Traffic sign recognition method and device, electronic equipment and storage medium

Abstract

The invention provides a traffic sign recognition method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining a traffic sign image, carrying out the global pooling operation of the traffic sign image, obtaining the global feature information, inputting the global feature information into a channel attention module, and obtaining a channel attention module; attention image features are obtained by calculating the attention weight of each channel in the global feature information, the size value of a filter in a channel attention module is adaptively adjusted based on the number of channels of the global feature information, and the attention image features are input into a big kernel attention module. The attention image features are deeply extracted through a plurality of convolutional layers and then spliced to obtain large-kernel image features, traffic sign features are obtained based on the large-kernel image features and the attention image features, and target detection is performed on the traffic sign features to obtain a traffic sign recognition result. Through the traffic sign recognition method, the technical problem of low accuracy of long-distance traffic sign recognition is solved.

Description

Technical Field

[0001] The present application relates to the field of image recognition technology, and in particular to a traffic sign recognition method, device, electronic device and storage medium. Background Art

[0002] In recent years, with the increasing busyness of urban traffic and the continuous expansion of road networks, vehicle traffic sign recognition technology has gradually become a key component of intelligent transportation systems. If drivers cannot accurately recognize long-distance traffic signs, such as ignoring stop signs, speed limit signs or turn signs, they may make incorrect driving decisions and increase the risk of traffic accidents. Their wrong behavior may also cause danger to other road users. For example, sudden lane changes or parking in the wrong place may cause accidents to other vehicles or traffic congestion. Therefore, the development of vehicle traffic sign recognition technology is of great significance to the construction of intelligent, safe and efficient transportation systems.

[0003] Although the existing vehicle traffic sign recognition technology has achieved certain results and can perform some close-range traffic sign recognition, at long distances, due to the small size of the target object, it is often mixed with the background and may also be affected by image blur and other issues, further reducing the recognition accuracy. In addition, the traditional convolution operation cannot fully utilize the spatial information in the image, which will lead to inaccurate matching problems. Summary of the invention

[0004] In view of the shortcomings of the related technologies mentioned above, the present application provides a traffic sign recognition method, device, electronic device and storage medium to solve the technical problem of low accuracy in long-distance traffic sign recognition.

[0005] The present application provides a traffic sign recognition method, which includes: acquiring a traffic sign image, and performing a global pooling operation on the traffic sign image to obtain global feature information, wherein the global feature information includes image features of multiple channels; inputting the global feature information into a channel attention module, and obtaining an attention image feature by calculating the attention weights of each channel in the global feature information, wherein the filter size value in the channel attention module is adaptively adjusted based on the number of channels of the global feature information; inputting the attention image feature into a large core attention module, performing deep feature extraction on the attention image feature through multiple convolutional layers and then splicing it to obtain a large core image feature, and obtaining a traffic sign feature based on the large core image feature and the attention image feature; and performing target detection on the traffic sign feature to obtain a traffic sign recognition result.

[0006] In one embodiment of the present application, obtaining traffic sign features based on the large core image features and the attention image features includes: performing maximum pooling operations and average pooling operations on each channel in the large core image features to obtain maximum pooling features and average pooling features of the large core image features; obtaining feature weights based on the maximum pooling features and average pooling features of the large core image features, and fusing the attention image features and the feature weights with the output of each convolutional layer one by one to obtain traffic sign features.

[0007] In one embodiment of the present application, the traffic sign recognition method also includes: taking the traffic sign feature as a first scale feature, inputting the first scale feature into the channel attention module, and inputting the output of the channel attention module corresponding to the first scale feature into the large core attention module to obtain a second scale feature; inputting the second scale feature into the channel attention module, and inputting the output of the channel attention module corresponding to the second scale feature into the large core attention module to obtain a third scale feature; performing feature fusion on the first scale feature, the second scale feature and the third scale feature to obtain a scale fusion feature; performing target detection on the scale fusion feature to obtain a traffic sign recognition result.

