Method, device and apparatus for correcting detection signals of sensors on self-driving cars

Abstract

The present application relates to a detection signal correction method, device and equipment for a sensor on an autonomous vehicle. The method comprises: firstly obtaining a detection signal collected by a sensor, and extracting features from the detection signal to obtain a first feature, performing discrete wavelet transform processing on the first feature to obtain a first high-frequency feature and a first low-frequency feature, processing the first low-frequency feature using an adaptive frequency weighting algorithm to obtain a second feature, determining a target low-frequency feature according to the second feature, a discrete wavelet transform algorithm and an inverse discrete wavelet transform algorithm, and correcting the detection signal according to the first high-frequency feature, the target low-frequency feature and the inverse discrete wavelet transform algorithm, wherein the frequency of the first high-frequency feature is greater than or equal to a preset frequency threshold, and the frequency of the first low-frequency feature is less than the preset frequency threshold. The present application takes into account the frequency aliasing in the detection signal, improves the accuracy of the correction of the detection signal, and thus ensures the safe driving of the autonomous vehicle.

Description

Technical Field

[0001] The present application relates to the technical field of autonomous vehicles, and in particular to a method, device and apparatus for correcting detection signals of sensors on autonomous vehicles. Background Art

[0002] In recent years, autonomous vehicles have achieved rapid breakthroughs and have successfully operated autonomously on closed or semi-closed urban roads. This rapid progress is primarily due to the rapid development of artificial intelligence (AI) and sensor technologies. Autonomous vehicles utilize a variety of sensors to perceive their environment and recognize their own posture, and make autonomous decisions through AI-based autonomous driving models. Autonomous vehicle sensors, such as lidar and cameras, primarily provide external perception. Furthermore, sensors such as brake pressure, throttle pressure, and steering angle are crucial to the vehicle's driving safety. However, over extended use, these sensors inevitably experience fatigue and degradation, leading to functional failure, which can seriously compromise the safety of autonomous vehicles. Sensor failure compromises the accuracy of the sensor's detection signals. If inaccurate detection signals are received by autonomous vehicles, they can lead to unforeseen accidents. Therefore, accurate correction of detection signals is crucial for autonomous vehicles to prevent potentially dangerous situations.

[0003] Traditionally, convolutional neural networks (CNNs) have been used to perform hierarchical feature learning using multiple randomly initialized convolution kernels to correct sensor signals. However, convolutional neural networks have poor frequency analysis capabilities, resulting in poor accuracy in correcting detection signals. Summary of the Invention

[0004] Based on this, it is necessary to provide a detection signal correction method, device and equipment for sensors on autonomous driving vehicles that can improve the accuracy of detection signal correction of sensors in response to the above technical problems.

[0005] In a first aspect, the present application provides a method for calibrating detection signals from sensors on an autonomous vehicle. The method comprises:

[0006] Acquire a detection signal collected by the sensor, and perform feature extraction on the detection signal to obtain a first feature;

[0007] Performing discrete wavelet transform on the first feature to obtain a first high-frequency feature and a first low-frequency feature; the frequency of the first high-frequency feature is greater than or equal to a preset frequency threshold, and the frequency of the first low-frequency feature is less than the preset frequency threshold;

[0008] Processing the first low-frequency feature using an adaptive frequency weighting algorithm to obtain a second feature;

[0009] Determining a target low-frequency feature based on the second feature, a discrete wavelet transform algorithm, and an inverse discrete wavelet transform algorithm;

[0010] The detection signal is corrected according to the first high-frequency feature, the target low-frequency feature and an inverse discrete wavelet transform algorithm.

[0011] In one embodiment, determining the target low-frequency feature based on the second feature, a discrete wavelet transform algorithm, and an inverse discrete wavelet transform algorithm includes:

[0012] Performing a discrete wavelet transform on the second feature using the discrete wavelet transform algorithm to obtain a second high-frequency feature and a second low-frequency feature; the frequency of the second high-frequency feature is greater than or equal to the preset frequency threshold, and the frequency of the second low-frequency feature is less than the preset frequency threshold;

[0013] determining a third high frequency feature based on the adaptive frequency discarding and the second high frequency feature;

[0014] Determining a third low-frequency feature based on the second low-frequency feature, the discrete wavelet transform algorithm, and the inverse discrete wavelet transform algorithm;

[0015] The target low-frequency feature is determined according to the third high-frequency feature, the third low-frequency feature, and the inverse discrete wavelet transform algorithm.

[0016] In one embodiment, determining the third low-frequency feature according to the second low-frequency feature, the discrete wavelet transform algorithm, and the inverse discrete wavelet transform algorithm includes:

[0017] The second low-frequency feature is processed using an adaptive frequency weighting algorithm to obtain a third feature;

[0018] Performing discrete wavelet transform on the third feature to obtain a fourth high-frequency feature and a fourth low-frequency feature;

[0019] Processing the fourth high-frequency feature using an adaptive frequency discarding algorithm to obtain a processed fourth high-frequency feature;

[0020] The third low-frequency feature is determined according to the fourth low-frequency feature, the processed fourth high-frequency feature, and an inverse discrete wavelet transform algorithm.

[0021] In one embodiment, determining the target low-frequency feature according to the third high-frequency feature, the third low-frequency feature, and the inverse discrete wavelet transform algorithm includes:

[0022] The third high-frequency feature and the third low-frequency feature are processed by using the inverse discrete wavelet transform algorithm to obtain a fourth feature;

[0023] The fourth feature is convolved to obtain the target low-frequency feature.

[0024] In one embodiment, the detection signal is corrected according to the first high-frequency feature, the target low-frequency feature, and an inverse discrete wavelet transform algorithm, including:

[0025] Processing the first high-frequency feature using an adaptive frequency discarding algorithm to obtain a processed first high-frequency feature;

[0026] Using the inverse discrete wavelet transform algorithm to process the processed first high-frequency feature and the target low-frequency feature to obtain a fifth feature;

[0027] A convolution process is performed on the fifth feature to correct the detection signal.

[0028] In one embodiment, the method further comprises:

[0029] The fourth low-frequency feature is input into a fault classification model to obtain a fault category of the sensor.

[0030] In a second aspect, the present application also provides a device for correcting detection signals of sensors on an autonomous vehicle. The device comprises:

[0031] an acquisition module, configured to acquire a detection signal collected by the sensor and perform feature extraction on the detection signal to obtain a first feature;

[0032] a first determining module, configured to perform discrete wavelet transform processing on the first feature to obtain a first high-frequency feature and a first low-frequency feature; wherein the frequency of the first high-frequency feature is greater than or equal to a preset frequency threshold, and the frequency of the first low-frequency feature is less than the preset frequency threshold;

[0033] A second determining module is configured to process the first low-frequency feature using an adaptive frequency weighting algorithm to obtain a second feature;

[0034] A third determining module is used to determine the target low-frequency feature according to the second feature, a discrete wavelet transform algorithm and an inverse discrete wavelet transform algorithm;

[0035] The correction module is used to correct the detection signal according to the first high-frequency feature, the target low-frequency feature and an inverse discrete wavelet transform algorithm.

[0036] In a third aspect, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0037] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the above method when executed by a processor.

[0038] In a fifth aspect, the present application further provides a computer program product, which includes a computer program that implements the steps of the above method when executed by a processor.

[0039] The detection signal correction method, device and equipment of the sensor on the above-mentioned autonomous driving vehicle first obtains the detection signal collected by the sensor, and performs feature extraction on the detection signal to obtain a first feature, performs discrete wavelet transform processing on the first feature to obtain a first high-frequency feature and a first low-frequency feature, uses an adaptive frequency weighting algorithm to process the first low-frequency feature to obtain a second feature, determines the target low-frequency feature based on the second feature, the discrete wavelet transform algorithm and the inverse discrete wavelet transform algorithm, and corrects the detection signal based on the first high-frequency feature, the target low-frequency feature and the inverse discrete wavelet transform algorithm, wherein the frequency of the first high-frequency feature is greater than or equal to a preset frequency threshold, and the frequency of the first low-frequency feature is less than the preset frequency threshold. In traditional technology, convolutional neural networks are used for hierarchical feature learning to correct the detection signals detected by sensors. However, the frequency analysis capability of convolutional neural networks is very poor. In this application, discrete wavelet transform is used to divide the first feature into a first high-frequency feature and a first low-frequency feature, and then the first low-frequency feature is processed using an adaptive frequency weighting algorithm. After that, the corrected detection signal is obtained according to the inverse discrete wavelet transform algorithm. Since the adaptive frequency weighting algorithm can automatically identify valuable frequency information and give it a higher weight, and ignore or suppress useless frequency information, this application takes into account the frequency aliasing in the detection signal, improves the accuracy of the correction of the detection signal, and thus ensures the safe driving of the autonomous driving vehicle. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 This is a flowchart illustrating a method for correcting detection signals of sensors on a self-driving vehicle provided in an embodiment of the present application;

[0041] Figure 2 Schematic diagram of the frequency decomposition layer architecture driven by discrete wavelet transform;

[0042] Figure 3 A schematic diagram of the architecture of adaptive frequency weighting provided in an embodiment of the present application;

[0043] Figure 4 This is a flow chart of a method for determining target low-frequency characteristics provided in an embodiment of the present application;

[0044] Figure 5 A schematic diagram of the architecture of the adaptive frequency drop algorithm provided in an embodiment of the present application;

[0045] Figure 6A schematic flow chart of a third low-frequency feature determination method provided in an embodiment of the present application;

[0046] Figure 7 The second flowchart of a method for determining target low-frequency characteristics provided in an embodiment of the present application;

[0047] Figure 8 A flow chart of a method for correcting a detection signal provided in an embodiment of the present application;

[0048] Figure 9 A model structure diagram of a detection signal correction algorithm provided in an embodiment of the present application;

[0049] Figure 10 A structural diagram of a car provided in an embodiment of the present application;

[0050] Figure 11 An experimental analysis result diagram provided in an embodiment of the present application;

[0051] Figure 12 This is a structural block diagram of a detection signal correction device for a sensor on an autonomous vehicle provided in an embodiment of the present application;

[0052] Figure 13 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0054] In one embodiment, Figure 1 This is a flowchart illustrating a method for calibrating a detection signal of a sensor on a self-driving vehicle provided in an embodiment of the present application. The method includes the following steps:

[0055] S101: Acquire a detection signal collected by a sensor, and perform feature extraction on the detection signal to obtain a first feature.

[0056] The detection signal may include signals related to the vehicle's bus data, such as Global Positioning System (GPS) coordinates / latitude and longitude, vehicle pitch angle, vehicle roll angle, vehicle speed, accelerator pedal pressure, brake pressure, steering angle, XYZ axis acceleration, XYZ axis angular velocity, etc., where X represents the horizontal axis of the coordinate system, Y represents the vertical axis of the coordinate system, and Z represents the vertical axis of the coordinate system.

[0057] In an embodiment of the present application, a computer device obtains a detection signal collected by a sensor, first processes the detection signal using a sliding segmentation method to obtain a first detection signal, then standardizes the processed detection signal using a 0-1 normalization principle to obtain a second detection signal, and extracts features from the second detection signal through convolution to obtain a first feature.

[0058] S102 , performing discrete wavelet transform processing on the first feature to obtain a first high-frequency feature and a first low-frequency feature; the frequency of the first high-frequency feature is greater than or equal to a preset frequency threshold, and the frequency of the first low-frequency feature is less than the preset frequency threshold.

[0059] The preset frequency threshold is a critical value between high-frequency features and low-frequency features.

[0060] In an embodiment of the present application, the computer device performs discrete wavelet transform processing on the first feature to obtain a first high-frequency feature and a first low-frequency feature, where the frequency of the first high-frequency feature is greater than or equal to a preset frequency threshold, and the frequency of the first low-frequency feature is less than the preset frequency threshold.

[0061] Specifically, refer to Figure 2 , Figure 2 Schematic diagram of the frequency decomposition layer architecture driven by discrete wavelet transform. Discrete wavelet transform is a time-frequency signal analysis method that uses predefined wavelet basis functions to perform multi-scale time-frequency decomposition of information to obtain frequency domain features. To process discrete signals, the scale factor and shift factor are discretized to obtain discrete wavelet transform (DWT). Integrating DWT into deep convolutional neural networks can achieve hierarchical frequency decomposition and learning. In one-dimensional convolution, the information flow is represented by a two-dimensional feature map, which consists of multiple feature vectors containing different feature information. Assume that the feature map in the one-dimensional convolution is F∈R C×L , C represents the number of eigenvectors, i.e. the number of network channels, and L is the length of the eigenvector. The Mallat algorithm is used to implement discrete wavelet transform in the convolutional neural network, and channel-independent wavelet transform is performed. Each eigencomponent is decomposed into low-frequency components and high-frequency components by discrete wavelet transform. Assume that F i (n) is the i-th eigenvector of the feature graph F, given the approximate expression Based on the Mallat algorithm, the following discrete wavelet transform recursive formula can be obtained:

[0062]

[0063] in is the signal F i The approximate coefficient of (n) (low frequency part), is the signal F i(n) detail coefficient (high frequency part). Coefficients h*(n) and g * (n) is a low-pass and high-pass wavelet filter bank determined by the scaling function and the wavelet function. Similarly, based on the Mallat algorithm, the following expression process of the inverse discrete wavelet transform can be obtained:

[0064]

[0065] The above process can be embedded in the encoder of the training model, and the feature map can be decomposed into low-frequency components and high-frequency components. The low-frequency component describes the basic object structure of the signal, and the high-frequency component describes the degree of signal oscillation and transformation and contains a certain noise component. It can achieve downsampling and complete frequency decomposition at the same time, solve the frequency aliasing problem in the detection signal of multi-sensor, and improve the frequency learning ability of convolutional neural network.

[0066] S103: Process the first low-frequency feature using an adaptive frequency weighting algorithm to obtain a second feature.

[0067] In an embodiment of the present application, the computer device performs convolution processing on the first low-frequency feature obtained in S102, and then processes the convolved first low-frequency feature using an adaptive frequency weighting algorithm to obtain a second feature. The adaptive frequency weighting algorithm can automatically identify valuable frequency information and assign it a higher weight, while ignoring or suppressing useless frequency information.

[0068] Specifically, since the low-frequency features in the encoded signal of the encoder layer in the training model can generate robust high-level features and provide highly distinguishable fault-related features for the fault classifier, in order to dynamically enhance the frequency feature learning ability of the encoder, an adaptive frequency weighting algorithm (AFW) is proposed. Figure 3 , Figure 3 Schematic diagram of the adaptive frequency weighting architecture provided in the embodiment of the present application. Assume that the frequency characteristic input to the adaptive frequency weighting module is L l ∈R C×L , which contains C channels with a length of L feature vector. First, the adaptive frequency weighting module encodes the relative importance of different channels through feature compression, encoding and decoding. Feature L l The global average pooling is performed on each channel to generate a channel information representation vector. This process can be expressed as:

[0069]

[0070] ω represents the channel representation vector, represents the jth eigenvalue of the i-th channel. Subsequently, the adaptive frequency weighting module designs an encoder-decoder architecture, and the vector ω is used to encode the correlation between different features in a nonlinear way. Figure 3 As shown in Figure 2, the vector is first compressed through two convolutional layers and then restored to its original dimension. Subsequently, the Sigmoid function is used to map the weight of the vector to between 0 and 1 to obtain the frequency weight vector As the network is trained and optimized, the frequency weight vector will represent the importance of its corresponding channel features. The adaptive frequency weighting module uses channel multiplication to pass Adaptively optimize and adjust the frequency characteristics of network information flow. It is automatically generated through network optimization. Therefore, the model will generate different weight vectors for different inputs, so that each input signal is optimized for its frequency characteristics, thereby enabling the encoder to have more effective frequency feature encoding capabilities.

[0071] S104 : Determine the target low-frequency feature according to the second feature, a discrete wavelet transform algorithm, and an inverse discrete wavelet transform algorithm.

[0072] In an embodiment of the present application, the computer device can use discrete wavelet transform to process the second feature to obtain a second high-frequency feature and a second low-frequency feature, and then perform inverse discrete wavelet transform on the third high-frequency feature obtained after adaptive frequency discarding the second high-frequency feature and the second low-frequency feature to determine the target low-frequency feature.

[0073] The second feature can also be processed by a discrete wavelet transform algorithm to obtain a second high-frequency feature and a second low-frequency feature, and then the second low-frequency feature is convolved, and then the second low-frequency feature after the convolution is processed using an adaptive frequency weighting algorithm to obtain a third feature; the third feature is processed by a discrete wavelet transform to obtain a fourth low-frequency feature and a fourth high-frequency feature; the fourth high-frequency feature is processed using an adaptive frequency discarding algorithm to obtain a feature processed by the adaptive frequency discarding algorithm, and the feature and the fourth low-frequency feature are processed using an inverse discrete wavelet transform algorithm (IDWT) to determine the target low-frequency feature.

[0074] S105 , correcting the detection signal according to the first high-frequency feature, the target low-frequency feature, and an inverse discrete wavelet transform algorithm.

[0075] In an embodiment of the present application, a computer device processes the first high-frequency feature and the target low-frequency feature using an inverse discrete wavelet transform algorithm to obtain a fifth feature. After performing convolution extraction on the fifth feature, a corrected detection signal can be obtained.

[0076] In an embodiment of the present application, a detection signal collected by a sensor is first obtained, and a feature extraction is performed on the detection signal to obtain a first feature. The first feature is processed by discrete wavelet transform to obtain a first high-frequency feature and a first low-frequency feature. The first low-frequency feature is processed using an adaptive frequency weighting algorithm to obtain a second feature. The target low-frequency feature is determined based on the second feature, the discrete wavelet transform algorithm, and the inverse discrete wavelet transform algorithm. The detection signal is corrected based on the first high-frequency feature, the target low-frequency feature, and the inverse discrete wavelet transform algorithm, wherein the frequency of the first high-frequency feature is greater than or equal to a preset frequency threshold, and the frequency of the first low-frequency feature is less than the preset frequency threshold. In traditional technology, a convolutional neural network is used for hierarchical feature learning to correct the detection signal detected by the sensor. However, the frequency analysis capability of the convolutional neural network is poor. In this application, discrete wavelet transform is used to divide the first feature into a first high-frequency feature and a first low-frequency feature, and then the first low-frequency feature is processed using an adaptive frequency weighting algorithm. Thereafter, the corrected detection signal is obtained according to the inverse discrete wavelet transform algorithm. The adaptive frequency weighting algorithm can automatically identify valuable frequency information and give it a higher weight, while ignoring or suppressing useless frequency information. Therefore, this application takes into account the frequency aliasing in the detection signal, improves the accuracy of the correction of the detection signal, and thus ensures the safe driving of the autonomous driving vehicle.

[0077] In one embodiment, Figure 4 This is one of the flow charts of a method for determining a target low-frequency feature provided in an embodiment of the present application. This embodiment relates to a possible implementation method for determining the target low-frequency feature based on the second feature, a discrete wavelet transform algorithm, and an inverse discrete wavelet transform algorithm. Based on the above embodiment, the above S104 includes:

[0078] S401. Perform discrete wavelet transform on the second feature using a discrete wavelet transform algorithm to obtain a second high-frequency feature and a second low-frequency feature; the frequency of the second high-frequency feature is greater than or equal to a preset frequency threshold, and the frequency of the second low-frequency feature is less than the preset frequency threshold.

[0079] In an embodiment of the present application, the computer device can use a discrete wavelet transform algorithm to perform discrete wavelet transform processing on the second feature to obtain a second high-frequency feature and a second low-frequency feature, wherein the frequency of the second high-frequency feature is greater than or equal to a preset frequency threshold, and the frequency of the second low-frequency feature is less than the preset frequency threshold.

[0080] S402: Determine a third high-frequency feature according to the adaptive frequency discarding algorithm and the second high-frequency feature.

[0081] In an embodiment of the present application, the computer device may process the second high-frequency feature using an adaptive frequency discarding (AFD) algorithm to obtain a third high-frequency feature.

[0082] It should be noted that the reference Figure 5 , Figure 5 Schematic diagram of the architecture of the adaptive frequency discarding algorithm provided in the embodiment of the present application. Since the high-frequency features not only contain the characteristics of the signal itself, but also contain certain noise information and some fault characteristics. The high-frequency features are mainly used to reconstruct the signal in the decoder. In order to remove the fault information or noise information in the detection signal of the sensor, so as to obtain the normal and pure detection signal of the sensor and achieve the purpose of correcting the detection signal, the ability of the decoder to remove irrelevant frequency information can be dynamically improved. The adaptive frequency discarding algorithm can automatically identify the fault or noise frequency information, and set the fault or noise frequency information to 0, and give different weights to valuable features. Assume that the frequency feature input to the adaptive frequency discarding algorithm is L h ∈R C×L , which contains C channels of feature vectors of length L. The adaptive frequency drop algorithm uses feature compression, encoding and decoding to encode the correlation between frequency features. First, two 3x3 convolutional layers are used to encode the feature L h Feature extraction and encoding are performed, which is encoded into a feature map with a channel number of C / 2. Subsequently, a 3x3 convolution is used to reconstruct the feature L h , and uses the linear rectification function (Rectified Linear Unit, ReLU) to automatically set part of the information to 0, thereby obtaining the frequency weight map ω.

[0083] ω=relu(conv c (conv c / 2 (conv c (L h ))))

[0084] where conv c Represents a convolutional layer with C channels. As the network is trained and optimized, the frequency weight map will characterize the importance of the features at the corresponding position. Finally, the adaptive frequency drop algorithm adaptively optimizes and discards the frequency features in the network information flow by multiplying the elements at the corresponding positions of the two matrices. The difference from the adaptive frequency weighting algorithm is that the frequency weight map ω in the adaptive frequency drop algorithm is related to the feature L h The dimensions of the features are the same, and each feature value can be given an independent weight to obtain a more detailed frequency optimization effect, which can prompt the decoder to retain valuable information more carefully and accurately and discard noise or fault information.

[0085] S403 : Determine a third low-frequency feature according to the second low-frequency feature, a discrete wavelet transform algorithm, and an inverse discrete wavelet transform algorithm.

[0086] In an embodiment of the present application, the computer device can perform a convolution operation on the second low-frequency feature and then perform an adaptive weighting algorithm to obtain a third feature, and then perform discrete wavelet transform processing on the third feature to obtain a low-frequency feature and a high-frequency feature, and then use an adaptive frequency discarding algorithm to process the high-frequency feature, and perform inverse discrete wavelet transform processing on the high-frequency feature obtained at this time and the low-frequency feature obtained by performing discrete wavelet transform processing on the third feature to obtain a low-frequency feature, so as to determine the third low-frequency feature.

[0087] S404: Determine a target low-frequency feature according to the third high-frequency feature, the third low-frequency feature, and an inverse discrete wavelet transform algorithm.

[0088] In an embodiment of the present application, the computer device can use an inverse discrete wavelet transform algorithm to perform inverse discrete wavelet transform on the third high-frequency feature and the third low-frequency feature to obtain a fourth feature. After convolution processing of the fourth feature, the target low-frequency feature can be determined.

[0089] In this embodiment, the second feature is subjected to discrete wavelet transform using a discrete wavelet transform algorithm to obtain a second high-frequency feature and a second low-frequency feature. The third high-frequency feature is determined based on the adaptive frequency discarding algorithm and the second high-frequency feature. Finally, the third low-frequency feature is determined based on the second low-frequency feature, the discrete wavelet transform algorithm and the inverse discrete wavelet transform algorithm. The third low-frequency feature determined by this method is more accurate.

[0090] In one embodiment, Figure 6 A flowchart of a method for determining a third low-frequency feature provided in an embodiment of the present application is provided. This embodiment relates to a possible implementation method for determining the third low-frequency feature based on the second low-frequency feature, a discrete wavelet transform algorithm, and an inverse discrete wavelet transform algorithm. Based on the above embodiment, the above S403 includes:

[0091] S601: Process the second low-frequency feature using an adaptive frequency weighting algorithm to obtain a third feature.

[0092] In an embodiment of the present application, the computer device can use an adaptive frequency weighting algorithm to perform weighted processing on the second low-frequency feature to obtain a third feature.

[0093] S602: Perform discrete wavelet transform on the third feature to obtain a fourth high-frequency feature and a fourth low-frequency feature.

[0094] In the embodiment of the present application, the computer device may obtain a fourth high-frequency feature and a fourth low-frequency feature after performing discrete wavelet transform processing on the third feature using discrete wavelet transform.

[0095] S603: Process the fourth high-frequency feature using an adaptive frequency discarding algorithm to obtain a processed fourth high-frequency feature.

[0096] In an embodiment of the present application, the computer device may utilize an adaptive frequency discarding algorithm to perform useless noise discarding processing on the fourth high-frequency feature, and may obtain a processed fourth high-frequency feature.

[0097] S604: Determine a third low-frequency feature according to the fourth low-frequency feature, the processed fourth high-frequency feature, and an inverse discrete wavelet transform algorithm.

[0098] In an embodiment of the present application, a computer device processes the fourth low-frequency feature and the processed fourth high-frequency feature using an inverse discrete wavelet transform algorithm to obtain a restored feature, and then performs convolution processing on the restored feature to determine the third low-frequency feature.

[0099] In this embodiment, the second low-frequency feature is processed by using an adaptive frequency weighting algorithm to obtain the third feature, the third feature is processed by discrete wavelet transform to obtain the fourth high-frequency feature and the fourth low-frequency feature, and then the fourth high-frequency feature is processed by an adaptive frequency discarding algorithm to obtain the processed fourth high-frequency feature. Finally, the method of determining the third low-frequency feature is based on the fourth low-frequency feature, the processed fourth high-frequency feature and the inverse discrete wavelet transform algorithm. This can obtain a more detailed frequency optimization effect and improve the accuracy of correction of the detection signal.

[0100] In one embodiment, Figure 7 This is a second flow chart of a method for determining a target low-frequency feature provided in an embodiment of the present application. This embodiment relates to a possible implementation method for determining the target low-frequency feature based on the third high-frequency feature, the third low-frequency feature, and the inverse discrete wavelet transform algorithm. Based on the above embodiment, the above S404 includes:

[0101] S701. Process the third high-frequency feature and the third low-frequency feature using an inverse discrete wavelet transform algorithm to obtain a fourth feature.

[0102] In an embodiment of the present application, the computer device uses an inverse discrete wavelet transform algorithm to restore the third high-frequency feature and the third low-frequency feature to obtain a restored fourth feature.

[0103] S702: Perform convolution processing on the fourth feature to obtain a target low-frequency feature.

[0104] In an embodiment of the present application, the computer device can perform convolution processing on the fourth feature, extract feature information in the fourth feature, and obtain the target low-frequency feature.

[0105] In this embodiment, by using the inverse discrete wavelet transform algorithm to process the third high-frequency feature and the third low-frequency feature to obtain the fourth feature, and then performing convolution processing on the fourth feature to obtain the target low-frequency feature, a more accurate target low-frequency feature can be obtained.

[0106] In one embodiment, Figure 8 A flow chart of a method for correcting a detection signal provided in an embodiment of the present application is provided. This embodiment relates to a possible implementation of how to correct a detection signal based on a first high-frequency feature, a target low-frequency feature, and an inverse discrete wavelet transform algorithm. Based on the above embodiment, the above S105 includes:

[0107] S801: Process the first high-frequency feature using an adaptive frequency discarding algorithm to obtain a processed first high-frequency feature.

[0108] In an embodiment of the present application, the computer device discards the first high-frequency feature using an adaptive frequency discarding algorithm to obtain a processed first high-frequency feature.

[0109] S802: Use an inverse discrete wavelet transform algorithm to process the processed first high-frequency feature and the target low-frequency feature to obtain a fifth feature.

[0110] In an embodiment of the present application, the computer device uses an inverse discrete wavelet transform algorithm to restore the processed first high-frequency feature and the target low-frequency feature to obtain a restored fifth feature.

[0111] S803: Perform convolution processing on the fifth feature to correct the detection signal.

[0112] In an embodiment of the present application, the computer device can use convolution processing to extract the fifth feature and obtain a corrected detection signal.

[0113] In this embodiment, the first high-frequency feature is processed by using an adaptive frequency discarding algorithm to obtain a processed first high-frequency feature, and then the processed first high-frequency feature and the target low-frequency feature are processed by an inverse discrete wavelet transform algorithm to obtain a fifth feature. Finally, the fifth feature is convolved to correct the detection signal, which can improve the accuracy of the detection signal correction of the sensor.

[0114] In one embodiment, the method further comprises:

[0115] The fourth low-frequency feature is input into the fault classification model to obtain the fault category of the sensor.

[0116] In this embodiment of the present application, the computer device inputs the fourth low-frequency feature into a fault classification model to obtain the sensor fault category. The fault classification model may be composed of convolutional layers and fully connected layers, and is responsible for mapping the high-level feature information obtained by the encoder in the model into a fault category space.

[0117] In this embodiment, the accuracy of detecting the fault type of the sensor can be improved by inputting the fourth low-frequency feature into the fault classification model to obtain the fault type of the sensor.

[0118] In one embodiment, Figure 9 This is a model structure diagram of a detection signal correction algorithm provided in an embodiment of the present application. The model (WavePHF) obtains sensor fault-related features from multivariate sensor data through layered frequency feature learning, and adopts a convolutional encoder-decoder architecture and a layered frequency discarding module to filter the fault information in the sensor's detection signal, and accurately restore the real sensor information. The encoder consists of multiple convolution layers, a wavelet-driven frequency decomposition layer (Wavelet-Powered Frequency Decomposition Layer, WPFD-Layer) and an adaptive frequency weighting module. The wavelet-driven frequency decomposition layer decomposes the features layer by layer into low-frequency features and high-frequency features. The convolution layer is responsible for learning useful features from the decomposed frequency features. In order to enhance the frequency learning ability of the encoder, an adaptive frequency weighting algorithm is used to efficiently and dynamically aggregate fault-related frequency information.

[0119] Among them, the fault classification module is placed after the encoder. It mainly consists of a convolutional layer and a fully connected layer, which is responsible for mapping the high-level feature information obtained by the encoder to the fault category space.

[0120] The decoder is then responsible for decoding the obtained high-level features and outputting the detection signals of a normal sensor without fault information. The decoder consists of multiple convolutional layers, an inverse wavelet transform layer, and an adaptive frequency dropout module. The inverse wavelet transform layer receives high-frequency and low-frequency features from the encoder to restore the signal. In particular, the adaptive frequency dropout module is proposed to adaptively discard useless and irrelevant frequency features to assist the model in restoring fault-free sensor information. Similarly, the convolutional layers implement feature learning and information encoding in a hierarchical manner to restore the detection signals of normal sensors layer by layer. This model fully integrates the two tasks of fault identification and fault restoration, which allows for deep sharing of task-specific features between different tasks, achieving mutual learning and joint optimization. It should be noted that this model, WavePHF, can diagnose faults and then correct erroneous detection signals, that is, it can correct erroneous sensor information.

[0121] Figure 10 A vehicle structure diagram provided for an embodiment of the present application. Using the published autonomous driving dataset (A2D2), an autonomous vehicle drives on highways, rural and urban roads, and then collects data from the autonomous vehicle's camera, lidar, acceleration of the autonomous vehicle's XYZ axis, angular velocity of the XYZ axis, GPS coordinates (latitude and longitude), pitch and roll angles, vehicle speed, accelerator pedal pressure, brake pressure, steering angle, etc. The collected multivariate signal samples can be expressed as: Z = {z1, z2, ..., z ∈},∈ represents the number of detection signals of the sensor. i Represents the detection signal of the i-th sensor. Due to sensor degradation or contamination and interference, the sensor's detection signal will have a variety of different failure modes. Through a large amount of signal acquisition and fault injection work, a data set can be obtained, where N is the number of data samples, Z i ∈R ∈×d is the i-th multivariate detection signal, and its corresponding fault label is H i , the data also gives the fault signal Z i The corresponding normal signal The joint architecture of sensor fault classification and detection signal correction is: γ(υ F ,υ R ),υ F ,υ R is a learnable parameter used for fault classification and correction of detection signals. F ,υ R ) uses hierarchical feature representation to map multivariate signals to fault category space and removes fault modes to obtain normal fault samples. Using the obtained dataset and the constructed Figure 9 During the model training process, the adaptive moment estimation (Adam) optimization algorithm was used, the learning rate was set to 0.0001, and the batch size was set to 64.

[0122] The accuracy score and balanced F score (i.e., Accuracy score and F1 score) are used to measure the performance of the model. The larger their values, the better the model performance. The correction of the detection signal is defined as a reconstruction task. Therefore, the mean squared error (MSE), mean absolute error (MAE), and coefficient of determination (R2-Score) are used to measure the performance of the model's detection signal correction. They are defined as:

[0123]

[0124]

[0125]

[0126] in Indicates the true wear value, represents the predicted wear value of the model, Represents the average of the true wear values. The smaller the MSE and MAE values, the better the model performance. Conversely, the larger the R2-Score value, the better the model performance.

[0127] It should be noted that the fault classification model consists of a convolutional module, an adaptive average pooling layer, and a fully connected layer. Finally, the cross entropy loss function is used to measure the distance between the categories predicted by this model (WavePHF) and the actual distribution, which can be expressed as:

[0128]

[0129] K represents the number of categories, N represents the number of samples of the detection signal. P is the target distribution, and Q is the estimated distribution. Using MSE loss, it can be expressed as:

[0130]

[0131] Assumption Z i is the i-th detection signal in the sample of multiple detection signals, To effectively make the model focus on the fault recovery of important sensors, the total loss function of the joint framework can be expressed as:

[0132]

[0133] Among them, g i is the predicted i-th detection signal.

[0134] In order to verify the effectiveness of the proposed model and each module, an ablation experiment was set up and five network models were experimentally compared. The terms in the table are explained as follows: WavePHF-F is the encoder part removed, and it can only handle fault diagnosis tasks. WavePHF-C is the fault classifier removed, and it can only handle the correction task of the detection signal. WaveNet-AFW is all AFD-Blocks removed, and only AFW-Block is retained, which can complete fault diagnosis and correction at the same time. WaveNet-AFD is all AFW-Blocks removed, and only AFD-Block is retained, which can complete fault diagnosis and correction at the same time. WaveNet is only the backbone network based on wavelet Powered, and both AFW-Block and AFD-Block are removed, which can complete fault diagnosis and correction at the same time.

[0135] Refer to Table 1, which shows the results of the ablation experiment.

[0136] Table 1

[0137]

[0138]

[0139] Table 1 shows the results of fault diagnosis and correction of detection signals for the accelerator pedal sensor using WavePHF, WavePHF-F, WavePHF-C, WaveNet-AFW, WaveNet-AFD, and WaveNet. Table 1 shows that both AFD-Block and AFW-Block effectively improve the model's fault diagnosis and correction performance. For example, the fault diagnosis accuracy of WaveNet-AFW and WaveNet-AFD is 97.15% and 96.71%, respectively, while WaveNet's diagnostic accuracy is only 96.10%. Furthermore, the mean square error (MSE) of WaveNet-AFW and WaveNet-AFD improves from 0.0037 for WaveNet to 0.0028 and 0.0033, respectively. This strongly demonstrates that the hierarchical frequency filtering mechanism can effectively improve the model's performance and enhance its frequency learning capabilities. Furthermore, the joint learning architecture significantly improves model performance by sharing deep features between the two tasks. For example, WavePHF achieved a fault diagnosis performance of 97.66%, while WavePHF-F achieved only 97.05%. For fault correction tasks, the performance gains brought by the joint learning architecture are very significant. WavePHF achieved an MSE of 0.0025, while WavePHF-C only achieved an MSE of 0.0043. This demonstrates that introducing the fault diagnosis task significantly improves the model's ability to identify and remove fault information. Finally, we can see that the proposed WavePHF achieved very good fault diagnosis performance, with an R2-Score of 0.9677, demonstrating that the proposed method can accurately identify and correct faults, ensuring the safe operation of autonomous vehicles.

[0140] Furthermore, existing fault diagnosis and reconstruction models were compared to demonstrate the superior performance of the proposed method. For the fault diagnosis task, the proposed model (WavePHF) was compared with six mainstream fault diagnosis models: Multi-GRU, Multi-LSTM, MA1DCNN, ResNet18, WDCNN, WenCNN, Multi-GRU, and Multi-LSTM. Multi-GRU and Multi-LSTM use different recurrent neural units to learn sequence information, followed by fully connected layers for fault identification. To accommodate multi-sensor data processing, three recurrent units were adopted, with 128 hidden layers. MA1DCNN is a diagnostic architecture based on a multi-attention mechanism, employing multiple hybrid attention mechanisms at different levels to extract fault features. ResNet is a classic multi-classification network architecture. Since the encoder and fault classifier of the proposed WavePHF only use six convolutional modules in total, a ResNet18 network architecture was used in this experiment for a fair comparison. WDCNN is a diagnostic model based on wide convolutions, which capture a wide range of information. WenCNN is a 2DCNN-based architecture that converts a one-dimensional signal into a two-dimensional map. The 2DCNN is then used to learn the spatial representation of the signal. Tables 2 and 3 show the comparative results of the fault classification and diagnosis performance of each model, while Table 2 shows the comparative results of the correction performance of each model for detection signals.

[0141] Table 2

[0142]

[0143] Table 3

[0144] Method WavePHF CAE ATT-CAE U-Net ATT-U-Net MSE 0.0025±0.0002 0.0063±0.0003 0.0057±0.0004 0.0053±0.0002 0.0036±0.0003 MAE 0.0276±0.0017 0.0432±0.0008 0.0419±0.0014 0.0385±0.0016 0.0317±0.0013 <![CDATA[R 2 -Score]]> 0.9677±0.0038 0.9122±0.0049 0.9242±0.0057 0.9310±0.0051 0.9512±0.0052

[0145] Since there is currently little research on fault correction for autonomous vehicle sensors based on deep learning models, this experiment compares the proposed WavePHF with four mainstream reconstruction-based models: CAE, ATT-CAE, U-Net, and ATT-U-Net. CAE is a convolutional autoencoder, and like the WavePHF model, its encoder and decoder each utilize five convolutional modules. ATT-CAE is an attention-based convolutional autoencoder that builds on the CAE model by introducing an attention mechanism to improve feature learning. U-Net is an excellent reconstruction model that introduces skip links between the encoder and decoder to achieve optimal reconstruction. ATT-U-Net is an attention-based U-Net that uses attention to optimize features, resulting in better reconstruction performance. Table II shows that the proposed WavePHF model achieves the best sensor fault identification performance. For example, WavePHM achieves a sensor fault identification accuracy of 97.66%. However, Multi-LSTM achieved only 95.41% diagnostic accuracy; ResNet18 only 95.76%; and WenCNN only 94.35%. This demonstrates that by deeply integrating the advantages of wavelet transform and CNN frequency decomposition and feature learning, the model's frequency learning capability can be significantly improved, thereby extracting more valuable fault features from the frequency level. As shown in Table 3, WavePHF clearly achieved the best fault correction performance and significantly outperformed the other four compared methods. For example, WavePHF achieved an MSE of 0.0025 and an R²-score of 0.9677, while ATT-CAE's MSE and R²-score were only 0.0057 and 0.9242, respectively; and ATT-U-Net's MSE and R²-score were only 0.0036 and 0.9512, respectively. This demonstrates that WavePHF has excellent reconstruction performance, accurately removing fault information from sensor detection signals and restoring normal sensor detection signals. It can be seen that the proposed WavePHF can not only detect the abnormal conditions of the sensor in time, but also correct the abnormal conditions to ensure the safety of the vehicle.

[0146] To further analyze the fault correction performance of the model, refer to Figure 11 , Figure 11An experimental analysis result diagram provided for an embodiment of the present application shows the visualization results of the correction of the detection signals of the three models WavePHF, ATT-U-Net, and ATT-CAE. It can be seen that the fault correction effect of the present model WavePHF is the best. It basically removes all fault information, and the predicted sensor detection signal is basically consistent with the pure detection signal. For ATT-U-Net, although its predicted signal basically removes the fault information, there is a partial deviation between it and the real pure signal. For ATT-CAE, the deviation between its predicted signal and the real pure signal is more significant, so its correction performance is relatively poor. The above results also illustrate from another level that the proposed WavePHF has considerable fault correction capabilities and can recover from abnormal situations to ensure vehicle safety.

[0147] It should be understood that, although the steps in the flowcharts of the above embodiments are shown in sequence as indicated by the arrows, these steps are not necessarily performed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be performed in other orders. Moreover, at least a portion of the steps in the flowcharts of the above embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily performed at the same time, but can be performed at different times. The execution order of these steps or stages is not necessarily to be performed in sequence, but can be performed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0148] Based on the same inventive concept, embodiments of the present application also provide a detection signal correction device for a sensor on an autonomous vehicle, which is used to implement the aforementioned detection signal correction method for a sensor on an autonomous vehicle. The solution provided by this device is similar to the solution described in the aforementioned method. Therefore, the specific limitations of the following embodiments of the detection signal correction device for one or more autonomous vehicle sensors can be found in the aforementioned definitions of the detection signal correction method for an autonomous vehicle sensor, and will not be further elaborated here.

[0149] In one embodiment, Figure 12 This is a structural block diagram of a detection signal correction device for a sensor on an autonomous vehicle provided in an embodiment of the present application. The detection signal correction device 1200 for a sensor on an autonomous vehicle includes: an acquisition module 1201, a first determination module 1202, a second determination module 1203, a third determination module 1204, and a correction module 1205, wherein:

[0150] The acquisition module 1201 is configured to acquire a detection signal collected by a sensor and perform feature extraction on the detection signal to obtain a first feature.

[0151] The first determining module 1202 is configured to perform discrete wavelet transform processing on the first feature to obtain a first high-frequency feature and a first low-frequency feature; the frequency of the first high-frequency feature is greater than or equal to a preset frequency threshold, and the frequency of the first low-frequency feature is less than the preset frequency threshold.

[0152] The second determining module 1203 is configured to process the first low-frequency feature using an adaptive frequency weighting algorithm to obtain a second feature.

[0153] The third determining module 1204 is configured to determine the target low-frequency feature according to the second feature, a discrete wavelet transform algorithm, and an inverse discrete wavelet transform algorithm.

[0154] The correction module 1205 is configured to correct the detection signal according to the first high-frequency feature, the target low-frequency feature, and an inverse discrete wavelet transform algorithm.

[0155] In one embodiment, the third determining module 1204 includes:

[0156] The first determining unit is used to perform discrete wavelet transform on the second feature using a discrete wavelet transform algorithm to obtain a second high-frequency feature and a second low-frequency feature; the frequency of the second high-frequency feature is greater than or equal to a preset frequency threshold, and the frequency of the second low-frequency feature is less than the preset frequency threshold.

[0157] The second determining unit is configured to determine a third high frequency feature according to the adaptive frequency discarding algorithm and the second high frequency feature.

[0158] The third determining unit is configured to determine the third low-frequency feature according to the second low-frequency feature, a discrete wavelet transform algorithm, and an inverse discrete wavelet transform algorithm.

[0159] The fourth determining unit is configured to determine the target low-frequency feature according to the third high-frequency feature, the third low-frequency feature, and an inverse discrete wavelet transform algorithm.

[0160] In one embodiment, the third determining unit includes:

[0161] The first determining subunit is configured to process the second low-frequency feature using an adaptive frequency weighting algorithm to obtain a third feature.

[0162] The first processing subunit is configured to perform discrete wavelet transform on the third feature to obtain a fourth high-frequency feature and a fourth low-frequency feature.

[0163] The second processing subunit is configured to process the fourth high-frequency feature by using an adaptive frequency discarding algorithm to obtain a processed fourth high-frequency feature.

[0164] The second determining subunit is configured to determine the third low-frequency feature according to the fourth low-frequency feature, the processed fourth high-frequency feature, and an inverse discrete wavelet transform algorithm.

[0165] In one embodiment, the fourth determining unit includes:

[0166] The third processing subunit is used to process the third high-frequency feature and the third low-frequency feature by using an inverse discrete wavelet transform algorithm to obtain a fourth feature.

[0167] The third determining subunit is configured to perform convolution processing on the fourth feature to obtain a target low-frequency feature.

[0168] In one embodiment, the correction module 1205 includes:

[0169] The first processing unit is configured to process the first high-frequency feature by using an adaptive frequency discarding algorithm to obtain a processed first high-frequency feature.

[0170] The second processing unit is used to process the processed first high-frequency feature and the target low-frequency feature by using an inverse discrete wavelet transform algorithm to obtain a fifth feature.

[0171] The third processing unit is configured to perform convolution processing on the fifth feature to correct the detection signal.

[0172] In one embodiment, the detection signal correction device 1200 for the sensor on the autonomous driving vehicle further includes:

[0173] The fourth determination module is configured to input the fourth low-frequency feature into the fault classification model to obtain a fault category of the sensor.

[0174] Each module in the aforementioned sensor detection signal correction device for an autonomous vehicle can be implemented in whole or in part through software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in hardware form, or stored in a computer device memory in software form, so that the processor can call and execute the corresponding operations of each module.

[0175] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 13As shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected via a system bus, and the communication interface, the display unit and the input device are connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a detection signal correction method for a sensor on an autonomous vehicle is implemented. The display unit of the computer device is used to form a visually visible image, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device casing, or an external keyboard, touchpad or mouse.

[0176] Those skilled in the art will understand that Figure 13 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0177] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0178] Acquire a detection signal collected by a sensor, and perform feature extraction on the detection signal to obtain a first feature;

[0179] Performing discrete wavelet transform processing on the first feature to obtain a first high-frequency feature and a first low-frequency feature; the frequency of the first high-frequency feature is greater than or equal to a preset frequency threshold, and the frequency of the first low-frequency feature is less than the preset frequency threshold;

[0180] Processing the first low-frequency feature using an adaptive frequency weighting algorithm to obtain a second feature;

[0181] Determine the target low-frequency feature according to the second feature, a discrete wavelet transform algorithm, and an inverse discrete wavelet transform algorithm;

[0182] The detection signal is corrected according to the first high-frequency feature, the target low-frequency feature and an inverse discrete wavelet transform algorithm.

[0183] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0184] Performing a discrete wavelet transform on the second feature using a discrete wavelet transform algorithm to obtain a second high-frequency feature and a second low-frequency feature; the frequency of the second high-frequency feature is greater than or equal to a preset frequency threshold, and the frequency of the second low-frequency feature is less than the preset frequency threshold;

[0185] determining a third high frequency feature based on an adaptive frequency discarding algorithm and the second high frequency feature;

[0186] determining a third low-frequency feature based on the second low-frequency feature, a discrete wavelet transform algorithm, and an inverse discrete wavelet transform algorithm;

[0187] The target low-frequency feature is determined according to the third high-frequency feature, the third low-frequency feature and an inverse discrete wavelet transform algorithm.

[0188] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0189] The second low-frequency feature is processed using an adaptive frequency weighting algorithm to obtain a third feature;

[0190] Performing discrete wavelet transform on the third feature to obtain a fourth high-frequency feature and a fourth low-frequency feature;

[0191] Processing the fourth high-frequency feature using an adaptive frequency discarding algorithm to obtain a processed fourth high-frequency feature;

[0192] The third low-frequency feature is determined according to the fourth low-frequency feature, the processed fourth high-frequency feature, and an inverse discrete wavelet transform algorithm.

[0193] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0194] The third high-frequency feature and the third low-frequency feature are processed using an inverse discrete wavelet transform algorithm to obtain a fourth feature;

[0195] The fourth feature is convolved to obtain the target low-frequency feature.

[0196] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0197] Processing the first high-frequency feature using an adaptive frequency discarding algorithm to obtain a processed first high-frequency feature;

[0198] The processed first high-frequency feature and the target low-frequency feature are processed using an inverse discrete wavelet transform algorithm to obtain a fifth feature;

[0199] The fifth feature is convolved to correct the detection signal.

[0200] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0201] The fourth low-frequency feature is input into the fault classification model to obtain the fault category of the sensor.

[0202] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0203] Acquire a detection signal collected by a sensor, and perform feature extraction on the detection signal to obtain a first feature;

[0204] Performing discrete wavelet transform processing on the first feature to obtain a first high-frequency feature and a first low-frequency feature; the frequency of the first high-frequency feature is greater than or equal to a preset frequency threshold, and the frequency of the first low-frequency feature is less than the preset frequency threshold;

[0205] Processing the first low-frequency feature using an adaptive frequency weighting algorithm to obtain a second feature;

[0206] Determine the target low-frequency feature according to the second feature, a discrete wavelet transform algorithm, and an inverse discrete wavelet transform algorithm;

[0207] The detection signal is corrected according to the first high-frequency feature, the target low-frequency feature and an inverse discrete wavelet transform algorithm.

[0208] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0209] Performing a discrete wavelet transform on the second feature using a discrete wavelet transform algorithm to obtain a second high-frequency feature and a second low-frequency feature; the frequency of the second high-frequency feature is greater than or equal to a preset frequency threshold, and the frequency of the second low-frequency feature is less than the preset frequency threshold;

[0210] determining a third high frequency feature based on an adaptive frequency discarding algorithm and the second high frequency feature;

[0211] determining a third low-frequency feature based on the second low-frequency feature, a discrete wavelet transform algorithm, and an inverse discrete wavelet transform algorithm;

[0212] The target low-frequency feature is determined according to the third high-frequency feature, the third low-frequency feature and an inverse discrete wavelet transform algorithm.

[0213] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0214] The second low-frequency feature is processed using an adaptive frequency weighting algorithm to obtain a third feature;

[0215] Performing discrete wavelet transform on the third feature to obtain a fourth high-frequency feature and a fourth low-frequency feature;

[0216] Processing the fourth high-frequency feature using an adaptive frequency discarding algorithm to obtain a processed fourth high-frequency feature;

[0217] The third low-frequency feature is determined according to the fourth low-frequency feature, the processed fourth high-frequency feature, and an inverse discrete wavelet transform algorithm.

[0218] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0219] The third high-frequency feature and the third low-frequency feature are processed using an inverse discrete wavelet transform algorithm to obtain a fourth feature;

[0220] The fourth feature is convolved to obtain the target low-frequency feature.

[0221] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0222] Processing the first high-frequency feature using an adaptive frequency discarding algorithm to obtain a processed first high-frequency feature;

[0223] The processed first high-frequency feature and the target low-frequency feature are processed using an inverse discrete wavelet transform algorithm to obtain a fifth feature;

[0224] The fifth feature is convolved to correct the detection signal.

[0225] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0226] The fourth low-frequency feature is input into the fault classification model to obtain the fault category of the sensor.

[0227] In one embodiment, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the following steps:

[0228] Acquire a detection signal collected by a sensor, and perform feature extraction on the detection signal to obtain a first feature;

[0229] Performing discrete wavelet transform processing on the first feature to obtain a first high-frequency feature and a first low-frequency feature; the frequency of the first high-frequency feature is greater than or equal to a preset frequency threshold, and the frequency of the first low-frequency feature is less than the preset frequency threshold;

[0230] Processing the first low-frequency feature using an adaptive frequency weighting algorithm to obtain a second feature;

[0231] Determine the target low-frequency feature according to the second feature, a discrete wavelet transform algorithm, and an inverse discrete wavelet transform algorithm;

[0232] The detection signal is corrected according to the first high-frequency feature, the target low-frequency feature and an inverse discrete wavelet transform algorithm.

[0233] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0234] Performing a discrete wavelet transform on the second feature using a discrete wavelet transform algorithm to obtain a second high-frequency feature and a second low-frequency feature; the frequency of the second high-frequency feature is greater than or equal to a preset frequency threshold, and the frequency of the second low-frequency feature is less than the preset frequency threshold;

[0235] determining a third high frequency feature based on an adaptive frequency discarding algorithm and the second high frequency feature;

[0236] determining a third low-frequency feature based on the second low-frequency feature, a discrete wavelet transform algorithm, and an inverse discrete wavelet transform algorithm;

[0237] The target low-frequency feature is determined according to the third high-frequency feature, the third low-frequency feature and an inverse discrete wavelet transform algorithm.

[0238] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0239] The second low-frequency feature is processed using an adaptive frequency weighting algorithm to obtain a third feature;

[0240] Performing discrete wavelet transform on the third feature to obtain a fourth high-frequency feature and a fourth low-frequency feature;

[0241] Processing the fourth high-frequency feature using an adaptive frequency discarding algorithm to obtain a processed fourth high-frequency feature;

[0242] The third low-frequency feature is determined according to the fourth low-frequency feature, the processed fourth high-frequency feature, and an inverse discrete wavelet transform algorithm.

[0243] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0244] The third high-frequency feature and the third low-frequency feature are processed using an inverse discrete wavelet transform algorithm to obtain a fourth feature;

[0245] The fourth feature is convolved to obtain the target low-frequency feature.

[0246] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0247] Processing the first high-frequency feature using an adaptive frequency discarding algorithm to obtain a processed first high-frequency feature;

[0248] The processed first high-frequency feature and the target low-frequency feature are processed using an inverse discrete wavelet transform algorithm to obtain a fifth feature;

[0249] The fifth feature is convolved to correct the detection signal.

[0250] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0251] The fourth low-frequency feature is input into the fault classification model to obtain the fault category of the sensor.

[0252] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions.

[0253] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processor involved in the various embodiments provided herein may be, but are not limited to, a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic unit, a data processing logic unit based on quantum computing, and the like.

[0254] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0255] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A method for calibrating a detection signal of a sensor on an autonomous driving vehicle, characterized in that: The method comprises: Acquire a detection signal collected by the sensor, and perform feature extraction on the detection signal to obtain a first feature; Performing discrete wavelet transform processing on the first feature to obtain a first high-frequency feature and a first low-frequency feature; the frequency of the first high-frequency feature is greater than or equal to a preset frequency threshold, and the frequency of the first low-frequency feature is less than the preset frequency threshold; Processing the first low-frequency feature using an adaptive frequency weighting algorithm to obtain a second feature; Determine the target low-frequency feature according to the second feature, a discrete wavelet transform algorithm and an inverse discrete wavelet transform algorithm; Correcting the detection signal according to the first high-frequency feature, the target low-frequency feature and an inverse discrete wavelet transform algorithm; The step of determining the target low-frequency feature according to the second feature, a discrete wavelet transform algorithm, and an inverse discrete wavelet transform algorithm includes: Performing discrete wavelet transform on the second feature using the discrete wavelet transform algorithm to obtain a second high-frequency feature and a second low-frequency feature; the frequency of the second high-frequency feature is greater than or equal to the preset frequency threshold, and the frequency of the second low-frequency feature is less than the preset frequency threshold; determining a third high frequency feature according to an adaptive frequency discarding algorithm and the second high frequency feature; Determining a third low-frequency feature according to the second low-frequency feature, the discrete wavelet transform algorithm and the inverse discrete wavelet transform algorithm; Determining the target low-frequency feature according to the third high-frequency feature, the third low-frequency feature and the inverse discrete wavelet transform algorithm; The step of determining a third low-frequency feature according to the second low-frequency feature, the discrete wavelet transform algorithm, and the inverse discrete wavelet transform algorithm comprises: Processing the second low-frequency feature by using an adaptive frequency weighting algorithm to obtain a third feature; Performing discrete wavelet transform processing on the third feature to obtain a fourth high-frequency feature and a fourth low-frequency feature; Processing the fourth high-frequency feature using the adaptive frequency discarding algorithm to obtain a processed fourth high-frequency feature; Determining the third low-frequency feature according to the fourth low-frequency feature, the processed fourth high-frequency feature, and an inverse discrete wavelet transform algorithm; The step of determining the target low-frequency feature according to the third high-frequency feature, the third low-frequency feature and the inverse discrete wavelet transform algorithm comprises: Processing the third high-frequency feature and the third low-frequency feature using the inverse discrete wavelet transform algorithm to obtain a fourth feature; The fourth feature is convolved to obtain the target low-frequency feature.

2. The method according to claim 1, characterized in that The correcting the detection signal according to the first high-frequency feature, the target low-frequency feature and an inverse discrete wavelet transform algorithm comprises: Processing the first high-frequency feature using an adaptive frequency discarding algorithm to obtain a processed first high-frequency feature; Using the inverse discrete wavelet transform algorithm to process the processed first high-frequency feature and the target low-frequency feature to obtain a fifth feature; The fifth feature is subjected to convolution processing to correct the detection signal.

3. The method according to claim 1, characterized in that The method further comprises: The fourth low-frequency feature is input into a fault classification model to obtain a fault category of the sensor.

4. A detection signal correction device for a sensor on an autonomous driving vehicle, characterized in that: The device comprises: an acquisition module, used for acquiring the detection signal collected by the sensor, and performing feature extraction on the detection signal to obtain a first feature; A first determination module is used to perform discrete wavelet transform processing on the first feature to obtain a first high-frequency feature and a first low-frequency feature; the frequency of the first high-frequency feature is greater than or equal to a preset frequency threshold, and the frequency of the first low-frequency feature is less than the preset frequency threshold; A second determination module, configured to process the first low-frequency feature using an adaptive frequency weighting algorithm to obtain a second feature; A third determination module is used to determine the target low-frequency feature according to the second feature, a discrete wavelet transform algorithm and an inverse discrete wavelet transform algorithm; A correction module, used for correcting the detection signal according to the first high-frequency feature, the target low-frequency feature and an inverse discrete wavelet transform algorithm; Wherein, the third determination module includes: A first determining unit is configured to perform a discrete wavelet transform on the second feature using the discrete wavelet transform algorithm to obtain a second high-frequency feature and a second low-frequency feature; the frequency of the second high-frequency feature is greater than or equal to the preset frequency threshold, and the frequency of the second low-frequency feature is less than the preset frequency threshold; A second determining unit, configured to determine a third high frequency feature according to an adaptive frequency discarding algorithm and the second high frequency feature; a third determining unit, configured to determine a third low-frequency feature according to the second low-frequency feature, the discrete wavelet transform algorithm and the inverse discrete wavelet transform algorithm; a fourth determining unit, configured to determine the target low-frequency feature according to the third high-frequency feature, the third low-frequency feature, and the inverse discrete wavelet transform algorithm; The third determining unit includes: A first determining subunit, configured to process the second low-frequency feature by using an adaptive frequency weighting algorithm to obtain a third feature; A first processing subunit is used to perform discrete wavelet transform processing on the third feature to obtain a fourth high-frequency feature and a fourth low-frequency feature; A second processing subunit is used to process the fourth high-frequency feature by using the adaptive frequency discarding algorithm to obtain a processed fourth high-frequency feature; A second determining subunit is used to determine the third low-frequency feature according to the fourth low-frequency feature, the processed fourth high-frequency feature and an inverse discrete wavelet transform algorithm; The fourth determining unit includes: A third processing subunit is used to process the third high-frequency feature and the third low-frequency feature by using the inverse discrete wavelet transform algorithm to obtain a fourth feature; The third determination subunit is used to perform convolution processing on the fourth feature to obtain the target low-frequency feature.

5. The device according to claim 4, characterized in that The correction module comprises: A first processing unit, configured to process the first high-frequency feature by using an adaptive frequency discarding algorithm to obtain a processed first high-frequency feature; A second processing unit is used to process the processed first high-frequency feature and the target low-frequency feature by using the inverse discrete wavelet transform algorithm to obtain a fifth feature; The third processing unit is used to perform convolution processing on the fifth feature to correct the detection signal.

6. The device according to claim 4, characterized in that The device also includes: The fourth determination module is used to input the fourth low-frequency feature into a fault classification model to obtain a fault category of the sensor.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 3 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 3 are implemented.

9. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 3 are implemented.

Images