Large-scale small target detection method and device in traffic scene, equipment and medium

Abstract

The embodiment of the invention provides a large-scale small target detection method and device in a traffic scene, equipment and a medium, and the method comprises the steps: obtaining image information in the traffic scene, and extracting image features; processing the image features by using strip pooling based on a longitudinal and transverse attention network to generate a pair of direction perception feature maps; respectively processing the pooled feature maps, and outputting a first convolution feature function; splitting an output result of the first convolution characteristic function into two independent tensors according to the spatial dimension, and respectively carrying out convolution on the two tensors; processing the two convoluted tensors based on input features to generate a final convolution feature function so as to extract target key features; the correlation between the features is enhanced, so that the features of randomly distributed small targets with low discrimination are enhanced, the network can capture key features of the targets more accurately, and the problems of small target positioning and missing detection are relieved.

Description

Technical Field

[0001] The present application relates to the field of traffic detection technology, and more specifically, to a large-scale small target detection method, device, equipment and medium in a traffic scene. Background Art

[0002] With the rise of deep convolutional neural networks, general object detection has made significant progress. These general object detection algorithms can effectively detect and locate objects of various sizes, benefiting many practical application scenarios. However, despite the prosperity of the general object detection field, the field of small object detection has not made the significant progress that people expected. Small object detection (SOD) is a subfield of object detection that focuses on detecting objects with smaller volumes. It has important theoretical and practical significance in various scenarios such as surveillance, drone scene analysis, pedestrian detection, and traffic sign detection in autonomous driving. At present, small object detection technology has inaccurate position perception, which affects the capture of key features of the target. Summary of the invention

[0003] The purpose of the embodiments of the present application is to provide a large-scale small target detection method, device, equipment and medium in a traffic scenario, so as to solve the problem that the existing small target detection technology has inaccurate position perception, which affects the capture of key features of the target.

[0004] In a first aspect, an embodiment of the present application provides a large-scale small target detection method in a traffic scenario, including:

[0005] Obtain image information in traffic scenes and extract image features;

[0006] Based on the vertical and horizontal attention network, the image features are processed using strip pooling to generate a pair of direction-aware feature maps;

[0007] Process the pooled feature maps respectively and output the first convolution feature function;

[0008] Split the output of the first convolution feature function into two independent tensors according to the spatial dimension, and convolve the two tensors separately;

[0009] The two tensors after convolution are processed based on the input features to generate the final convolution feature function to extract the target key features.

[0010] In the above implementation process, image information in the traffic scene is obtained, and image features are extracted; based on the vertical and horizontal attention networks, the image features are processed using strip pooling to generate a pair of direction-aware feature maps; the pooled feature maps are processed separately to output the first convolution feature function; the output result of the first convolution feature function is split into two independent tensors according to the spatial dimension, and the two tensors are convolved separately; the two tensors after convolution are processed based on the input features to generate the final convolution feature function to extract the key features of the target; the correlation between the features is strengthened, so that the small targets with random distribution and poor discriminability are enhanced in features, so that the network can more accurately capture the key features of the target and alleviate the problems of small target positioning and missed detection.

[0011] Furthermore, the image features are processed by strip pooling based on the vertical and horizontal attention network to generate a pair of direction-aware feature maps, including:

[0012] Receive input features of an image:

[0013] X=R H×W ;

[0014] Among them, H is the row eigenvalue and W is the column eigenvalue;

[0015] Based on the vertical and horizontal attention network, the first pooling kernel is used to process the input features of the image, and each channel is encoded horizontally to obtain the output feature value after horizontal strip pooling:

[0016]

[0017] Use the second pooling kernel to process the input features of the image, encode each channel along the vertical direction, and obtain the output feature value after vertical strip pooling:

[0018]

[0019] Where h is any value of the row eigenvalue, w is any value of the column eigenvalue, 0≤i <h,0≤j<w;

[0020] Features are aggregated horizontally and vertically respectively to generate a pair of direction-aware feature maps after pooling.

[0021] In the above implementation process, the attention module is allowed to capture long-distance dependencies in one spatial direction and retain precise location information in another spatial direction, so that the network can more accurately locate the target of interest, achieve a global receptive field and encode precise location information. In addition, it can effectively capture the relationship between channels.