[0008] In one embodiment of the present application, performing feature fusion on the first scale feature, the second scale feature, and the third scale feature to obtain a scale fusion feature includes: performing an upsampling operation on the third scale feature to obtain a third scale upsampled feature; performing feature fusion on the third scale upsampled feature and the second scale feature to obtain a fused second scale feature; performing an upsampling operation on the fused second scale feature to obtain a second scale upsampled feature; and performing feature fusion on the second scale upsampled feature and the first scale feature to obtain a scale fusion feature.

[0009] In one embodiment of the present application, the filter size value in the channel attention module is adaptively adjusted based on the number of channels of the global feature information, including:

[0010]

[0011] Wherein, k is the filter size value, C is the number of channels, and η, b are preset linear mapping parameters.

[0012] In one embodiment of the present application, deep feature extraction is performed on the attention image features through multiple convolution layers and then spliced ​​to obtain large core image features, including: decomposing a large core convolution into a first small core convolution and a second small core convolution, and performing deep feature extraction on the attention image features based on the first small core convolution and the second small core convolution respectively; the features extracted by the first small core convolution and the second small core convolution are spliced ​​to obtain the large core image features.

[0013] In one embodiment of the present application, obtaining traffic sign features based on the large core image features and the attention image features includes:

[0014]

[0015] in, represents the traffic sign feature, X represents the attention image feature, F(·) represents the fusion convolution layer, represents the feature weight, which is obtained after activating the maximum pooling feature and the average pooling feature, Y i is the image feature output after the i-th convolutional layer.

[0016] An embodiment of the present application provides a traffic sign recognition device, which includes: an image input module, used to acquire a traffic sign image, and perform a global pooling operation on the traffic sign image to obtain global feature information, wherein the global feature information includes image features of multiple channels; a channel attention module, used to input the global feature information into the channel attention module, and obtain attention image features by calculating the attention weights of each channel in the global feature information, and the filter size value in the channel attention module is adaptively adjusted based on the number of channels of the global feature information; a large core attention module, used to input the attention image features into the large core attention module, perform deep feature extraction on the attention image features through multiple convolutional layers and then splice them to obtain large core image features, and obtain traffic sign features based on the large core image features and the attention image features; a sign recognition module, used to perform target detection on the traffic sign features to obtain traffic sign recognition results.

[0017] An embodiment of the present application also provides an electronic device, comprising: one or more processors; a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the electronic device implements the traffic sign recognition method as described in any of the above embodiments.

[0018] An embodiment of the present application further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor of a computer, the computer is enabled to execute a traffic sign recognition method as described in any one of the above embodiments.

[0019] Beneficial effects of the present application: The embodiments of the present application provide a traffic sign recognition method, device, electronic device and storage medium. The method obtains a traffic sign image and performs a global pooling operation on the traffic sign image to obtain global feature information, inputs the global feature information into a channel attention module, obtains attention image features by calculating the attention weights of each channel in the global feature information, adaptively adjusts the filter size value in the channel attention module based on the number of channels of the global feature information, inputs the attention image features into a large core attention module, deeply extracts the attention image features through multiple convolutional layers and then splices them to obtain large core image features, obtains traffic sign features based on the large core image features and the attention image features, performs target detection on the traffic sign features to obtain traffic sign recognition results, adaptively adjusts the number of channels to better capture important features in the data, processes the output of the channel attention module based on the large core attention module, retains more detailed feature information while maintaining the receptive field, so as to realize feature extraction of long-distance traffic signs, thereby preventing the problem of unclear recognition due to the long distance of the traffic sign.

[0020] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 is a schematic diagram of an implementation environment of a traffic sign recognition method shown in an exemplary embodiment of the present application;

[0022] Figure 2 is a flow chart of a traffic sign recognition method shown in an exemplary embodiment of the present application;

[0023] Figure 3 is a schematic diagram of a large core attention module shown in an exemplary embodiment of the present application;

[0024] Figure 4 is a schematic diagram of a channel attention module shown in an exemplary embodiment of the present application;

[0025] Figure 5 is a block diagram of a traffic sign recognition device shown in an exemplary embodiment of the present application;

[0026] Figure 6 It is a structural schematic diagram of an electronic device shown in an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0027] The following describes the embodiments of the present application through specific examples, and those skilled in the art can easily understand other advantages and effects of the present application from the contents disclosed in this specification. The present application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present application. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0028] It should be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present application, and thus the drawings only show components related to the present application rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed at will, and the component layout may also be more complicated.

