CN115984822A - SNN traffic sign identification method based on space attention and related device - Google Patents
SNN traffic sign identification method based on space attention and related device Download PDFInfo
- Publication number
- CN115984822A CN115984822A CN202310156238.2A CN202310156238A CN115984822A CN 115984822 A CN115984822 A CN 115984822A CN 202310156238 A CN202310156238 A CN 202310156238A CN 115984822 A CN115984822 A CN 115984822A
- Authority
- CN
- China
- Prior art keywords
- snn
- preset
- traffic sign
- network model
- feature
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Image Analysis (AREA)
Abstract
The application discloses a SNN traffic sign identification method based on space attention and a related device, wherein the method comprises the following steps: inputting the target traffic sign image into a neuron coding layer of a preset SNN network model for coding to obtain a pulse coding pattern, wherein the neuron coding layer comprises soft reset LIF neurons; extracting an attention feature map in a pulse code map based on a spatial attention network layer in a preset SNN network model; performing feature analysis processing on the attention feature map according to a feature analysis network layer in a preset SNN network model to obtain a feature map to be recognized; and carrying out mark classification processing by adopting a classification network layer in a preset SNN network model according to the characteristic graph to be recognized to obtain the target traffic mark category. The method and the device can solve the technical problem that the prior art cannot meet the requirements of accuracy and calculation efficiency at the same time, and the traffic sign recognition effect is poor.
Description
Technical Field
The present application relates to the field of image processing technologies, and in particular, to a spatial attention-based SNN traffic sign recognition method and a related apparatus.
Background
Traffic sign recognition is one of the core technologies in the field of automatic driving, and plays an important role in safe driving and traffic management. Due to the complexity of traffic environment, the improvement of the accuracy and the real-time performance of a traffic sign recognition algorithm is a key problem to be solved in the practical application process, and most of the existing methods combine a deep learning technology and adopt a Convolutional Neural Network (CNN) to realize the breakthrough of a detection effect, so that the high-precision requirement can be basically met.
However, the problems of high operation power consumption, low recognition speed, large network occupied resource and the like exist in the current deep learning, the requirement of real-time detection cannot be met, and the further development of the traffic sign recognition technology is hindered. When the method is particularly applied to a scheme of traffic sign identification, the method is mainly embodied in that the identification time is too long, real-time processing cannot be realized, or important features are ignored in the feature extraction process, so that the identification result precision is not high, or the neural network design is too complex, so that the calculation amount is large, the occupied resources are more, and the calculation efficiency is lower; the feature extraction process is also accompanied with the problem that the reliability of the recognition result cannot be ensured due to poor image feature expression capability.
Disclosure of Invention
The application provides a SNN traffic sign identification method based on space attention and a related device, which are used for solving the technical problem that the prior art cannot meet the requirements of accuracy and calculation efficiency at the same time, so that the traffic sign identification effect is poor.
In view of this, a first aspect of the present application provides a SNN traffic sign identification method based on spatial attention, including:
inputting a target traffic sign image into a neuron coding layer of a preset SNN network model for coding to obtain a pulse coding pattern, wherein the neuron coding layer comprises soft reset LIF neurons;
extracting an attention feature map in the pulse code map based on a spatial attention network layer in the preset SNN network model;
performing feature analysis processing on the attention feature map according to a feature analysis network layer in the preset SNN network model to obtain a feature map to be identified;
and carrying out mark classification processing by adopting a classification network layer in the preset SNN network model according to the characteristic diagram to be recognized to obtain the target traffic mark category.
Preferably, the target traffic sign image is input into a neuron coding layer of a preset SNN network model for coding to obtain a pulse coding pattern, the neuron coding layer includes soft-reset LIF neurons, and the method further includes:
different traffic sign images are obtained in real time through a preset monitoring device, and preprocessing operation is carried out to obtain a target traffic sign image.
Preferably, the target traffic sign image is input into a neuron coding layer of a preset SNN network model for coding processing to obtain a pulse coding pattern, where the neuron coding layer includes soft-reset LIF neurons, and the method further includes:
carrying out simplified analysis on the initial LIF neurons to obtain soft reset LIF neurons, wherein the soft reset LIF neurons comprise a subtraction threshold reset mode and a preset voltage reset mode;
after a neuron coding layer is built based on the soft reset LIF neurons, an initial SNN network model is built by combining a space attention mechanism;
and performing mark classification training on the initial SNN network model by adopting a preset traffic mark image training set to obtain a preset SNN network model.