[0022] Furthermore, the processing of the pooled feature maps respectively and outputting a first convolution feature function includes:

[0023] The perceptual feature maps in the two directions after pooling are processed by the first convolution transformation function and the second convolution transformation function respectively, and the processed features are spliced ​​to obtain spliced ​​features;

[0024] The concatenated features are processed based on the third convolution change function, and the first convolution feature function is output:

[0025]

[0026] in, represents the cascade operation along the spatial dimension, δ is a nonlinear activation function, F1 is the first convolution transformation function, F2 is the second convolution transformation function, F3 is the third convolution transformation function, and Z h is the perceptual feature map in the horizontal direction, Z w It is a perceptual feature map in the vertical direction;

[0027] The intermediate feature maps encoding spatial information in the horizontal and vertical directions are:

[0028] f∈R C / r×(H+W) ;

[0029] Among them, C represents the number of channels, and r represents the coefficient of channel compression.

[0030] In the above implementation process, the feature maps are convolved and then spliced, and then further convolved to output intermediate feature maps that encode spatial information in the horizontal and vertical directions.

[0031] Furthermore, the step of splitting the output result of the first convolution feature function into two independent tensors according to the spatial dimension and convolving the two tensors separately includes:

[0032] The first convolution feature function is split into two independent tensors according to the spatial dimension. The horizontal tensor is:

[0033] f h ∈R C / r×H ;

[0034] The longitudinal tensor is:

[0035] f w ∈R C / r×W ;

[0036] The horizontal tensor is convolved based on the fourth convolution transformation function, and the convolved function is transformed into a tensor with the same number of channels as the input feature:

[0037] g h =σ(Fh (f h ));

[0038] The vertical tensor is convolved based on the fifth convolution transformation function, and the convolved function is transformed into a tensor with the same number of channels as the input feature:

[0039] g w =σ(F w (f w ));

[0040] Among them, σ is the activation function.

[0041] In the above implementation process, the obtained intermediate feature map is split into two independent tensors, and then transformed into tensors with the same number of channels as the input features after convolution. The horizontal and vertical codes are added to the input tensor at the same time. In these two coding maps, each element reflects whether the target of interest exists in the corresponding row and column. Through this encoding process, the module can more accurately locate the exact position of the target of interest, thereby helping the entire model to better identify it.

[0042] Furthermore, the two tensors after convolution are processed based on the input features to generate a final convolution feature function to extract the target key features, including:

[0043] The two tensors after convolution are processed based on the input features, and the number of channels of the input features is reduced by setting the reduction ratio to obtain the final convolution feature function:

[0044]

[0045] In the above implementation process, in order to reduce the complexity of the overhead model, the number of channels of the input features is reduced by setting a reduction ratio.

[0046] Furthermore, the obtaining of image information in a traffic scene and extracting image features include:

[0047] The image information of the traffic scene is obtained through the camera, and the image information is extracted to obtain row features and column features.

[0048] In the above implementation process, row features and column features are obtained to achieve subsequent extraction of target key features.

[0049] In a second aspect, an embodiment of the present application provides a large-scale small target detection device in a traffic scenario, including:

[0050] Image acquisition module, used to acquire image information in traffic scenes and extract image features;

[0051] An image processing module is used to process image features using strip pooling based on the vertical and horizontal attention networks to generate a pair of direction-aware feature maps;

[0052] A first convolution module, used to process the pooled feature maps respectively and output a first convolution feature function;

[0053] A second convolution module is used to split the output result of the first convolution feature function into two independent tensors according to the spatial dimension, and convolve the two tensors separately;

[0054] The third convolution module is used to process the two tensors after convolution based on the input features to generate the final convolution feature function to extract the target key features.

[0055] In a third aspect, an embodiment of the present application provides an electronic device, including:

[0056] A processor, a memory and a bus, wherein the processor is connected to the memory via the bus, and the memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, they are used to implement the large-scale small target detection method in the traffic scenario as described above.

[0057] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a server, the method for detecting large-scale small targets in a traffic scenario as described above is implemented.