[0029] It should be noted that in this application, "first", "second", etc. are only used to distinguish similar objects, and are not used to limit the order or precedence of similar objects. The variations of "including", "having", etc. described above indicate that the scope covered by the subject of the word is not exclusive except for the examples shown by the word.

[0030] It is understood that the various numbers, step numbers, and other reference numerals recorded in this application are distinguished for the convenience of description and are not intended to limit the scope of this application. The size of the reference numerals in this application does not mean the order of execution. The execution order of each process should be determined by its function and internal logic.

[0031] In the following description, a large number of details are discussed to provide a more thorough explanation of the embodiments of the present application. However, it is obvious to those skilled in the art that the embodiments of the present application can be implemented without these specific details. In other embodiments, well-known structures and devices are shown in the form of block diagrams rather than in detail to avoid making the embodiments of the present application difficult to understand.

[0032] The embodiments of the present application respectively propose a traffic sign recognition method, a traffic sign recognition device, an electronic device, a computer-readable storage medium, and a computer program product, which will be described in detail below.

[0033] See also Figure 1 , Figure 1 It is a schematic diagram of an implementation environment of a traffic sign recognition method shown in an exemplary embodiment of the present application.

[0034] like Figure 1As shown, the implementation environment may include an image acquisition device 101 and a computer device 102, wherein the image acquisition device 101 may be a camera installed on a road or on a vehicle, and the computer device 102 may be at least one of a microcomputer, an embedded computer, a neural network computer, etc. The trained model parameters are saved in the computer device 102 as a weight file, and then the traffic sign image obtained by the image acquisition device 101 is inferred to obtain a recognition result, and the recognition result is superimposed on the original image to generate a picture with an identification label, so as to accurately identify the traffic sign of a small target vehicle at a long distance.

[0035] See also Figure 2 , Figure 2 is a flow chart of a traffic sign recognition method shown in an exemplary embodiment of the present application. The method can be applied to Figure 1 The implementation environment shown is as follows. The method may also be applicable to other exemplary implementation environments and be specifically executed by devices in other implementation environments. This embodiment does not limit the implementation environment to which the method is applicable.

[0036] like Figure 2 As shown, in an exemplary embodiment, the traffic sign recognition method includes at least steps S210 to S240, which are described in detail as follows:

[0037] Step S210 , acquiring a traffic sign image, and performing a global pooling operation on the traffic sign image to obtain global feature information, where the global feature information includes image features of multiple channels.

[0038] In one embodiment of the present application, the original sign image is scaled to a size of 1024×1024 and randomly horizontally flipped to obtain a traffic sign image, and a global pooling operation is performed on the traffic sign image including global average pooling and global maximum pooling to obtain global feature information containing multiple channels.

[0039] In step S220, the global feature information is input into the channel attention module, and the attention image feature is obtained by calculating the attention weight of each channel in the global feature information. The filter size value in the channel attention module is adaptively adjusted based on the number of channels of the global feature information.

[0040] In one embodiment of the present application, in order to enhance the connection between channels, all channels share the same learning parameters, that is:

[0041]

[0042] In formula (1), w is the preset parameter matrix, x i is the pixel point in the image feature, k is the receptive field within the range of k units, Represents the set of k adjacent channels of the feature map in the global feature information

[0043] In one embodiment of the present application, the channel attention module is implemented by a two-dimensional convolution with a convolution kernel of k, and the high-dimensional channel information has a longer-range interaction, while the low-dimensional channel information has a shorter-range interaction by using nonlinear mapping.

[0044] In one embodiment of the present application, the filter size value in the channel attention module is adaptively adjusted based on the number of channels of the global feature information, including:

[0045]

[0046] Formula (2), where k is the filter size value, C is the number of channels, and η,b are the preset linear mapping parameters.