Preferably, the performing feature analysis processing on the attention feature map according to a feature analysis network layer in the preset SNN network model to obtain a feature map to be recognized includes:
carrying out feature extraction operation on the attention feature map according to a feature analysis network layer in the preset SNN network model to obtain a feature expression map;
and performing dimensionality reduction processing on the feature expression graph through a pooling layer in the feature analysis network layer to obtain a feature graph to be identified.
The second aspect of the present application provides an SNN traffic sign recognition apparatus based on spatial attention, including:
the image characteristic coding unit is used for inputting the target traffic sign image into a neuron coding layer of a preset SNN network model for coding to obtain a pulse coding pattern, and the neuron coding layer comprises soft reset LIF neurons;
a first feature extraction unit, configured to extract an attention feature map in the pulse code map based on a spatial attention network layer in the preset SNN network model;
the second feature extraction unit is used for performing feature analysis processing on the attention feature map according to a feature analysis network layer in the preset SNN network model to obtain a feature map to be identified;
and the traffic sign identification unit is used for carrying out sign classification processing by adopting a classification network layer in the preset SNN network model according to the characteristic diagram to be identified to obtain a target traffic sign category.
Preferably, the method further comprises the following steps:
and the image acquisition unit is used for acquiring different traffic sign images in real time through a preset monitoring device and carrying out preprocessing operation to obtain a target traffic sign image.
Preferably, the method further comprises the following steps:
the neuron optimization unit is used for carrying out simplified analysis on the initial LIF neurons to obtain soft reset LIF neurons, and the soft reset LIF neurons comprise a subtraction threshold reset mode and a preset voltage reset mode;
the network structure building unit is used for building a neuron coding layer based on the soft reset LIF neurons and then building an initial SNN network model by combining a space attention mechanism;
and the network model training unit is used for performing mark classification training on the initial SNN network model by adopting a preset traffic mark image training set to obtain a preset SNN network model.
Preferably, the second feature extraction unit is specifically configured to:
performing feature extraction operation on the attention feature map according to a feature analysis network layer in the preset SNN network model to obtain a feature expression map;
and performing dimensionality reduction processing on the feature expression graph through a pooling layer in the feature analysis network layer to obtain a feature graph to be identified.
A third aspect of the present application provides a spatial attention based SNN traffic sign recognition apparatus, the apparatus comprising a processor and a memory;
the memory is used for storing program codes and transmitting the program codes to the processor;
the processor is configured to execute the spatial attention based SNN traffic sign recognition method according to the first aspect according to instructions in the program code.
A fourth aspect of the present application provides a computer-readable storage medium for storing a program code for executing the method for SNN traffic sign recognition based on spatial attention according to the first aspect.
According to the technical scheme, the embodiment of the application has the following advantages:
the application provides a SNN traffic sign identification method based on space attention and a related device, and the method comprises the following steps: inputting the target traffic sign image into a neuron coding layer of a preset SNN network model for coding to obtain a pulse coding pattern, wherein the neuron coding layer comprises soft reset LIF neurons; extracting an attention feature map in a pulse code map based on a spatial attention network layer in a preset SNN network model; performing feature analysis processing on the attention feature map according to a feature analysis network layer in a preset SNN network model to obtain a feature map to be recognized; and carrying out mark classification processing by adopting a classification network layer in a preset SNN network model according to the characteristic graph to be recognized to obtain the target traffic mark category.
According to the SNN traffic sign recognition method and the related device based on the space attention, the SNN network model is used for carrying out feature analysis and classification processing on the target traffic sign image, in the process, the neuron coding layer and the feature extraction network layer are used for carrying out improvement and adjustment, the image analysis capability of the model is improved according to the soft reset LIF neuron and the attention mechanism, and the accuracy of a classification result is ensured; and the soft reset LIF neurons in the model can reduce the calculation complexity and improve the calculation efficiency. Therefore, the method and the device can solve the technical problem that the prior art cannot meet the requirements of accuracy and calculation efficiency at the same time and the traffic sign recognition effect is poor.
