Ground material identification method, device, electronic equipment and storage medium
By using the 1D-SNN model in intelligent vehicles combined with 1D-CNN and pulsed neuron technology, the local and global characteristics of ground materials are extracted, and the problem of low accuracy of traditional ground material recognition methods in diversity and complex environments is solved. Improves the adaptability and safety of smart vehicles.
Patent Information
- Application Number
- CN202410834192.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-26
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2044-06-26
AI Technical Summary
Traditional ground material recognition methods face the diversity of ground materials and the complexity of the environment, low recognition accuracy affects the adaptability and safe driving ability of smart vehicles.
Using a method based on the 1D-SNN model, local features of pressure change timing data are extracted through 1D-CNN, global features are extracted in combination with pulsed neurons, and ground material recognition is used to improve the accuracy and stability of the recognition.
When facing diverse and complex environments, the accuracy and stability of ground material recognition are significantly improved, and the adaptability and safe driving capabilities of smart vehicles are enhanced.
Smart Images

Figure CN118378153B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of data processing technology, and in particular to a ground material recognition method, device, electronic equipment and storage medium. Background Art
[0002] With the rapid development of intelligent vehicle technology, accurate environmental perception has become the key to improving autonomous driving capabilities and safety. In this regard, real-time and accurate recognition of ground materials is crucial to improving the adaptability and driving decision-making process of intelligent vehicles.
[0003] Traditional ground material recognition technology mainly obtains ground features through ultrasonic signals, optical signals, audio signals fed back from the ground or analyzes the impact of motion resistance on work efficiency, and then uses traditional discriminators to discriminate and analyze ground materials.
[0004] However, most traditional ground material recognition methods rely on a large number of signal processing and analysis steps. When faced with the diversity of ground materials and the complexity of the environment, the accuracy of ground material recognition is low. Summary of the invention
[0005] The present invention provides a ground material recognition method, device, electronic device and storage medium, which are used to solve the defect of low ground material recognition accuracy in the traditional ground material recognition method in the prior art when facing the diversity of ground materials and the complexity of the environment, and effectively improve the accuracy and stability of ground material recognition when facing the diversity of ground materials and the complexity of the environment, thereby enhancing the adaptability and safe driving ability of intelligent vehicles.
[0006] The present invention provides a ground material recognition method, comprising:
[0007] Collect the time series data of the pressure change of the bleed air of the ground material to be identified;
[0008] The pre-processed pressure change time series data is input into the trained 1D-SNN model to perform ground material recognition to obtain a recognition result;
[0009] Among them, the 1D-SNN model is constructed based on 1D-CNN and pulse neurons, the 1D-CNN is used to extract the local time series features of the pressure change time series data, and the pulse neurons are used to extract the global time series features of the pressure change time series data.
[0010] In some embodiments, the 1D-SNN model includes: at least one one-dimensional convolutional layer, a plurality of residual blocks, and a fully connected layer;
[0011] The signal input into the 1D-SNN model is output after passing through at least one one-dimensional convolutional layer, multiple residual blocks and a fully connected layer in sequence;
[0012] Each of the residual blocks includes: three one-dimensional convolutional layers, two batch normalization layers and two LIF spike neurons;
[0013] The signal input to the residual block is divided into two paths. One path of the signal passes through the one-dimensional convolution layer, the batch normalization layer, the LIF pulse neuron, the one-dimensional convolution layer and the batch normalization layer in sequence, and then is merged with the other path of the signal passing through the one-dimensional convolution layer and output through the LIF pulse neuron.
[0014] In some embodiments, the method further comprises:
[0015] Inputting the preprocessed sample time series data into the 1D-SNN model to obtain time series features;
[0016] According to the cross entropy loss function, the loss value corresponding to the time series feature is calculated;
[0017] Performing reverse iterative update on the 1D-SNN model using the loss value to update the loss value;
[0018] When the updated loss value is less than the target threshold, the trained 1D-SNN model is obtained.
[0019] In some embodiments, the reverse iterative updating of the 1D-SNN model using the loss value includes:
[0020] Determine a replacement gradient according to the loss value and the gradient function;
[0021] The 1D-SNN model is updated in reverse iteratively using the substitute gradient.