[0058] In a fifth aspect, an embodiment of the present application provides a computer program product, which includes instructions, and when the instructions are executed by a computer, the computer implements the method as described above. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0060] Figure 1 A flowchart of a large-scale small target detection method in a traffic scenario provided in an embodiment of the present application;

[0061] Figure 2 It is a structural schematic diagram of a large-scale small target detection device in a traffic scenario provided by an embodiment of the present application;

[0062] Figure 3It is a structural schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0063] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

[0064] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of this application, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0065] Please see Figure 1 , Figure 1 A flow chart of a large-scale small target detection method in a traffic scene provided in an embodiment of the present application. The large-scale small target detection method in a traffic scene includes:

[0066] 100. Obtain image information in traffic scenes and extract image features.

[0067] Specifically, image information of a traffic scene is acquired through a camera, and the image information is extracted to obtain row features and column features; wherein the row features and column features are obtained to realize subsequent extraction of target key features.

[0068] It should be noted that according to the requirements of traffic scenarios, a dataset suitable for large-scale small target detection in traffic scenarios is selected, and the dataset should contain a large number of small target samples related to driving scenarios.

[0069] Optionally, the image features in the embodiment of the present application may be image features obtained by processing pictures or videos of traffic scenes.

[0070] Optionally, after acquiring the image information in the traffic scene, the image is subjected to denoising, image enhancement, and frame selection and cropping processing.

[0071] Optional, row features: usually refers to the horizontal features in the image, such as lane lines, the contour of the top of the vehicle, etc.; can be extracted by edge detection (such as Canny edge detection), Hough transform, etc. Column features: focus on the longitudinal features, such as vehicle width, distance between vehicles, longitudinal edges, etc. These features are very useful for understanding vehicle distribution and lane occupancy.

[0072] 200. Based on the vertical and horizontal attention network, the image features are processed using strip pooling to generate a pair of direction-aware feature maps.

[0073] Specifically, the input features of the received image are:

[0074] X=R H×W ;

[0075] Among them, H is the row eigenvalue and W is the column eigenvalue;

[0076] Optionally, in order to utilize precise location information and capture long-range dependencies in space, strip pooling is used to average all feature values ​​in rows or columns, unlike two-dimensional average pooling. Specifically, for the input feature X = R H×W , using (H, 1) and (1, W) pooling kernels respectively, encode each channel horizontally and vertically to obtain the output after horizontal strip pooling and the output after vertical strip pooling.

[0077] Specifically, based on the vertical and horizontal attention network, the first pooling kernel is used to process the input features of the image, and each channel is encoded horizontally to obtain the output feature value after horizontal strip pooling:

[0078]

[0079] Use the second pooling kernel to process the input features of the image, encode each channel along the vertical direction, and obtain the output feature value after vertical strip pooling:

[0080]

[0081] Where h is any value of the row eigenvalue, w is any value of the column eigenvalue, 0≤i <h,0≤j<w;

[0082] Features are aggregated horizontally and vertically respectively to generate a pair of direction-aware feature maps after pooling.

[0083] Through the transformation in two directions, features can be aggregated along two spatial directions respectively to generate a pair of direction-aware feature maps, which allows the attention block to capture long-distance dependencies in one spatial direction and retain precise location information in another spatial direction, so that the network can more accurately locate the target of interest, achieve a global receptive field and encode precise location information. In addition, it can effectively capture the relationship between channels.

[0084] 300. Process the pooled feature maps respectively and output a first convolution feature function.

[0085] Specifically, the perceptual feature maps in the two directions after pooling are processed by the first convolution change function and the second convolution change function respectively, and the processed features are spliced ​​to obtain spliced ​​features; the spliced ​​features are processed based on the third convolution change function, and the first convolution feature function is output:

[0086] f=δ(F3[F1(Z h),F2(Z w )]);

[0087] Among them, [F1(Z h ),F2(Z w )] represents the cascade operation along the spatial dimension, i.e., concate, δ is a nonlinear activation function, F1 is the first convolution transformation function, F2 is the second convolution transformation function, F3 is the third convolution transformation function, and Z h is the horizontal perceptual feature map, Z w It is a perceptual feature map in the vertical direction.

[0088] The intermediate feature maps encoding spatial information in the horizontal and vertical directions are:

[0089] f∈R C / r×(H+W) ;

[0090] Among them, C represents the number of channels, and r represents the coefficient of channel compression.