Drawings
Fig. 1 is a schematic flowchart of an SNN traffic sign recognition method and a related apparatus method based on spatial attention according to an embodiment of the present disclosure;
fig. 2 is a schematic structural diagram of an SNN traffic sign recognition method and a related apparatus based on spatial attention according to an embodiment of the present disclosure;
fig. 3 is a schematic network structure diagram of a predetermined SNN network model according to an embodiment of the present disclosure;
fig. 4 is a schematic diagram of coding of a neuron coding layer constructed based on a soft-reset LIF neuron according to an embodiment of the present application;
fig. 5 is a schematic diagram of a hardware module of an SNN traffic sign recognition system provided in an application example of the present application;
fig. 6 is a state transition diagram of an FSM of a control module in a hardware module according to an application example of the present application.
Detailed Description
In order to make the technical solutions of the present application better understood, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
For easy understanding, please refer to fig. 1, an embodiment of the SNN traffic sign recognition method based on spatial attention and the related apparatus method provided in the present application includes:
step 101, inputting the target traffic sign image into a neuron coding layer of a preset SNN network model for coding to obtain a pulse coding pattern, wherein the neuron coding layer comprises soft reset LIF neurons.
The target traffic sign image can be an image to be recognized acquired in real time or an image for target research and analysis, and the size of the image input into the preset SNN network model is consistent. The preset SNN network model is obtained by pre-configuration and training and can be directly applied to the actual traffic sign image processing and analyzing scene. Referring to fig. 3, the structure of the preset SNN network model in the embodiment mainly includes a neuron coding layer, a spatial attention network layer (module), two convolution layers, a dimension processing layer, and two full connection layers; wherein, the convolution layer comprises an average pooling layer, and finally, the network outputs a classification result, namely an identification result; the dimension processing layer is arranged between the second convolution layer and the first full-connection layer and is used for converting multidimensional input into one dimension to be used as a transition layer from the convolution layer to the full-connection layer.
It should be noted that the neurons in the neuron coding layer in this embodiment are soft reset LIF neurons, and the LIF neurons mainly include three processes: leakage, integration and emission, the kinetic equation for which is:
wherein, V and V reset Membrane potential and reset voltage, τ, of LIF neurons, respectively m For the membrane potential time constant, X (t) is the input of the LIF neuron at time t, and the leakage of the membrane potential is generally simulated by using an exponential equation, which can be expressed as:
wherein, t 1 、t 2 Respectively, at adjacent time points in front and back, let Δ t = t 2 -t 1 The time difference between the front and rear voltages.
The exponential operation of the neurons is relatively complex, which may cause an increase in operation power consumption, so the present embodiment designs a Simplified LIF neuron, that is, a Simplified Soft-reset LIF neuron (SS-LIF) in the present embodiment, and if the input of the LIF neuron is not attenuated, discretizing and recursive processing are performed on the above equation, an approximate difference equation of the mohr voltage change may be obtained:
V(t 2 )=V(t 1 )/2+X(t)
wherein, tau m Set to 2, the neuron fires when its membrane potential reaches a threshold, which dissipates the charge accumulated by the neuron, so that the membrane potential will decrease momentarily, i.e. reset. In the embodiment, two reset methods are provided, one is SOFT-SET, i.e. a subtraction threshold reset mode, and the other is HARD-SET, i.e. a preset voltage reset mode; the former is the membrane potential minus the threshold voltage after the release pulse, i.e.: v (t) = V (t) -V threshold (ii) a The latter, after the release pulse, the membrane potential is set to the reset voltage, i.e.: v (t) = V reset (ii) a And will V reset Is set to 0. IF the IF neuron is an IF neuron, the membrane potential does not need to be attenuated, but only the input needs to be accumulated.
The essence of the coding process is to convert the input real number form pixel data into discrete pulse values, in the process, the target traffic sign image needs to be expanded according to the time sequence, for each time step, the data after normalization processing is input into a neuron coding layer for coding processing, the membrane potential of the neuron is updated by combining the input data and a state updating equation, and meanwhile, whether the membrane potential of the neuron is larger than a set threshold value is judged, if so, a pulse is issued and reset, otherwise, the current state is kept to know the next time step, and the specific coding process can refer to fig. 4.
Further, step 101, before, further includes:
and acquiring different traffic sign images in real time through a preset monitoring device, and performing preprocessing operation to obtain a target traffic sign image. The traffic sign image may be subjected to some image processing, so as to improve the quality of the image, or unify the image size, and the specific selected preprocessing means may be set according to the actual situation without limitation.