[0022] In some embodiments, before the preprocessed sample time series data is input into the 1D-SNN model, the method further includes:
[0023] Perform outlier identification and elimination processing on the sample time series data, and adjust the size of the processed sample time series data to obtain the target sample time series data;
[0024] The target sample time series data is normalized to obtain the preprocessed sample time series data.
[0025] In some embodiments, the pressure variation time series data is obtained by a pressure sensor installed at a level with the bottom of the vehicle tire.
[0026] In some embodiments, the pressure sensor is a FSR402 ultra-thin resistive pressure sensor.
[0027] The present invention also provides a ground material identification device, comprising:
[0028] An acquisition module is used to collect the time series data of the pressure change of the bleed air of the ground material to be identified;
[0029] An identification module is used to input the pre-processed pressure change time series data into a trained 1D-SNN model to perform ground material identification and obtain an identification result;
[0030] Among them, the 1D-SNN model is constructed based on 1D-CNN and pulse neurons, the 1D-CNN is used to extract the local time series features of the pressure change time series data, and the pulse neurons are used to extract the global time series features of the pressure change time series data.
[0031] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, any of the above-mentioned ground material recognition methods is implemented.
[0032] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the ground material identification method described in any one of the above methods is implemented.
[0033] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the ground material recognition method described in any one of the above is implemented.
[0034] The ground material identification method, device, electronic device and storage medium provided by the present invention perform ground material identification by inputting the pressure change time series data of the air induction of the ground material to be identified into the 1D-SNN model. The 1D-CNN in the 1D-SNN model is responsible for extracting the local time series features of the pressure change time series data, and the pulse neuron in the 1D-SNN model is responsible for extracting the global time series features of the pressure change time series data, thereby obtaining more comprehensive time series features. In the face of the diversity of ground materials and the complexity of the environment, the accuracy and stability of ground material identification are effectively improved, thereby enhancing the adaptability and safe driving capabilities of intelligent vehicles. BRIEF DESCRIPTION OF THE DRAWINGS
[0035] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0036] Figure 1 It is a schematic diagram of the flow of the ground material identification method provided by the present invention;
[0037] Figure 2 It is a structural schematic diagram of the 1D-SNN model provided by the present invention;
[0038] Figure 3 is a schematic diagram of the structure of the residual block provided by the present invention;
[0039] Figure 4 It is a schematic diagram of the working principle of the 1D-SNN model provided by the present invention;
[0040] Figure 5 It is a schematic diagram of the process of training a 1D-SNN model provided by the present invention;
[0041] Figure 6 It is a structural schematic diagram of the ground material identification device provided by the present invention;
[0042] Figure 7 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION
[0043] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0044] Figure 1 is a flow chart of the ground material identification method provided by the present invention, such as Figure 1 As shown, the present invention provides a ground material recognition method, comprising the following steps:
[0045] Step 110, collecting time series data of pressure change of bleed air of the ground material to be identified;
[0046] Step 120, input the pre-processed pressure change time series data into the trained 1D-SNN model to perform ground material recognition to obtain a recognition result;
[0047] Among them, the 1D-SNN model is built based on 1D-CNN and pulse neurons. 1D-CNN is used to extract the local time series features of pressure change time series data, and pulse neurons are used to extract the global time series features of pressure change time series data.
[0048] Specifically, the hardness, texture and other physical characteristics of the ground material can be reflected by the pressure change data of the ground material air entrainment. Therefore, the pressure change time series data of the ground material air entrainment to be identified is collected. The pressure change time series data comes from the test of various common road surfaces, which is used to analyze the impact of different ground materials on data collection.
[0049] The collected pressure change time series data is preprocessed, such as data normalization, outlier removal, etc., to obtain preprocessed pressure change time series data.
[0050] The 1D-SNN model is a network structure that combines a spiking neural network (SNN) and a one-dimensional convolutional neural network (1D-CNN). The bio-simulation characteristics of SNNs enable them to maintain efficient and stable working performance in complex and dynamically changing environments, significantly improving the environmental adaptability of the system. In addition, 1D-SNN has excellent time series data extraction capabilities, and can capture subtle differences between ground materials from continuous signal data, improving recognition accuracy.
[0051] The preprocessed pressure change time series data is input into the trained 1D-SNN model for ground material recognition. The 1D-SNN model includes 1D-CNN and pulse neurons. Among them, 1D-CNN extracts the local time series features of the pressure change time series data, and the pulse neurons extract the global time series features of the pressure change time series data, thereby obtaining the characteristics of the ground material time series data more comprehensively and improving the accuracy of ground material recognition.