[0047] In step S230, the attention image features are input into the large core attention module, and the attention image features are subjected to deep feature extraction through multiple convolutional layers and then concatenated to obtain large core image features, and traffic sign features are obtained based on the large core image features and the attention image features.

[0048] In one embodiment of the present application, performing deep feature extraction on attention image features through multiple convolutional layers and then splicing them to obtain large core image features includes: decomposing a large core convolution into a first small core convolution and a second small core convolution, and performing deep feature extraction on the attention image features based on the first small core convolution and the second small core convolution respectively; splicing the features extracted by the first small core convolution and the second small core convolution to obtain the large core image features.

[0049] In one embodiment of the present application, determining the number of decomposed kernels is a key to the large-core attention module, and the increase in kernel size and dilation rate ensures that the receptive field expands quickly enough. In the present application, the kernel size k, the dilation rate d, and the receptive field R of the i-th convolution in the large-core attention module are defined as follows:

[0050] k i-1 ≤k i , d1=1,d i-1 <d i ≤R i-1 Formula (3)

[0051] R1=k1,R i =d i (k i -1)+R i-1 Formula (4)

[0052] In equations (3) and (4), k is the kernel size, d is the dilation rate, and R is the receptive field.

[0053] According to the definitions of formula (3) and formula (4), in this embodiment, a convolution with a kernel of 23 is split into a convolution with a kernel of 5 and a convolution with a kernel of 7, so that the extracted features are more accurate. The splitting operation is implemented by convolutions with different receptive fields, which is expressed as follows:

[0054] y0=X,y i+1 =F i (y i ) Formula (5)

[0055] In equation (5), y0 and X represent the original image, and F is the kernel k. i The depth of the convolution, y i Represents the image features after the i-th convolution, y i+1 Represents y i Image features after deep convolution.

[0056] In one embodiment of the present application, obtaining traffic sign features based on large core image features and attention image features includes: performing maximum pooling operations and average pooling operations on each channel in the large core image features to obtain maximum pooling features and average pooling features of the large core image features; obtaining feature weights based on the maximum pooling features and average pooling features of the large core image features, and fusing the attention image features and feature weights with the output of each convolutional layer one by one to obtain traffic sign features.

[0057] In one embodiment of the present application, a maximum pooling operation and an average pooling operation are performed on each channel in the large core image feature, so as to enhance the perception of the feature information of each channel in the model and make the network more flexible in processing the information of each channel:

[0058] S avg =P avg (Y),S max =P max (Y) Formula (6)

[0059] In formula (6), S avg is the average pooling feature, S max is the maximum pooling feature, P avg is the average pooling operation, P max is the maximum pooling operation, and (Y) is the large core image feature.

[0060] In one embodiment of the present application, in order to reflect the information interaction between different spatial descriptions, the two spatial pooling features are connected, and then the sigmoid activation function is applied to obtain a separate spatial selection mask for each decomposition kernel, allowing the model to adaptively select features in different decomposition kernels:

[0061]

[0062] In formula (7), σ(·) represents the sigmoid function, S avg is the average pooling feature, S max is the maximum pooling feature, is the feature weight.

[0063] In one embodiment of the present application, the features decomposed by the small kernel are weighted by the corresponding spatial selection mask and fused by the convolution layer F(·). Finally, the attention weight is introduced to strengthen the attention to the input features. The traffic sign features obtained based on the large kernel image features and the attention image features include:

[0064]

[0065] In formula (8), represents the traffic sign feature, X represents the attention image feature, F(·) represents the fusion convolution layer, Represents the feature weight, which is obtained after activating the maximum pooling feature and the average pooling feature. i is the image feature output after the i-th convolutional layer.

[0066] Step S240: performing target detection on the traffic sign features to obtain a traffic sign recognition result.

[0067] In one embodiment of the present application, the traffic sign recognition method also includes: taking the traffic sign feature as the first scale feature, inputting the first scale feature into the channel attention module, and inputting the output of the channel attention module corresponding to the first scale feature into the large core attention module to obtain the second scale feature; inputting the second scale feature into the channel attention module, and inputting the output of the channel attention module corresponding to the second scale feature into the large core attention module to obtain the third scale feature; performing feature fusion on the first scale feature, the second scale feature and the third scale feature to obtain the scale fusion feature; performing target detection on the scale fusion feature to obtain the traffic sign recognition result.