Further, step 101, before, further includes:
carrying out simplified analysis on the initial LIF neurons to obtain soft reset LIF neurons, wherein the soft reset LIF neurons comprise a subtraction threshold resetting mode and a preset voltage resetting mode;
after a neuron coding layer is built based on the soft reset LIF neurons, an initial SNN network model is built by combining a space attention mechanism;
and performing mark classification training on the initial SNN network model by adopting a preset traffic mark image training set to obtain a preset SNN network model.
It should be noted that the LIF neuron in this embodiment is a simplified neuron, which can reduce the complexity of the model and improve the computation efficiency; and the soft reset LIF neuron is configured with two different reset modes, so that the method is more accurate and reliable.
And 102, extracting an attention feature map in the pulse code map based on a spatial attention network layer in a preset SNN network model. An attention feature map can be extracted based on a spatial attention mechanism in a preset SNN network model; the mechanism can enable the model to pay more attention to the main features of the image, thereby neglecting the secondary features and improving the accuracy of the model in identifying the traffic signs.
And 103, performing feature analysis processing on the attention feature map according to a feature analysis network layer in a preset SNN network model to obtain a feature map to be recognized.
Further, step 103 includes:
carrying out feature extraction operation on the attention feature map according to a feature analysis network layer in a preset SNN network model to obtain a feature expression map;
and performing dimensionality reduction processing on the feature expression graph through a pooling layer in the feature analysis network layer to obtain a feature graph to be identified.
The alternation of the convolutional layer and the pooling layer in fig. 3 is the feature analysis network layer referred to in this embodiment, and feature analysis and dimension reduction processing can be performed on the feature map, and the processed feature map can be input to the final classification network layer for classification, so as to obtain a classification result.
And step 104, carrying out mark classification processing by adopting a classification network layer in a preset SNN network model according to the characteristic diagram to be recognized to obtain the target traffic mark category.
For the convenience of understanding, the application provides an SNN traffic sign recognition system based on space attention, fig. 5 is a system hardware structure diagram, and fig. 6 is an FSM state transition diagram of a control module. The system mainly comprises an optimization input coding module, a space attention module, a pulse sequencing module, a data storage module, a control module, a pulse cache module, a pulse counting module, a PE array module and a full-connection computing module.
Referring to fig. 6, the control module is mainly divided into five parts, namely, membrane potential leakage, data preparation, convolution calculation, full-connection calculation and data storage, and the control module is implemented by using a state machine. If the system uses LIF neurons to calculate, the calculation flow is (1) (2) (3) (4) (5); IF the system uses IF neurons to perform the calculation, the leakage of membrane potential is not needed, i.e. (1) the calculation flow is (2), (3), (4) and (5). In the control core, there is a network mapping memory, which stores the network layer number, the input channel number of each layer image, the size of convolution kernel, the size of output image and the stride. Different network topologies may be implemented by configuring the network mapping memory.
And the pulse ordering module is used for coding the input pixel data by using pulses and caching the input pixel data into the pulse buffering module, and is also used for completing data prefetching of convolution (ConV) and Pooling (Pooling) calculation, namely, the pulse data and the weight data are prefetched for data rearrangement, and the data type is converted into a form suitable for calculation by using a pulse array.
And the full-connection calculation module realizes full-connection calculation in the SCNN network and adopts event driving to realize that calculation is started when the neurons send pulses. The implementation process comprises the steps of firstly, sequentially obtaining pulse data from a pulse sequencing module, and buffering an address corresponding to a pulse into an FIFO (first in first out) of a full-connection module when the high-level pulse exists. And sequentially traversing the source neuron addresses stored in the FIFO, acquiring the weight relative to the target neuron according to the source address, updating the states of 64 neurons once by adopting a one-to-many and parallel computing mode, and repeating the computation until the state of the target neuron is updated, thereby realizing the full-connection computation.