[0091] Exemplarily, the perceptual feature maps in the two directions after pooling are processed by 13×1 convolution and 7×1 convolution transformation function F1 (first convolution transformation function) and 1×13 convolution and 1×7 convolution transformation function F2 (third convolution transformation function), and then spliced ​​together. Finally, they are sent to the shared 1×1 convolution transformation function F3 (third convolution transformation function) to obtain the final output.

[0092] In the above implementation process, the feature maps are convolved and then spliced, and then further convolved to output intermediate feature maps that encode spatial information in the horizontal and vertical directions.

[0093] 400. Split the output result of the first convolution feature function into two independent tensors according to the spatial dimension, and convolve the two tensors separately.

[0094] Specifically, the first convolution feature function is split into two independent tensors along the spatial dimension, and the horizontal tensor is:

[0095] f h ∈R C / r×H ;

[0096] The longitudinal tensor is:

[0097] f w ∈R C / r×W ;

[0098] The horizontal tensor is convolved based on the fourth convolution transformation function, and the convolved function is transformed into a tensor with the same number of channels as the input feature:

[0099] g h =σ(Fh (f h ));

[0100] The vertical tensor is convolved based on the fifth convolution transformation function, and the convolved function is transformed into a tensor with the same number of channels as the input feature:

[0101] g w =σ(F w (f w ));

[0102] Where σ is the sigmoid activation function.

[0103] Exemplarily, using 7×1 convolution and 5×1 convolution (the fourth convolution transformation function), 1×7 convolution and 1×5 convolution (the fifth convolution transformation function), the horizontal tensor and the vertical tensor are convolved separately, and the horizontal tensor and the vertical tensor are transformed into tensors with the same number of channels as the input feature X.

[0104] In the above, the obtained intermediate feature map is split into two independent tensors, and then transformed into tensors with the same number of channels as the input features after convolution. The horizontal and vertical codes are added to the input tensor at the same time. In these two coding maps, each element reflects whether the target of interest exists in the corresponding row and column. Through this encoding process, the module can more accurately locate the exact position of the target of interest, thereby helping the entire model to better identify it.

[0105] 500. The two tensors after convolution are processed based on the input features to generate a final convolution feature function to extract the target key features.

[0106] Specifically, the two tensors after convolution are processed based on the input features, and the number of channels of the input features is reduced by setting the reduction ratio to obtain the final convolution feature function:

[0107]

[0108] Optionally, in order to reduce the complexity of the overhead model, the number of channels of the input features is reduced by setting the reduction ratio r=32.

[0109] As described above, the embodiment of the present application obtains image information in a traffic scene and extracts image features; based on the vertical and horizontal attention network, the image features are processed using strip pooling to generate a pair of direction-aware feature maps; the pooled feature maps are processed separately to output a first convolution feature function; the output result of the first convolution feature function is split into two independent tensors according to the spatial dimension, and the two tensors are convolved separately; the two tensors after convolution are processed based on the input features to generate a final convolution feature function to extract the key features of the target; the correlation between the features is strengthened, so that small targets with random distribution and poor discriminability are enhanced in features, so that the network can more accurately capture the key features of the target and alleviate the problems of small target positioning and missed detection.

[0110] The above steps are not to be performed in a strict order as described in the numbers, but should be understood as an overall solution.

[0111] In the second aspect, based on the above embodiments, Figure 2 This is a schematic diagram of the structure of a large-scale small target detection device in a traffic scene provided by an embodiment of the present application. Figure 2 The large-scale small target detection device in the traffic scene provided by this embodiment specifically includes: an image acquisition module 201, an image processing module 202, a first convolution module 203, a second convolution module 204 and a third convolution module 205.

[0112] Among them, the image acquisition module 201 is used to acquire image information in traffic scenes and extract image features; the image processing module 202 is used to process image features using strip pooling based on the vertical and horizontal attention networks to generate a pair of direction-aware feature maps; the first convolution module 203 is used to process the pooled feature maps separately and output the first convolution feature function; the second convolution module 204 is used to split the output result of the first convolution feature function into two independent tensors according to the spatial dimension, and convolve the two tensors separately; the third convolution module 205 is used to process the two tensors after convolution based on the input features to generate the final convolution feature function to extract the target key features.