[0068] In one embodiment of the present application, performing feature fusion on the first scale feature, the second scale feature, and the third scale feature to obtain a scale fusion feature includes: performing an upsampling operation on the third scale feature to obtain a third scale upsampled feature; performing feature fusion on the third scale upsampled feature and the second scale feature to obtain a fused second scale feature; performing an upsampling operation on the fused second scale feature to obtain a second scale upsampled feature; and performing feature fusion on the second scale upsampled feature and the first scale feature to obtain a scale fusion feature.

[0069] In one embodiment of the present application, fusing features of different scales includes: upsampling the feature map of the last layer to obtain a richer feature expression, fusing the convolved feature map with the feature map of the previous layer to obtain a richer feature expression, and repeating the above steps until the highest layer is reached.

[0070] In one embodiment of the present application, target detection is performed on traffic sign features to obtain traffic sign recognition results, and a trained target detection model (SVM or CNN) can be used to classify the traffic recognition features to identify the type of traffic sign features.

[0071] In one embodiment of the present application, the result of target detection is compared with the true label, the loss of bounding box regression is calculated using the complete intersection-over-union loss function and the distribution focus loss function, and the binary cross entropy loss function is used to calculate the loss of the category, and the performance of the model is evaluated by measuring the difference between the model prediction and the actual sample. Then, the parameters of the model are adjusted using the back propagation mechanism so that the gradient information of the loss function can be propagated from the output layer to each layer of the network. The weights and biases of the network are updated according to the gradient information to gradually reduce the loss function.

[0072] In one embodiment of the present application, processing the calculated loss function and outputting the model weight includes: comparing the model's prediction results with the true label, using CIoU and DFL to calculate the loss function of the bounding box regression, using BCE to calculate the loss function of the category, and evaluating the performance of the model by measuring the difference between the model prediction and the actual sample. The following are the formulas of various loss functions:

[0073]

[0074] In equations (9) and (10), CIOU represents the loss function, IoU represents the intersection over union ratio, d is the distance between the center points of the predicted box and the true box, c is the diagonal distance of the minimum circumscribed rectangle, and v is the preset correction factor.

[0075] DFL(S i ,S i+1 )=-((y i+1 -y)log(S i )+(yy i )log(S i+1 )) Formula (11)

[0076] In formula (11), S i Represents a single peak distribution of n+1 units processed by the Softmax layer. i and i+1 Represents two predicted values ​​near the label y.

[0077]

[0078] In formula (12), y is a binary label 0 or 1, p(y) is the probability that the output belongs to the label, and N represents the number of groups of objects predicted by the model.

[0079] See also Figure 3 , Figure 3 is a schematic diagram of a large core attention module shown in an exemplary embodiment of the present application. Figure 3 According to the rules set by formula (3) and formula (4), a large kernel convolution is split into multiple small convolution kernels with different sizes of receptive fields in the LBN module. The size, hole rate and receptive field of each split convolution are determined to ensure the rapid expansion of the receptive field. Then, the 1×1 convolution layer conv2D is used to further process the output of LBN. After activation by GeLU, the original large kernel convolution is decomposed using the matrix multiplication strategy in KABlock to increase the receptive field. Assuming there are N decomposed convolution kernels, after the depth convolution, a 1×1 convolution layer is used for further processing. The number of channels of each decomposed convolution kernel becomes the original number. times and concatenates them to enhance the model's ability to capture features of receptive fields of different sizes, enabling the model to consider information at multiple scales simultaneously. DWConv is a lightweight convolution operation that can achieve the same effect as traditional convolution with fewer parameters and computation. Compared with traditional convolution operations, the main difference between DWConv and conv2D is that its convolution kernel is separated, that is, it is divided into two parts for convolution, including depth convolution and point-by-point convolution.

[0080] See also Figure 4 , Figure 4 : is a schematic diagram of a channel attention module shown in an exemplary embodiment of the present application. The original traffic sign image X is subjected to a GAP operation to obtain global feature information containing multiple channels, where the number of channels is C. An adaptive parameter k is used to perform adaptive convolution on features of different dimensions, where k = τ(C). The feature maps of multiple channels in the global feature information are activated after convolution. Traffic sign images are used to obtain attention image features.

[0081] See also Figure 5 , Figure 5 FIG. 1 is a block diagram of a traffic sign recognition device shown in an exemplary embodiment of the present application. Figure 5 As shown, the exemplary traffic sign recognition device includes:

[0082] An image input module 501 is used to obtain a traffic sign image and perform a global pooling operation on the traffic sign image to obtain global feature information, where the global feature information includes image features of multiple channels;

[0083] A channel attention module 502 is used to input the global feature information into the channel attention module, and obtain the attention image feature by calculating the attention weight of each channel in the global feature information. The filter size value in the channel attention module is adaptively adjusted based on the number of channels of the global feature information;

[0084] The large core attention module 503 is used to input the attention image features into the large core attention module, perform deep feature extraction on the attention image features through multiple convolutional layers and then concatenate them to obtain large core image features, and obtain traffic sign features based on the large core image features and the attention image features;

[0085] The sign recognition module 504 is used to perform target detection on traffic sign features to obtain traffic sign recognition results.

[0086] Figure 6 The structure diagram of the computer system suitable for implementing the electronic device of the embodiment of the present application is shown. It should be noted that: Figure 6 The computer system 600 of the electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0087] like Figure 6 As shown, the computer system 600 includes a central processing unit (CPU) 601, which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 602 or the program loaded from the storage part 608 to the random access memory (RAM) 603, such as executing the method described in the above embodiment. In the RAM 603, various programs and data required for system operation are also stored. The CPU 601, ROM 602 and RAM 603 are connected to each other through a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0088] The following components are connected to the I / O interface 605: an input section 606 including a keyboard, a mouse, etc.; an output section 607 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN (Local Area Network) card, a modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the I / O interface 605 as needed. A removable medium 611, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 610 as needed so that a computer program read therefrom is installed into the storage section 608 as needed.

[0089] In particular, according to an embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a computer program for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication section 609, and / or installed from a removable medium 611. When the computer program is executed by a central processing unit (CPU) 601, various functions defined in the system of the present application are executed.

[0090] It should be noted that the computer-readable medium shown in the embodiment of the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium can be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media can include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, wherein a computer-readable computer program is carried. This propagated data signal can take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. A computer program contained on a computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0091] The flowchart and block diagram in the accompanying drawings illustrate the possible architecture, functions and operations of the system, method and computer program product according to various embodiments of the present application. Wherein, each box in the flowchart or block diagram can represent a module, a program segment, or a part of the code, and the above-mentioned module, program segment, or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0092] The units involved in the embodiments described in this application may be implemented by software or hardware, and the units described may also be set in a processor. The names of these units do not, in some cases, constitute limitations on the units themselves.

[0093] Another aspect of the present application further provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor of a computer, the computer executes the traffic sign recognition method as described above. The computer-readable storage medium may be included in the electronic device described in the above embodiment, or may exist independently without being assembled into the electronic device.

[0094] Another aspect of the present application also provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the traffic sign recognition method provided in each of the above embodiments.

[0095] The above embodiments are merely illustrative of the principles and effects of the present application, and are not intended to limit the present application. Anyone familiar with the technology may modify or change the above embodiments without violating the spirit and scope of the present application. Therefore, all equivalent modifications or changes made by a person of ordinary skill in the art without departing from the spirit and technical ideas disclosed in the present application shall still be covered by the claims of the present application.

Claims

1. A method for identifying traffic signs, characterized in that: The traffic sign recognition method comprises: Acquire a traffic sign image, and perform a global pooling operation on the traffic sign image to obtain global feature information, wherein the global feature information includes image features of multiple channels; The global feature information is input into a channel attention module, and an attention image feature is obtained by calculating the attention weight of each channel in the global feature information, and the filter size value in the channel attention module is adaptively adjusted based on the number of channels of the global feature information; Input the attention image features into the large core attention module, perform deep feature extraction on the attention image features through multiple convolutional layers and then concatenate them to obtain large core image features, and obtain traffic sign features based on the large core image features and the attention image features; Target detection is performed on the traffic sign features to obtain a traffic sign recognition result.

2. The traffic sign recognition method according to claim 1, characterized in that: Obtaining traffic sign features based on the large core image features and the attention image features includes: Performing a maximum pooling operation and an average pooling operation on each channel of the large core image feature to obtain a maximum pooling feature and an average pooling feature of the large core image feature; A feature weight is obtained according to the maximum pooling feature and the average pooling feature of the large core image feature, and a traffic sign feature is obtained by fusing the attention image feature and the feature weight with the output of each convolutional layer one by one.

3. The traffic sign recognition method according to claim 1, characterized in that: The traffic sign recognition method further comprises: The traffic sign feature is used as a first scale feature, the first scale feature is input into the channel attention module, and the output of the channel attention module corresponding to the first scale feature is input into the large core attention module to obtain a second scale feature; Inputting the second scale feature into the channel attention module, and inputting the output of the channel attention module corresponding to the second scale feature into the large core attention module to obtain a third scale feature; Performing feature fusion on the first scale feature, the second scale feature, and the third scale feature to obtain a scale fusion feature; The scale fusion features are used to perform target detection to obtain a traffic sign recognition result.

4. The traffic sign recognition method according to claim 3, characterized in that: Performing feature fusion on the first scale feature, the second scale feature, and the third scale feature to obtain a scale fusion feature includes: Performing an upsampling operation on the third-scale feature to obtain a third-scale upsampled feature; Performing feature fusion on the third scale sampled features and the second scale features to obtain fused second scale features; Performing an upsampling operation on the fused second-scale features to obtain second-scale upsampled features; The second-scale upsampled features and the first-scale features are fused to obtain scale fusion features.

5. The traffic sign recognition method according to any one of claims 1 to 3, characterized in that: The filter size value in the channel attention module is adaptively adjusted based on the number of channels of the global feature information, including: Wherein, k is the filter size value, C is the number of channels, and η, b are preset linear mapping parameters.

6. The traffic sign recognition method according to any one of claims 1 to 3, characterized in that: The attention image features are extracted through multiple convolutional layers and then concatenated to obtain large core image features, including: Decomposing a large kernel convolution into a first small kernel convolution and a second small kernel convolution, and performing deep feature extraction on the attention image features based on the first small kernel convolution and the second small kernel convolution respectively; The features extracted by the first small kernel convolution and the second small kernel convolution are concatenated to obtain the large kernel image features.

7. The traffic sign recognition method according to any one of claim 2, characterized in that: Obtaining traffic sign features based on the large core image features and the attention image features includes: in, represents the traffic sign feature, X represents the attention image feature, F(·) represents the fusion convolution layer, represents the feature weight, Y i is the image feature output after the i-th convolutional layer.

8. A traffic sign recognition device, characterized in that: The traffic sign recognition device comprises: An image input module is used to obtain a traffic sign image and perform a global pooling operation on the traffic sign image to obtain global feature information, wherein the global feature information includes image features of multiple channels; A channel attention module, used for inputting the global feature information into the channel attention module, obtaining the attention image feature by calculating the attention weight of each channel in the global feature information, and the filter size value in the channel attention module is adaptively adjusted based on the number of channels of the global feature information; A large core attention module, used for inputting the attention image features into the large core attention module, performing deep feature extraction on the attention image features through multiple convolutional layers and then concatenating them to obtain large core image features, and obtaining traffic sign features based on the large core image features and the attention image features; The sign recognition module is used to perform target detection on the traffic sign features to obtain a traffic sign recognition result.

9. An electronic device, characterized in that: The electronic device comprises: one or more processors; A storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, enables the electronic device to implement the traffic sign recognition method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor of a computer, the computer is caused to execute the traffic sign recognition method as described in any one of claims 1 to 7.

Images