And a PE array module, wherein the convolution calculation is realized by a 2D systolic array consisting of 16 multiplied by 16 PEs. Each PE contains a neuron and can implement the computation of convolution and pooling. When convolution calculation is carried out, the pulse data stream after mapping and rearrangement is input into the pulse array from each row, corresponding weight data stream is input into each column, and calculation is carried out in a mode of fixing the middle membrane potential. For the multi-input channel image, firstly, the calculation of one input channel for all output channels is completed, the potential value of the intermediate membrane is saved, and then the calculation of the rest input channels is sequentially completed. When the pooling calculation is carried out, the input pulses are accumulated, and after the accumulation is finished, the average pooling calculation of 2 multiplied by 2 can be finished by adopting right shift of 2 bits.
And the data storage module is responsible for storing the weight of the network and the membrane potential of the neuron, acquiring the weight of a corresponding address when the system updates the membrane potential, mainly the weight of convolution operation, and completing reading and writing of the membrane potential of the neuron, wherein the weight in the embodiment is represented by a 16-bit fixed point decimal.
And the pulse buffer module is responsible for storing internal pulse data issued by the neurons in the network computing process and sending the data to the pulse sequencing module for data sequencing when computing is started.
And the pulse counting module is responsible for recording the number of pulses sent by the pulse neurons when the calculation is finished and providing a basis for a final result.
And the optimized input coding module is responsible for coding the input pixel data and converting the pixel value of the real number into a discrete pulse value.
And the space attention module is responsible for extracting attention of the feature map, so that the system pays more attention to the main features of the image, the secondary features are ignored, and the accuracy of the system in identifying the traffic sign is improved.
According to the SNN traffic sign recognition method and the related device based on the space attention, the SNN network model is used for carrying out feature analysis and classification processing on the target traffic sign image, in the process, the neuron coding layer and the feature extraction network layer are used for carrying out improvement and adjustment, the image analysis capability of the model is improved according to the soft reset LIF neuron and the attention mechanism, and the accuracy of the classification result is ensured; and the soft reset LIF neurons in the model can reduce the calculation complexity and improve the calculation efficiency. Therefore, the embodiment of the application can solve the technical problem that the prior art cannot meet the requirements of accuracy and calculation efficiency at the same time, so that the traffic sign identification effect is poor.
For easy understanding, please refer to fig. 2, the present application provides an embodiment of an SNN traffic sign recognition method based on spatial attention and a related device, including:
the image feature coding unit 201 is configured to input the target traffic sign image into a neuron coding layer of a preset SNN network model for coding, so as to obtain a pulse coding pattern, where the neuron coding layer includes soft-reset LIF neurons;
a first feature extraction unit 202, configured to extract an attention feature map in a pulse code map based on a spatial attention network layer in a preset SNN network model;
the second feature extraction unit 203 is configured to perform feature analysis processing on the attention feature map according to a feature analysis network layer in a preset SNN network model to obtain a feature map to be recognized;
and the traffic sign identification unit 204 is configured to perform sign classification processing according to the feature map to be identified by using a classification network layer in the preset SNN network model, so as to obtain a target traffic sign category.
Further, still include:
the image obtaining unit 205 is configured to obtain different traffic sign images in real time through a preset monitoring device, and perform a preprocessing operation to obtain a target traffic sign image.
Further, still include:
the neuron optimization unit 206 is configured to perform simplified analysis on the initial LIF neurons to obtain soft-reset LIF neurons, where the soft-reset LIF neurons include a subtraction threshold resetting mode and a preset voltage resetting mode;
the network structure building unit 207 is used for building a neuron coding layer based on the soft reset LIF neurons, and then building an initial SNN network model by combining a space attention mechanism;
and the network model training unit 208 is configured to perform label classification training on the initial SNN network model by using a preset traffic label image training set to obtain a preset SNN network model.
Further, the second feature extraction unit 203 is specifically configured to:
carrying out feature extraction operation on the attention feature map according to a feature analysis network layer in a preset SNN network model to obtain a feature expression map;
and performing dimensionality reduction processing on the feature expression graph through a pooling layer in the feature analysis network layer to obtain a feature graph to be identified.
The application also provides SNN traffic sign recognition equipment based on space attention, and the equipment comprises a processor and a memory;
the memory is used for storing the program codes and transmitting the program codes to the processor;
the processor is used for executing the SNN traffic sign identification method based on the space attention in the method embodiment according to the instructions in the program code.
The present application also provides a computer-readable storage medium for storing program code for executing the SNN traffic sign recognition method based on spatial attention in the above method embodiments.