[0113] As described above, the embodiment of the present application obtains image information in a traffic scene and extracts image features; based on the vertical and horizontal attention network, the image features are processed using strip pooling to generate a pair of direction-aware feature maps; the pooled feature maps are processed separately to output a first convolution feature function; the output result of the first convolution feature function is split into two independent tensors according to the spatial dimension, and the two tensors are convolved separately; the two tensors after convolution are processed based on the input features to generate a final convolution feature function to extract the key features of the target; the correlation between the features is strengthened, so that small targets with random distribution and poor discriminability are enhanced in features, so that the network can more accurately capture the key features of the target and alleviate the problems of small target positioning and missed detection.

[0114] In the embodiment of the present application, a spatial attention network based on Horizontal and Vertical convolution (SHV) is designed in the lateral connection, which realizes long-distance modeling with a very small number of parameters, and uses feature correlation to capture small targets that are omitted in the convolution process, providing richer and more accurate features for feature fusion of subsequent image processing.

[0115] The large-scale small target detection device in a traffic scene provided in an embodiment of the present application can be used to execute the large-scale small target detection method in a traffic scene provided in the above embodiment, and has corresponding functions and beneficial effects.

[0116] In a third aspect, an embodiment of the present application further provides an electronic device that can integrate the large-scale small target detection device in the traffic scenario provided by the embodiment of the present application. Figure 3 Schematic diagram of the structure of an electronic device provided by an embodiment of the present application. Figure 3 The electronic device includes: an input device 33, an output device 34, a memory 32 and one or more processors 31; the memory 32 is used to store one or more programs; when the one or more programs are executed by the one or more processors 41, the one or more processors 31 implement the large-scale small target detection method in the traffic scene provided by the above embodiment. The input device 33, the output device 34, the memory 32 and the processor 31 can be connected by a bus or other means. Figure 3 The example of connecting through bus is taken in the following.

[0117] The processor 31 executes various functional applications and data processing of the device by running the software programs, instructions and modules stored in the memory 32, that is, realizes the large-scale small target detection method in the above-mentioned traffic scene.

[0118] The electronic device provided above can be used to execute the large-scale small target detection method in the traffic scene provided in the above embodiment, and has corresponding functions and beneficial effects.

[0119] In a fourth aspect, an embodiment of the present application also provides a computer-readable storage medium, which includes a stored computer program; wherein, when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the large-scale small target detection method in the traffic scenario as described above, and can achieve the same beneficial effects.

[0120] Of course, the storage medium containing computer executable instructions provided in an embodiment of the present application is not limited to the large-scale small target detection method in the traffic scenario as described above, and can also execute related operations in the large-scale small target detection method in the traffic scenario provided in any embodiment of the present application.

[0121] In a fifth aspect, the embodiments of the present application also provide a computer program product, and the methods described in the various embodiments of the present application can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer programs or instructions. When the computer program or instruction is loaded and executed on a computer, the process or function described in the various embodiments of the present application is executed in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, a network device, a user device, a core network device, an OAM (Open Application Model) or other programmable device.

[0122] The computer program or instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer program or instructions may be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired or wireless means. The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium, such as a floppy disk, a hard disk, or a magnetic tape; it may also be an optical medium, such as a digital video disk; it may also be a semiconductor medium, such as a solid-state drive. The computer-readable storage medium may be a volatile or non-volatile storage medium, or may include both volatile and non-volatile types of storage media.

[0123] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for implementing 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 consecutive boxes 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 and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0124] In addition, the functional modules in the various embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

[0125] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application 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 for enabling an electronic device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0126] The above description is only an embodiment of the present application and is not intended to limit the scope of protection of the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings.

[0127] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0128] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

Claims

1. A large-scale small target detection method in a traffic scene, characterized in that: include: Obtain image information in traffic scenes and extract image features; Based on the vertical and horizontal attention network, the image features are processed using strip pooling to generate a pair of direction-aware feature maps; Process the pooled feature maps respectively and output the first convolution feature function; Split the output of the first convolution feature function into two independent tensors according to the spatial dimension, and convolve the two tensors separately; The two tensors after convolution are processed based on the input features to generate the final convolution feature function to extract the target key features.

2. The large-scale small target detection method in traffic scenes according to claim 1 is characterized in that: Based on the vertical and horizontal attention network, the image features are processed by strip pooling to generate a pair of direction-aware feature maps, including: Receive input features of an image: X=R H×W ; Among them, H is the row eigenvalue and W is the column eigenvalue; Based on the vertical and horizontal attention network, the first pooling kernel is used to process the input features of the image, and each channel is encoded horizontally to obtain the output feature value after horizontal strip pooling: Use the second pooling kernel to process the input features of the image, encode each channel along the vertical direction, and obtain the output feature value after vertical strip pooling: Where h is any value of the row eigenvalue, w is any value of the column eigenvalue, 0≤i <h,0≤j<w; Features are aggregated along the horizontal and vertical directions respectively to generate a pair of direction-aware feature maps after pooling.

3. The large-scale small target detection method in traffic scenes according to claim 1 is characterized in that: The processing of the pooled feature maps respectively and outputting a first convolution feature function comprises: The perceptual feature maps in the two directions after pooling are processed by the first convolution transformation function and the second convolution transformation function respectively, and the processed features are spliced ​​to obtain spliced ​​features; The concatenated features are processed based on the third convolution change function, and the first convolution feature function is output: in, represents the cascade operation along the spatial dimension, δ is a nonlinear activation function, F1 is the first convolution transformation function, F2 is the second convolution transformation function, F3 is the third convolution transformation function, and Z h is the horizontal perceptual feature map, Z w It is a perceptual feature map in the vertical direction; The intermediate feature maps encoding spatial information in the horizontal and vertical directions are: f∈R C / r×(H+W) ; Among them, C represents the number of channels, and r represents the coefficient of channel compression.

4. The large-scale small target detection method in traffic scenes according to claim 2 is characterized in that: The step of splitting the output result of the first convolution feature function into two independent tensors according to the spatial dimension and convolving the two tensors separately includes: The first convolution feature function is split into two independent tensors according to the spatial dimension. The horizontal tensor is: f h ∈R C / r×H ; The longitudinal tensor is: f w ∈R C / r×W ; The horizontal tensor is convolved based on the fourth convolution transformation function, and the convolved function is transformed into a tensor with the same number of channels as the input feature: g h =σ(F h (f h )); The vertical tensor is convolved based on the fifth convolution transformation function, and the convolved function is transformed into a tensor with the same number of channels as the input feature: g w =σ(F w (f w )); Among them, σ is the activation function.

5. The large-scale small target detection method in traffic scenes according to claim 4 is characterized in that: The two tensors after convolution are processed based on the input features to generate the final convolution feature function to extract the target key features, including: The two tensors after convolution are processed based on the input features, and the number of channels of the input features is reduced by setting the reduction ratio to obtain the final convolution feature function:

6. The large-scale small target detection method in traffic scenes according to claim 1 is characterized in that: The obtaining of image information in a traffic scene and extracting image features include: The image information of the traffic scene is obtained through the camera, and the image information is extracted to obtain row features and column features.

7. A large-scale small target detection device in a traffic scene, characterized in that: include: Image acquisition module, used to acquire image information in traffic scenes and extract image features; An image processing module is used to process image features using strip pooling based on the vertical and horizontal attention networks to generate a pair of direction-aware feature maps; A first convolution module, used to process the pooled feature maps respectively and output a first convolution feature function; A second convolution module is used to split the output result of the first convolution feature function into two independent tensors according to the spatial dimension, and convolve the two tensors separately; The third convolution module is used to process the two tensors after convolution based on the input features to generate the final convolution feature function to extract the target key features.

8. An electronic device, characterized in that: include: A processor, a memory and a bus, wherein the processor is connected to the memory via the bus, and the memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the method for large-scale small target detection in a traffic scenario as described in any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a server, implements the large-scale small target detection method in a traffic scenario as described in any one of claims 1-6.

10. A computer program product, characterized in that The computer program product comprises instructions which, when executed by a computer, cause the computer to implement the method according to any one of claims 1-6.

Images