[0052] The ground material recognition method provided by the present invention performs ground material recognition by inputting the pressure change time series data of the air induction of the ground material to be identified into the 1D-SNN model. The 1D-CNN in the 1D-SNN model is responsible for extracting the local time series features of the pressure change time series data, and the pulse neurons in the 1D-SNN model are responsible for extracting the global time series features of the pressure change time series data, thereby obtaining more comprehensive time series features. In the face of the diversity of ground materials and the complexity of the environment, the accuracy and stability of ground material recognition are effectively improved, thereby enhancing the adaptability and safe driving capabilities of intelligent vehicles.
[0053] In some embodiments, the pressure variation time series data is acquired by a pressure sensor installed at a level with the bottom of the vehicle tire.
[0054] Specifically, the pressure sensor installed at a position flush with the bottom of the vehicle tire includes a pressure sensor directly installed on the outer surface of the vehicle tire.
[0055] In some embodiments, the pressure sensor is a FSR402 ultra-thin resistive pressure sensor.
[0056] Specifically, the FSR402 ultra-thin resistive pressure sensor is made of conductive polymers and conductive films. It has the advantages of high sensitivity, ultra-thin design, high durability, low power consumption, fast response and easy integration. It can accurately, efficiently and low-energy-consumption identify ground materials in real time.
[0057] When pressure is applied to the surface of the FSR402 sensor, the conductive polymer layer inside it compresses, causing the contact area between it and the conductive film to change. This change in contact area causes the overall resistance value of the sensor to change.
[0058] In some embodiments, the 1D-SNN model includes: at least one one-dimensional convolutional layer, a plurality of residual blocks, and a fully connected layer;
[0059] The signal input into the 1D-SNN model is output after passing through at least one one-dimensional convolutional layer, multiple residual blocks and a fully connected layer in sequence;
[0060] Each residual block consists of three one-dimensional convolutional layers, two batch normalization layers, and two LIF spike neurons.
[0061] The signal of the input residual block is divided into two paths. One path passes through the one-dimensional convolution layer, batch normalization layer, LIF spike neuron, one-dimensional convolution layer and batch normalization layer in sequence, and then merges with the signal of the other path passing through the one-dimensional convolution layer and is output through the LIF spike neuron.
[0062] Specifically, Figure 2 It is a structural schematic diagram of the 1D-SNN model provided by the present invention, such as Figure 2 As shown in the figure, the 1D-SNN model includes at least one one-dimensional convolutional layer (1DConv), multiple residual blocks and a fully connected layer. The signal input to the 1D-SNN model is output after passing through at least one one-dimensional convolutional layer, multiple residual blocks and a fully connected layer in sequence. The one-dimensional convolutional layer here can adapt to one-dimensional time series data and extract local time series features of time series data.
[0063] Each residual block in the 1D-SNN model can perform image processing operations for multiple time steps. For example, after the residual block performs data processing for 4 time steps, the outputs of the 4 time steps are merged through the Add operation to obtain comprehensive features, realize comprehensive feature processing, and then input the comprehensive features into the fully connected layer. The number of neurons output by the fully connected layer is the number of categories (C).
[0064] Figure 3 is a schematic diagram of the structure of the residual block provided by the present invention, such as Figure 3As shown, each residual block includes: three one-dimensional convolutional layers, two batch normalization layers (BatchNorm), and two leaky integrate-and-fire (LIF) spike neurons.
[0065] The signal of the input residual block is divided into two paths. One path passes through the one-dimensional convolution layer, batch normalization layer, LIF spike neuron, one-dimensional convolution layer and batch normalization layer in sequence, and then merges with the signal of the other path passing through the one-dimensional convolution layer and is output through the LIF spike neuron.
[0066] In the right branch of the residual block, the one-dimensional convolution layer is used to extract the local time series features of the time series data, the BN layer is used to accelerate the convergence of the model, and the LIF spike neuron is used to extract the global time series features of the time series data. In the left branch of the residual block, the one-dimensional convolution layer is used to prevent the gradient from disappearing / diverging due to the formation of an overly deep network.