In the several embodiments provided in the present application, it should be understood that the disclosed apparatus and method may be implemented in other ways. For example, the above-described apparatus embodiments are merely illustrative, and for example, the division of the units is only one logical division, and other divisions may be realized in practice, for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit may be implemented in the form of hardware, or may also be implemented in the form of a software functional unit.
The integrated unit, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application, or portions or all or portions of the technical solutions that contribute to the prior art, may be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for executing all or part of the steps of the methods described in the embodiments of the present application through a computer device (which may be a personal computer, a server, or a network device). And the aforementioned storage medium includes: a usb disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk.
The above embodiments are only used for illustrating the technical solutions of the present application, and not for limiting the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions in the embodiments of the present application.
Claims (10)
1. The SNN traffic sign identification method based on space attention is characterized by comprising the following steps:
inputting a target traffic sign image into a neuron coding layer of a preset SNN network model for coding to obtain a pulse coding pattern, wherein the neuron coding layer comprises soft reset LIF neurons;
extracting an attention feature map in the pulse code map based on a spatial attention network layer in the preset SNN network model;
performing feature analysis processing on the attention feature map according to a feature analysis network layer in the preset SNN network model to obtain a feature map to be identified;
and carrying out mark classification processing by adopting a classification network layer in the preset SNN network model according to the characteristic diagram to be recognized to obtain the target traffic mark category.
2. The SNN traffic sign recognition method based on spatial attention according to claim 1, wherein the target traffic sign image is input into a neuron coding layer of a preset SNN network model for coding to obtain a pulse coding map, the neuron coding layer includes soft-reset LIF neurons, and the method further includes:
different traffic sign images are obtained in real time through a preset monitoring device, and preprocessing operation is carried out to obtain a target traffic sign image.
3. The SNN traffic sign recognition method based on spatial attention according to claim 1, wherein the target traffic sign image is input into a neuron coding layer of a preset SNN network model for coding to obtain a pulse coding map, the neuron coding layer includes soft-reset LIF neurons, and the method further includes:
carrying out simplified analysis on the initial LIF neurons to obtain soft reset LIF neurons, wherein the soft reset LIF neurons comprise a subtraction threshold reset mode and a preset voltage reset mode;
after a neuron coding layer is built based on the soft reset LIF neurons, an initial SNN network model is built by combining a space attention mechanism;
and performing mark classification training on the initial SNN network model by adopting a preset traffic mark image training set to obtain a preset SNN network model.
4. The SNN traffic sign recognition method based on spatial attention according to claim 1, wherein the performing feature analysis processing on the attention feature map according to a feature analysis network layer in the preset SNN network model to obtain a feature map to be recognized includes:
carrying out feature extraction operation on the attention feature map according to a feature analysis network layer in the preset SNN network model to obtain a feature expression map;
and performing dimensionality reduction processing on the feature expression graph through a pooling layer in the feature analysis network layer to obtain a feature graph to be identified.
5. SNN traffic sign recognition device based on space attention, characterized by including:
the image characteristic coding unit is used for inputting the target traffic sign image into a neuron coding layer of a preset SNN network model for coding to obtain a pulse coding pattern, and the neuron coding layer comprises soft reset LIF neurons;
a first feature extraction unit, configured to extract an attention feature map in the pulse code map based on a spatial attention network layer in the preset SNN network model;
the second feature extraction unit is used for performing feature analysis processing on the attention feature map according to a feature analysis network layer in the preset SNN network model to obtain a feature map to be identified;
and the traffic sign identification unit is used for carrying out sign classification processing by adopting a classification network layer in the preset SNN network model according to the characteristic diagram to be identified to obtain a target traffic sign category.
6. The SNN traffic sign recognition device based on spatial attention of claim 5, further comprising:
the image acquisition unit is used for acquiring different traffic sign images in real time through a preset monitoring device, and preprocessing operation is carried out to obtain the target traffic sign image.
7. The SNN traffic sign recognition device based on spatial attention of claim 5, further comprising:
the neuron optimization unit is used for carrying out simplified analysis on the initial LIF neurons to obtain soft reset LIF neurons, and the soft reset LIF neurons comprise a subtraction threshold reset mode and a preset voltage reset mode;
the network structure building unit is used for building a neuron coding layer based on the soft reset LIF neurons and then building an initial SNN network model by combining a space attention mechanism;
and the network model training unit is used for performing mark classification training on the initial SNN network model by adopting a preset traffic mark image training set to obtain a preset SNN network model.
