concatenating the time-domain features and frequency-domain features of multiple signal segments to obtain a time-series fusion feature vector includes: Normalize the time-domain features and the frequency-domain features; The normalized features are concatenated in chronological order to form a temporal fusion feature vector; Principal component analysis is performed on the time-series fusion feature vector to reduce its dimensionality to a preset dimension.

11. The method according to claim 1, characterized in that, The step of outputting the material classification of the touched material based on the temporal fusion feature vector includes: Determine the learning model to be used for classification; The temporal fusion feature vector is input into the learning model to calculate the probability distribution of each material category; The final material type is determined based on the maximum probability.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program performs the steps of the method described in any one of claims 1 to 11.

13. A controller having a computer program stored thereon, characterized in that, When executed by a processor, the computer program performs the steps of the method described in any one of claims 1 to 11.

14. A vehicle, characterized in that, Includes the controller as described in claim 13.

15. A computer program product, characterized in that, It includes a computer program or instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 11.

Images