[0067] The activation function of each residual block is replaced by LIF spike neurons for data nonlinear operations. The output of each one-dimensional convolutional layer in the residual block is fed into the LIF spike neuron and converted into a spike sequence. This process not only effectively extracts global temporal features, but also significantly reduces energy consumption.
[0068] The current equation of the LIF spiking neuron is expressed as:
[0069]
[0070] In the formula, is the current passing through the LIF neuron model at time t, is the current through the resistor R, is the current passing through capacitor C.
[0071] According to the above formula, we can deduce:
[0072]
[0073] In the formula, is the voltage across the LIF neuron model at time t, is the resting potential, is the membrane potential at t = 0, is the membrane time constant, , is the resistance value of resistor R, is the capacitance value of capacitor C.
[0074] The one-dimensional convolution layer outside the residual block is regarded as the first convolution kernel, and the one-dimensional convolution layer inside the residual block is regarded as the second convolution kernel. The sizes of the first convolution kernel and the second convolution kernel can be selected according to user needs. For example, the first convolution kernel is 5 5 convolution kernels, the second convolution kernel is 3 3 convolution kernels and 1 1 convolution.
[0075] For example, the 1D-SNN model includes a 5 5 convolution kernels and three residual blocks can ensure the effective transfer of key features, where each residual block consists of three 3 The network is composed of 3 convolution kernels and the activation function uses LIF neurons, which can balance biological rationality and practicality.
[0076] Figure 4 It is a schematic diagram of the working principle of the 1D-SNN model provided by the present invention, such as Figure 4 As shown in Figure 2, when the number of residual blocks N=3, the process of obtaining the comprehensive features is as follows:
[0077] (1) The pressure change time series data is input into the first residual block of the 1D-SNN. The data passes through the one-dimensional convolutional layer, the BatchNorm layer, and the LIF pulse neuron in sequence to obtain the first pulse output.
[0078] (2) The first pulse output is used as the input of the second residual block. The data passes through the one-dimensional convolution layer, BatchNorm layer, and LIF pulse neuron in sequence, and then the second pulse output is obtained. The first one-dimensional convolution layer of the second residual block will double the number of data channels.
[0079] (3) The second pulse output is used as the input of the third residual block. The data passes through the convolution layer, BatchNorm layer, and LIF pulse neuron in sequence, and then the third pulse output is obtained. A basic block includes two residual blocks. The first one-dimensional convolution layer of the third residual block will double the number of channels of the data.
[0080] (4) Merge the data at all time steps to obtain comprehensive features.
[0081] In some embodiments, the ground material recognition method provided by the present invention further includes:
[0082] Input the preprocessed sample time series data into the 1D-SNN model to obtain time series features;
[0083] According to the cross entropy loss function, calculate the loss value corresponding to the time series feature;
[0084] Use the loss value to perform reverse iterative update on the 1D-SNN model to update the loss value;
[0085] When the updated loss value is less than the target threshold, the trained 1D-SNN model is obtained.
[0086] Specifically, the preprocessed sample time series data (i.e., the pressure change sample time series data of ground material bleed air) is input into the 1D-SNN model. The 1D convolution in the 1D-SNN is responsible for extracting the local time series features of the preprocessed sample time series data, and the pulse neuron in the 1D-SNN is responsible for extracting the global time series features of the preprocessed sample time series data, so as to obtain more comprehensive time series features and output the recognition results corresponding to the time series features.
[0087] The loss value between the recognition result obtained by calculating the cross entropy loss function and the true recognition result corresponding to the sample time series data. The expression of the cross entropy loss function is as follows:
[0088]
[0089] In the formula, is the loss value, y is the true recognition result, 0 or 1, and p is the probability that the recognition result output by the 1D-SNN model is the true recognition result.
[0090] The 1D-SNN model is reversely iterated and updated using the loss value, and then the updated 1D-SNN model is used to update the loss value. The iterative update is continued, and when the updated loss value is less than the target threshold, a trained 1D-SNN model is obtained.
[0091] The ground material recognition method provided by the present invention obtains more comprehensive time series features by inputting the preprocessed sample time series data into the 1D-SNN model, and calculates the loss value corresponding to the time series features according to the cross entropy loss function, so as to reversely iterate and update the 1D-SNN model to obtain a trained 1D-SNN model, thereby improving the feature representation capability and further improving the classification performance of the 1D-SNN model.
[0092] In some embodiments, the 1D-SNN model is reversely iterated and updated using the loss value, including:
[0093] Determine the alternative gradient based on the loss value and the gradient function;
[0094] The surrogate gradient is used to perform reverse iterative updates on the 1D-SNN model.
[0095] Specifically, the expression of the gradient function is as follows:
[0096]
[0097] In the formula, is the gradient function, is the learning rate, is the loss value.
[0098] Substitute the loss value into the gradient function to obtain the substitute gradient, and use the substitute gradient to perform reverse iterative update on the 1D-SNN model.
[0099] The ground material recognition method provided by the present invention determines the substitute gradient through the loss value and the gradient function, and uses the substitute gradient to perform reverse iterative update on the 1D-SNN model, thereby further improving the classification performance of the 1D-SNN model.
[0100] In some embodiments, before the preprocessed sample time series data is input into the 1D-SNN model, the method further includes:
[0101] Perform outlier identification and elimination processing on the sample time series data, and adjust the size of the processed sample time series data to obtain the target sample time series data;
[0102] The target sample time series data is normalized to obtain preprocessed sample time series data.
[0103] Specifically, outlier identification and elimination can be performed on sample time series data through the following methods: manually setting thresholds, interquartile range (IQR) method, standard deviation method, sliding window statistics, etc. The following is a detailed introduction to IQR:
[0104] Arrange all sample time series data from small to large and divide them into four equal parts. The values at the three division points are called , and .
[0105] The expression for IQR is as follows:
[0106]
[0107] In the formula, is the interquartile range, is the value of the third split point, The value of the first split point.
[0108] The lower and upper bounds of the sample time series data can be obtained through IQR. When the sample time series data value exceeds the range of the lower and upper bounds, it will be judged as an outlier and removed. The expression of the lower bound of the sample time series data is as follows:
[0109]
[0110] In the formula, is the lower bound of the sample time series data, is the value of the first split point, is the interquartile range.
[0111] The expression for the upper bound of sample time series data is as follows:
[0112]
[0113] In the formula, is the upper bound of the sample time series data, is the value of the third split point, is the interquartile range.
[0114] The processed sample time series data is resized to be consistent with the data size allowed by the ID-SNN model to obtain the target sample time series data. The target sample time series data is normalized to obtain the preprocessed sample time series data.
[0115] The normalized expression is as follows:
[0116]
[0117] in, is the sample time series data after preprocessing, is the target sample time series data, is the minimum data value in the target sample time series data, It is the maximum data value in the target sample time series data.
[0118] The ground material recognition method provided by the present invention ensures the consistency of data by screening outliers and normalizing the sample time series data, thereby providing a more stable foundation for ID-SNN. These preprocessing steps help eliminate noise and outliers in the data, prevent abnormal data points from adversely affecting model training, and also speed up the learning process.
[0119] Figure 5 It is a schematic diagram of the process of training the 1D-SNN model provided by the present invention, such as Figure 5 As shown, training the 1D-SNN model includes:
[0120] (1) Initialize the network parameters (corresponding to neural network parameters) of the 1D-SNN model;
[0121] (2) The preprocessed sample time series data is input into the 1D-SNN model. The 1D-CNN extracts the local features of the time series data, and the pulse neurons are responsible for extracting the global features of the time series data, thereby obtaining a more comprehensive time series feature and outputting the recognition result corresponding to the time series feature;
[0122] (3) The loss value (corresponding error) between the recognition result calculated based on the cross entropy loss function and the true recognition result corresponding to the sample time series data;
[0123] (4) Perform back propagation update on the 1D-SNN model according to the loss value to update the loss value;
[0124] (5) If the updated loss value is greater than or equal to the target threshold, update the network parameters and return to step (2) to continue training; until the updated loss value is less than the target threshold, a trained 1D-SNN model is obtained.
[0125] The ground material recognition method provided by the present invention extracts the corresponding global output features by setting pulse neurons, and then builds characterization capabilities based on the local output feature information output by 1D-CNN to achieve more comprehensive features. On the one hand, it relies on pulse neurons to control power consumption, while at the same time improving the performance of the 1D-SNN model.
[0126] The ground material identification device provided by the present invention is described below. The ground material identification device described below and the ground material identification method described above can be referenced to each other.
[0127] Figure 6 is a schematic diagram of the structure of the ground material identification device provided by the present invention, such as Figure 6 As shown, the present invention provides a ground material recognition device, comprising:
[0128] The acquisition module 610 is used to acquire the pressure change time series data of the bleed air of the ground material to be identified;
[0129] An identification module 620 is used to input the pre-processed pressure change time series data into the trained 1D-SNN model to perform ground material identification and obtain an identification result;
[0130] Among them, the 1D-SNN model is constructed based on 1D-CNN and pulse neurons, the 1D-CNN is used to extract the local time series features of the pressure change time series data, and the pulse neurons are used to extract the global time series features of the pressure change time series data.
[0131] In some embodiments, the 1D-SNN model includes: at least one one-dimensional convolutional layer, a plurality of residual blocks, and a fully connected layer;
[0132] The signal input into the 1D-SNN model is output after passing through at least one one-dimensional convolutional layer, multiple residual blocks and a fully connected layer in sequence;
[0133] Each of the residual blocks includes: three one-dimensional convolutional layers, two batch normalization layers and two LIF spike neurons;
[0134] The signal input to the residual block is divided into two paths. One path of the signal passes through the one-dimensional convolution layer, the batch normalization layer, the LIF pulse neuron, the one-dimensional convolution layer and the batch normalization layer in sequence, and then is merged with the other path of the signal passing through the one-dimensional convolution layer and output through the LIF pulse neuron.
[0135] In some embodiments, the apparatus further comprises:
[0136] An input module, used to input the preprocessed sample time series data into the 1D-SNN model to obtain time series features;
[0137] A calculation module, used to calculate the loss value corresponding to the time series feature according to the cross entropy loss function;
[0138] An updating module, used for performing reverse iterative update on the 1D-SNN model using the loss value to update the loss value;
[0139] The obtaining module is used to obtain the trained 1D-SNN model when the updated loss value is less than the target threshold.
[0140] In some embodiments, the update module is specifically used to:
[0141] Determine a replacement gradient according to the loss value and the gradient function;
[0142] The 1D-SNN model is updated in reverse iteratively using the substitute gradient.
[0143] In some embodiments, the apparatus further comprises:
[0144] The first processing module is used to identify and remove outliers from the sample time series data, and adjust the size of the processed sample time series data to obtain target sample time series data;
[0145] The second processing module is used to perform normalization processing on the target sample time series data to obtain the pre-processed sample time series data.
[0146] In some embodiments, the pressure variation time series data is obtained by a pressure sensor installed at a level with the bottom of the vehicle tire.
[0147] In some embodiments, the pressure sensor is a FSR402 ultra-thin resistive pressure sensor.
[0148] Figure 7 is a schematic diagram of the structure of the electronic device provided by the present invention, such as Figure 7As shown, the electronic device may include: a processor 710, a communication interface 720, a memory 730 and a communication bus 740, wherein the processor 710, the communication interface 720 and the memory 730 communicate with each other through the communication bus 740. The processor 710 may call the logic instructions in the memory 730 to execute the ground material identification method, which includes: collecting the pressure change time series data of the bleed air of the ground material to be identified; inputting the pre-processed pressure change time series data into the trained 1D-SNN model to identify the ground material, and obtaining the identification result; wherein the 1D-SNN model is constructed based on 1D-CNN and pulse neurons, the 1D-CNN is used to extract the local time series features of the pressure change time series data, and the pulse neurons are used to extract the global time series features of the pressure change time series data.
[0149] In addition, the logic instructions in the above-mentioned memory 730 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.
[0150] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the ground material identification method provided by the above methods, and the method includes: collecting pressure change time series data of the air induction of the ground material to be identified; inputting the pre-processed pressure change time series data into a trained 1D-SNN model to perform ground material identification to obtain an identification result; wherein the 1D-SNN model is constructed based on 1D-CNN and pulse neurons, the 1D-CNN is used to extract the local time series features of the pressure change time series data, and the pulse neurons are used to extract the global time series features of the pressure change time series data.
[0151] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the ground material identification method provided by the above-mentioned methods, the method comprising: collecting pressure change time series data of the air induction of the ground material to be identified; inputting the pre-processed pressure change time series data into a trained 1D-SNN model to perform ground material identification to obtain an identification result; wherein the 1D-SNN model is constructed based on 1D-CNN and pulse neurons, the 1D-CNN is used to extract local time series features of the pressure change time series data, and the pulse neurons are used to extract global time series features of the pressure change time series data.
[0152] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.
[0153] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.
[0154] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for identifying ground material, characterized in that: include: Collect the time series data of the pressure change of the bleed air of the ground material to be identified; The pre-processed pressure change time series data is input into the trained 1D-SNN model to perform ground material recognition to obtain a recognition result; The 1D-SNN model is constructed based on 1D-CNN and pulse neurons, the 1D-CNN is used to extract the local time series features of the pressure change time series data, and the pulse neurons are used to extract the global time series features of the pressure change time series data; The 1D-SNN model includes: at least one one-dimensional convolutional layer, multiple residual blocks and a fully connected layer; The signal input into the 1D-SNN model is output after passing through at least one one-dimensional convolutional layer, multiple residual blocks and a fully connected layer in sequence; Each of the residual blocks includes: three one-dimensional convolutional layers, two batch normalization layers and two LIF spike neurons; The signal input to the residual block is divided into two paths. One path of the signal passes through the one-dimensional convolution layer, the batch normalization layer, the LIF pulse neuron, the one-dimensional convolution layer and the batch normalization layer in sequence, and then is merged with the other path of the signal passing through the one-dimensional convolution layer and output through the LIF pulse neuron.
2. The ground material recognition method according to claim 1, characterized in that: The method further comprises: Inputting the preprocessed sample time series data into the 1D-SNN model to obtain time series features; According to the cross entropy loss function, the loss value corresponding to the time series feature is calculated; Performing reverse iterative update on the 1D-SNN model using the loss value to update the loss value; When the updated loss value is less than the target threshold, the trained 1D-SNN model is obtained.
3. The ground material recognition method according to claim 2, characterized in that: The reverse iterative updating of the 1D-SNN model using the loss value includes: Determine a replacement gradient according to the loss value and the gradient function; The 1D-SNN model is updated in reverse iteratively using the substitute gradient.
4. The ground material recognition method according to claim 2, characterized in that: Before inputting the preprocessed sample time series data into the 1D-SNN model, it also includes: Perform outlier identification and elimination processing on the sample time series data, and adjust the size of the processed sample time series data to obtain the target sample time series data; The target sample time series data is normalized to obtain the preprocessed sample time series data.
5. The ground material recognition method according to claim 1, characterized in that: The pressure variation time series data is obtained through a pressure sensor installed at a position flush with the bottom of the vehicle tire.
6. The ground material recognition method according to claim 5, characterized in that: The pressure sensor is an FSR402 ultra-thin resistive pressure sensor.
7. A ground material recognition device, characterized in that: include: An acquisition module is used to collect the time series data of the pressure change of the bleed air of the ground material to be identified; An identification module is used to input the pre-processed pressure change time series data into a trained 1D-SNN model to perform ground material identification and obtain an identification result; The 1D-SNN model is constructed based on 1D-CNN and pulse neurons, the 1D-CNN is used to extract the local time series features of the pressure change time series data, and the pulse neurons are used to extract the global time series features of the pressure change time series data; The 1D-SNN model includes: at least one one-dimensional convolutional layer, multiple residual blocks and a fully connected layer; The signal input into the 1D-SNN model is output after passing through at least one one-dimensional convolutional layer, multiple residual blocks and a fully connected layer in sequence; Each of the residual blocks includes: three one-dimensional convolutional layers, two batch normalization layers and two LIF spike neurons; The signal input to the residual block is divided into two paths. One path of the signal passes through the one-dimensional convolution layer, the batch normalization layer, the LIF pulse neuron, the one-dimensional convolution layer and the batch normalization layer in sequence, and then is merged with the other path of the signal passing through the one-dimensional convolution layer and output through the LIF pulse neuron.
8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the ground material recognition method according to any one of claims 1 to 6 is implemented.
9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the ground material recognition method according to any one of claims 1 to 6 is implemented.
Citation Information
Patent Citations
Device and method for realizing object identification based on pressure-temperature sensing array
CN114201040A
Robot material identification method and system based on deep neural network
CN114519796A
Tactile attribute recognition method and device based on attention mechanism
CN116150684A
Ground material identification method, network training method and electronic equipment
CN117708599A
Computing device and electronic device
US20240013036A1