8. The SNN traffic sign recognition device according to claim 5, wherein the second feature extraction unit is specifically configured to:
performing feature extraction operation on the attention feature map according to a feature analysis network layer in the preset SNN network model to obtain a feature expression map;
and performing dimensionality reduction processing on the feature expression graph through a pooling layer in the feature analysis network layer to obtain a feature graph to be identified.
9. An SNN traffic sign recognition device based on spatial attention, the device comprising a processor and a memory;
the memory is used for storing program codes and transmitting the program codes to the processor;
the processor is configured to execute the SNN traffic sign recognition method based on spatial attention according to any one of claims 1-4 according to instructions in the program code.
10. A computer-readable storage medium for storing program code for performing the method for SNN traffic sign recognition based on spatial attention according to any one of claims 1 to 4.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310156238.2A CN115984822A (en) | 2023-02-22 | 2023-02-22 | SNN traffic sign identification method based on space attention and related device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310156238.2A CN115984822A (en) | 2023-02-22 | 2023-02-22 | SNN traffic sign identification method based on space attention and related device |
Publications (1)
Publication Number | Publication Date |
---|---|
CN115984822A true CN115984822A (en) | 2023-04-18 |
Family
ID=85968281
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310156238.2A Pending CN115984822A (en) | 2023-02-22 | 2023-02-22 | SNN traffic sign identification method based on space attention and related device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115984822A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN118038292A (en) * | 2024-04-11 | 2024-05-14 | 哈尔滨工业大学(深圳)(哈尔滨工业大学深圳科技创新研究院) | Object recognition device and method for satellite on-orbit calculation |
-
2023
- 2023-02-22 CN CN202310156238.2A patent/CN115984822A/en active Pending
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN118038292A (en) * | 2024-04-11 | 2024-05-14 | 哈尔滨工业大学(深圳)(哈尔滨工业大学深圳科技创新研究院) | Object recognition device and method for satellite on-orbit calculation |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110472531B (en) | Video processing method, device, electronic equipment and storage medium | |
CN109388807B (en) | Method, device and storage medium for identifying named entities of electronic medical records | |
WO2021022521A1 (en) | Method for processing data, and method and device for training neural network model | |
CN106951484B (en) | Picture retrieval method and device, computer equipment and computer readable medium | |
KR20210073569A (en) | Method, apparatus, device and storage medium for training image semantic segmentation network | |
CN113780149B (en) | Remote sensing image building target efficient extraction method based on attention mechanism | |
CN111860398B (en) | Remote sensing image target detection method and system and terminal equipment | |
CN110851641B (en) | Cross-modal retrieval method and device and readable storage medium | |
CN110929080B (en) | Optical remote sensing image retrieval method based on attention and generation countermeasure network | |
CN112529146B (en) | Neural network model training method and device | |
CN112200041B (en) | Video motion recognition method and device, storage medium and electronic equipment | |
CN111666919A (en) | Object identification method and device, computer equipment and storage medium | |
CN111144483A (en) | Image feature point filtering method and terminal | |
CN115984822A (en) | SNN traffic sign identification method based on space attention and related device | |
CN111368850A (en) | Image feature extraction method, image target detection method, image feature extraction device, image target detection device, convolution device, CNN network device and terminal | |
CN109886317B (en) | General image aesthetic evaluation method, system and equipment based on attention mechanism | |
CN110659641B (en) | Text recognition method and device and electronic equipment | |
CN111814653B (en) | Method, device, equipment and storage medium for detecting abnormal behavior in video | |
CN112330697A (en) | Image segmentation method and device, electronic equipment and readable storage medium | |
CN111931841A (en) | Deep learning-based tree processing method, terminal, chip and storage medium | |
CN115906861B (en) | Sentence emotion analysis method and device based on interaction aspect information fusion | |
CN115439848A (en) | Scene recognition method, device, equipment and storage medium | |
CN110781223A (en) | Data processing method and device, processor, electronic equipment and storage medium | |
CN112348855B (en) | Visual odometer characteristic point extraction method, system, electronic equipment and storage medium | |
CN111985542B (en) | Representative graph structure model, visual understanding model establishing method and application |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |