Lightweight SAR (Synthetic Aperture Radar) image segmentation network system, segmentation method, equipment and medium

Abstract

The invention provides a lightweight SAR image segmentation network system, a segmentation method, a device and a medium, and relates to the technical field of image processing, the system comprises a plurality of cascaded down-sampling modules configured to perform multiple times of down-sampling on an SAR image to obtain first feature maps of different scales; the cavity space pyramid pooling module is configured to perform cavity convolution on the first feature map output by the last stage of down-sampling and perform pooling on different scales to obtain a second feature map; the feature fusion module is configured to perform adaptive weighted fusion on the first feature map and the second feature map to obtain a third feature map; the lightweight channel attention module is configured to adaptively adjust the weight of each channel in the third feature map based on a lightweight channel attention mechanism to obtain a fourth feature map; and the up-sampling module is used for carrying out up-sampling on the fourth feature map to obtain a segmented image, and the segmented image comprises a segmentation result of the target object.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a lightweight SAR image segmentation network system, a segmentation method, a device and a medium. Background Art

[0002] The melting of polar sea ice has a significant impact on human activities. Real-time monitoring of sea ice is crucial to guiding maritime activities, preventing disasters and reducing losses. Using synthetic aperture radar (SAR) sea ice images and deep learning methods to classify sea ice has become a key technology for Arctic route planning and emergency response.

[0003] Convolutional neural networks have shown great potential in the recognition and classification of sea ice. Researchers began to use the basic CNN architecture to achieve preliminary distinction of sea ice based on the difference in backscatter signals of sea ice in SAR images. This method can automatically extract useful features from complex SAR images, reduce human intervention, and is favored for its good robustness. Image segmentation can accurately locate and identify which object each pixel in the image belongs to, which is crucial for understanding the detailed content of the image. Therefore, the SAR sea ice segmentation method based on convolutional neural networks is very suitable for sea ice classification in Arctic shipping routes.

[0004] At present, most of the deep learning-based sea ice segmentation methods tend to optimize the network structure or use multi-source data to improve classification accuracy. Remote sensing image acquisition time is short and the data volume is large. In scenarios where timeliness is not required, the classification accuracy can be improved by increasing the complexity of the model and increasing the data source. However, this method is not applicable to emergency situations, such as real-time route planning in the Arctic and emergency response to sea ice disasters at sea. Summary of the invention

[0005] In view of this, the present invention provides a lightweight SAR image segmentation network system, segmentation method, device and medium, which are used to at least partially solve the above technical problems.

[0006] A first aspect of the present invention provides a lightweight SAR image segmentation network system, comprising: a plurality of cascaded downsampling modules, configured to perform multiple downsampling on a SAR image to obtain first feature maps of different scales; a dilated spatial pyramid pooling module, configured to perform dilated convolution on the first feature map output by the last downsampling stage and perform pooling at different scales to obtain a second feature map; a feature fusion module, configured to perform adaptive weighted fusion on the first feature map and the second feature map to obtain a third feature map; a lightweight channel attention module, configured to adaptively adjust the weights of each channel in the third feature map based on a lightweight channel attention mechanism to obtain a fourth feature map; and an upsampling module, used to upsample the fourth feature map to obtain a segmented image, wherein the segmented image includes a segmentation result of a target object.

[0007] According to an embodiment of the present invention, a dilated spatial pyramid pooling module includes: a parallel dilated convolution branch and a pyramid pooling branch, a first fusion unit and a first convolution unit; the dilated convolution branch is configured to obtain contextual semantic information of a first feature map output by the last level of downsampling to obtain a first fused feature map; the pyramid pooling branch is configured to obtain information of different regions of the first feature map output by the last level of downsampling to obtain a second fused feature map; the first fusion unit is configured to fuse the first fused feature map, the first feature map output by the last level of downsampling, and the second fused feature map to obtain a fifth feature map; the first convolution unit is configured to perform convolution processing on the fifth feature map to compress the number of channels of the fifth feature map to obtain a second feature map.

[0008] According to an embodiment of the present invention, the hole convolution branch includes multiple parallel hole convolution units with different expansion rates, a second fusion unit and a second convolution unit; multiple parallel hole convolution units with different expansion rates are configured to capture the contextual semantic information of the first feature map output by the last level of downsampling based on different receptive fields; the first fusion unit is configured to fuse the captured contextual semantic information; the second convolution unit is configured to perform convolution processing on the fused contextual semantic information to obtain a first fused feature map; the pyramid pooling branch includes multiple parallel adaptive pooling layers with different scales, a third fusion unit and a third convolution unit; multiple parallel adaptive pooling layers with different scales are configured to pool the first feature map output by the last level of downsampling at different scales to obtain information of different regions; the third fusion unit is configured to fuse information of different regions; the third convolution unit is configured to perform convolution processing on the fused information of different regions to obtain a second fused feature map.

[0009] According to an embodiment of the present invention, a feature fusion module includes: a first upsampling unit, configured to upsample the second feature map to match the resolution of the first feature map; a fourth fusion unit, configured to perform element-wise addition of the first feature map and the upsampled second feature map to obtain a third fused feature map; a first branch, configured to obtain global features from the third fused feature map; a second branch, configured to obtain local features from the third fused feature map; a first activation layer, configured to determine a first weight corresponding to the first feature map and a second weight corresponding to the second feature map based on the fused global features and local features; and a fourth fusion unit, configured to perform weighted fusion of the first feature map and the second feature map based on the first weight and the second weight to obtain a third feature map.

[0010] According to an embodiment of the present invention, a lightweight channel attention module includes a global average pooling unit, a 1×1 convolution layer, a second activation layer and a fusion layer connected in sequence, and the input of the fusion layer is the output of the third feature map and the second activation layer; the global average pooling unit, the 1×1 convolution layer, and the second activation layer adaptively adjust the weights of each channel in the third feature map based on a lightweight channel attention mechanism.

[0011] According to an embodiment of the present invention, the convolution processing of multiple cascaded downsampling modules is depth-wise separable convolution, and the depth-wise separable convolution operation includes depth-wise convolution and point-wise convolution.

[0012] According to an embodiment of the present invention, a convolution kernel is independently applied to each input channel in the deep convolution; the point-by-point convolution uses a 1×1 convolution kernel to convolve the output of the deep convolution to fuse features of different channels.

[0013] The second aspect of the present invention provides a lightweight SAR image segmentation method, which is implemented based on the above-mentioned lightweight SAR image segmentation network system, including: performing multiple downsampling on the SAR image to obtain first feature maps of different scales; performing hole convolution on the first feature map output by the last level of downsampling and pooling at different scales to obtain a second feature map; performing adaptive weighted fusion on the first feature map and the second feature map to obtain a third feature map; based on an over-lightweight channel attention mechanism, adaptively adjusting the weights of each channel in the third feature map to obtain a fourth feature map; upsampling the fourth feature map to obtain a segmented image, and the segmented image includes a segmentation result of the target object.

[0014] A third aspect of the present invention provides an electronic device, comprising: one or more processors; a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the above method.

[0015] A fourth aspect of the present invention provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enables the processor to implement the above method.

[0016] The lightweight SAR image segmentation network system, segmentation method, device and medium according to the embodiments of the present invention have at least the following beneficial effects:

[0017] By performing atrous convolution on the image and pooling it at different scales through the atrous spatial pyramid pooling module and adaptively fusing the image based on the feature fusion module, richer features can be extracted and the detail segmentation effect of the image can be improved, thereby taking into account both classification accuracy and inference speed. Compared with the existing mainstream segmentation systems, both classification accuracy and inference speed are significantly improved.

[0018] A lightweight channel attention module is introduced in the decoding part to enhance the model's attention to key features by assigning importance weights to each channel. This mechanism not only improves the performance of the model in tasks such as image classification and semantic segmentation, but also, due to its small number of parameters and small amount of computation, is very suitable for integration into lightweight networks to improve efficiency without sacrificing too much accuracy.

[0019] Dividing the dilated convolution and pyramid pooling into two parallel branches enables the module to process multiple different feature data at the same time, reducing the data transmission within the model and improving the model operation efficiency.

[0020] An adaptive feature fusion module is used to replace the original simple splicing step. This module autonomously learns feature weights by fusing global features and local features, adaptively fuses shallow and deep features, and obtains richer semantic information.

[0021] A lightweight feature network is used to replace the original feature extraction network, and a depth-wise separable convolution is introduced to replace ordinary convolution. This can significantly reduce the number of parameters and the amount of calculation while having little impact on model performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The above and other objects, features and advantages of the present invention will become more apparent through the following description of the embodiments of the present invention with reference to the accompanying drawings, in which:

[0023] Figure 1 A structural block diagram of a lightweight SAR image segmentation network system according to an embodiment of the present invention is schematically shown;

[0024] Figure 2 The structural block diagram of the atrous spatial pyramid pooling module according to an embodiment of the present invention is schematically shown;

[0025] Figure 3A structural block diagram of a feature fusion module according to an embodiment of the present invention is schematically shown;

[0026] Figure 4 A schematic diagram of a structure of a lightweight channel attention module according to an embodiment of the present invention is shown;

[0027] Figure 5 The flowchart of the lightweight SAR image segmentation method according to an embodiment of the present invention is schematically shown.

[0028] Figure 6 A block diagram of an electronic device suitable for implementing the segmentation method according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0029] Below, embodiments of the present invention will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present invention. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of embodiments of the present invention. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of concepts of the present invention.

[0030] The terms used herein are only for describing specific embodiments and are not intended to limit the present invention. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0031] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0032] When using expressions such as "at least one of A, B, and C, etc.", they should generally be interpreted according to the meaning of the expression commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0033] Figure 1 The structural block diagram of a lightweight SAR image segmentation network system according to an embodiment of the present invention is schematically shown.

[0034] like Figure 1As shown, the lightweight SAR image segmentation network system of this embodiment may include:

[0035] The multiple cascaded downsampling modules are configured to perform multiple downsampling on the SAR image to obtain first feature maps of different scales.

[0036] The atrous spatial pyramid pooling module is configured to perform atrous convolution on the first feature map output by the last level of downsampling and to perform pooling at different scales to obtain a second feature map.

[0037] The feature fusion module is configured to perform adaptive weighted fusion on the first feature map and the second feature map to obtain a third feature map.

[0038] The lightweight channel attention module is configured to adaptively adjust the weights of each channel in the third feature map based on a lightweight channel attention mechanism to obtain a fourth feature map.

[0039] An upsampling module is used to upsample the fourth feature map to obtain a segmented image, wherein the segmented image includes a segmentation result of the target object.

[0040] For example, the model framework of the lightweight SAR image segmentation network system can be based on Deeplabv3+ and improved in a lightweight way, taking into account both classification accuracy and inference speed. The original complex backbone network can be replaced with a lightweight network, for example, the backbone network can be replaced with ResNet18.

[0041] For example, three downsamplings can be performed during the feature extraction process (the specific number of samplings is not limited in the present invention), and three feature maps of different scales of 256×256×C (denoted as F1), 128×128×C (denoted as F2), and 64×64×C (denoted as F3) can be obtained. Among them, F3 can be subsequently input into the Atrous Spatial Pyramid Pooling (ASPP) module to further extract richer semantic features. After that, F1 and F3 can be respectively input into the Feature Fusion Module (FFM) for feature fusion after adjusting the number of channels through a 1×1 convolution module, followed by a lightweight channel attention mechanism module (ECA). Finally, the final output result is obtained after a convolution module and then four times upsampling. Among them, the ASPP module, FFM module and ECA module will be introduced in detail later.

[0042] Through the lightweight SAR image segmentation network system of the embodiment of the present invention, the image is subjected to dilated convolution and pooling at different scales through the dilated spatial pyramid pooling module, and the image is adaptively fused based on the feature fusion module, so that richer features can be extracted and the detail segmentation effect of the image can be improved, thereby taking into account both classification accuracy and reasoning speed.

[0043] On the basis of the above embodiment, the convolution processing in multiple cascaded downsampling modules is depth-separable convolution, and the depth-separable convolution operation includes depth-wise convolution and point-wise convolution. Further, in the depth-wise convolution, a convolution kernel is applied independently to each input channel; the point-wise convolution uses a 1×1 convolution kernel to convolve the output of the depth-wise convolution to fuse the features of different channels.

[0044] Depthwise Separable Convolution is an efficient convolution operation that decomposes the standard convolution operation into two smaller operations: depthwise convolution and pointwise convolution, thereby reducing the number of parameters and computational complexity of the model.

[0045] In deep convolution, each input channel can apply a convolution kernel independently, which means that the network convolves the feature map of each channel without mixing the information between channels. Point-by-point convolution is to convolve the output of deep convolution with a 1×1 convolution kernel, the purpose of which is to combine features from different channels. This step can be seen as a process of feature fusion and channel number adjustment. Through 1×1 convolution, the network can increase or decrease the number of channels, while allowing features between different channels to be combined, enhancing the expressiveness of the features.

[0046] Compared with standard convolution, depthwise separable convolution can greatly reduce the number of model parameters. If the size of the input feature map is H×W×M, and a K×K convolution kernel is used for standard convolution, the number of parameters is K×K×M×N (where N is the number of output channels). The number of parameters of depthwise separable convolution is usually (K×K×M+M×N), which is obviously less than standard convolution.

[0047] On the basis of the above embodiments, the atrous spatial pyramid pooling module includes: a parallel atrous convolution branch and a pyramid pooling branch, a first fusion unit and a first convolution unit; the atrous convolution branch is configured to obtain contextual semantic information of a first feature map output by the last level of downsampling to obtain a first fused feature map; the pyramid pooling branch is configured to obtain information of different regions of the first feature map output by the last level of downsampling to obtain a second fused feature map; the first fusion unit is configured to fuse the first fused feature map, the first feature map output by the last level of downsampling, and the second fused feature map to obtain a fifth feature map; the first convolution unit is configured to perform convolution processing on the fifth feature map to compress the number of channels of the fifth feature map to obtain a second feature map.

[0048] Furthermore, the dilated convolution branch may include multiple parallel dilated convolution units with different dilated rates, a second fusion unit, and a second convolution unit; multiple parallel dilated convolution units with different dilated rates are configured to capture contextual semantic information of the first feature map output by the last level of downsampling based on different receptive fields; the first fusion unit is configured to fuse the captured contextual semantic information; the second convolution unit is configured to perform convolution processing on the fused contextual semantic information to obtain a first fused feature map. The pyramid pooling branch includes multiple parallel adaptive pooling layers with different scales, a third fusion unit, and a third convolution unit; multiple parallel adaptive pooling layers with different scales are configured to pool the first feature map output by the last level of downsampling at different scales to obtain information of different regions; the third fusion unit is configured to fuse information of different regions; the third convolution unit is configured to perform convolution processing on the fused information of different regions to obtain a second fused feature map.

[0049] Figure 2 The structure block diagram of the atrous spatial pyramid pooling module according to an embodiment of the present invention is schematically shown.

[0050] like Figure 2As shown in the figure, ASPP consists of two parts: the Atrous Convolution Module (ACM) and the Pyramid Pooling Module (PPM), which together constitute the ASPP module in deeplabV3+. The Atrous Convolution Module expands the receptive field without losing resolution by setting different dilation rates and captures more contextual semantic information. The Pyramid Pooling Module can fuse information from different sub-regions by pooling at different scales, which helps the network capture the features of objects at different scales, thereby improving the recognition ability of objects of different sizes. Considering that the parallel structure can process features of different scales at the same time, it may improve computational efficiency because it allows multiple operations to be performed simultaneously and reduces the number of data transfers in the network. Therefore, the Atrous Convolution Module and the Pyramid Pooling Module in the ASPP module can be designed as two parallel branches for processing, which are called the ACM branch and the PPM branch respectively. Each branch processes the input feature map independently. After the processing is completed, the outputs of the two branches are spliced ​​with the original input feature map. Finally, the concatenated feature maps are compressed through a 1×1 convolutional layer to reduce the number of channels, thereby generating the final output of the ASPP module. This structure not only enhances the diversity of features, but also improves the model's ability to integrate information of different scales.

[0051] Based on the above embodiments, the feature fusion module includes: a first upsampling unit, configured to upsample the second feature map to match the resolution of the first feature map; a fourth fusion unit, configured to perform element-wise addition of the first feature map and the upsampled second feature map to obtain a third fused feature map; a first branch, configured to obtain global features from the third fused feature map; a second branch, configured to obtain local features from the third fused feature map; a first activation layer, configured to determine a first weight corresponding to the first feature map and a second weight corresponding to the second feature map based on the fused global features and local features; and a fourth fusion unit, configured to perform weighted fusion of the first feature map and the second feature map based on the first weight and the second weight to obtain a third feature map.

[0052] Figure 3 The structural block diagram of the feature fusion module according to an embodiment of the present invention is schematically shown.

[0053] like Figure 3As shown in the figure, the FFM card can innovate the feature fusion strategy in the deeplabV3+ network and introduce an adaptive feature fusion mechanism. The module receives two sizes of feature maps as input: one is a shallow feature map with higher resolution, and the other is a deep feature map with lower resolution. During the fusion process, the deep feature map is first upsampled to match the resolution of the shallow feature map, and then the two are element-wise added. The added data is fed into two parallel branches: one path is used to capture global features and the other path is used to extract local features. The features extracted by these two paths are then merged and the sigmoid activation function is used to calculate their respective fusion weights. Finally, the calculated weights are applied to the shallow and deep feature maps respectively, and the output of the FFM module is obtained by weighted summation, thereby achieving effective fusion of feature maps.

[0054] The first branch may include, for example, a 1×1 convolution, a normalization layer, an activation layer, a 1×1 convolution, and a normalization layer connected in sequence. The second branch may include, for example, an adaptive pooling layer, a 1×1 convolution, a normalization layer, an activation layer, a 1×1 convolution, and a normalization layer connected in sequence.

[0055] Based on the above embodiment, the lightweight channel attention module includes a global average pooling unit, a 1×1 convolution layer, a second activation layer and a fusion layer connected in sequence, and the input of the fusion layer is the output of the third feature map and the second activation layer; the global average pooling unit, the 1×1 convolution layer, and the second activation layer are based on an over-lightweight channel attention mechanism to adaptively adjust the weights of each channel in the third feature map.

[0056] Figure 4 The structural block diagram of a lightweight channel attention module according to an embodiment of the present invention is schematically shown.

[0057] like Figure 4 As shown in the figure, after the data is processed by the FFM module, it is input into the ECA module to add an attention mechanism to the feature map according to the channel dimension. The ECA module is a lightweight channel attention module, which is an improvement on the SE module. It uses one-dimensional convolution to replace the fully connected layer, greatly reducing the number of parameters. In general, the ECA module uses a lightweight channel attention mechanism to adaptively adjust the weights of each channel in the feature map, thereby enhancing the model's ability to capture key information, improving the quality of feature representation and the generalization performance of the model, while maintaining a low computational cost, and is suitable for integration into various deep learning models to improve efficiency and performance.

[0058] Based on the above lightweight SAR image segmentation network system, an embodiment of the present invention further provides a lightweight SAR image segmentation method.

[0059] Figure 5The flowchart of the lightweight SAR image segmentation method according to an embodiment of the present invention is schematically shown.

[0060] like Figure 5 As shown, the lightweight SAR image segmentation method may include operations S510 to S520.

[0061] In operation S510, the SAR image is downsampled multiple times to obtain first feature maps of different scales.

[0062] In operation S520, a dilated convolution is performed on the first feature map output by the last level of downsampling, and pooling is performed at different scales to obtain a second feature map.

[0063] In operation S530, adaptive weighted fusion is performed on the first feature map and the second feature map to obtain a third feature map.

[0064] In operation S540, based on the lightweight channel attention mechanism, the weights of each channel in the third feature map are adaptively adjusted to obtain a fourth feature map.

[0065] In operation S550, the fourth feature map is upsampled to obtain a segmented image, where the segmented image includes a segmentation result of the target object.

[0066] It should be noted that the lightweight SAR image segmentation method part in the embodiment of the present invention corresponds to the lightweight SAR image segmentation network system part in the embodiment of the present invention, and their specific implementation details are also the same, which will not be repeated here.

[0067] Figure 6 A block diagram of an electronic device suitable for implementing the segmentation method according to an embodiment of the present disclosure is schematically shown.

[0068] like Figure 6 As shown, the electronic device 600 according to an embodiment of the present disclosure includes a processor 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage part 608 to a random access memory (RAM) 603. The processor 601 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a dedicated microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 601 may also include an onboard memory for caching purposes. The processor 601 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0069] In RAM 603, various programs and data required for the operation of electronic device 600 are stored. Processor 601, ROM 602 and RAM 603 are connected to each other via bus 604. Processor 601 performs various operations of the method flow according to the embodiment of the present disclosure by executing the program in ROM 602 and / or RAM 603. It should be noted that the program can also be stored in one or more memories other than ROM 602 and RAM 603. Processor 601 can also perform various operations of the method flow according to the embodiment of the present disclosure by executing the program stored in the one or more memories.

[0070] According to an embodiment of the present disclosure, the electronic device 600 may further include an input / output (I / O) interface 605, which is also connected to the bus 604. The electronic device 600 may further include one or more of the following components connected to the I / O interface 605: an input portion 606 including a keyboard, a mouse, etc.; an output portion 607 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage portion 608 including a hard disk, etc.; and a communication portion 609 including a network interface card such as a LAN card, a modem, etc. The communication portion 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 portion 608 as needed.

[0071] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or may exist independently without being assembled into the device / apparatus / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present disclosure is implemented.

[0072] According to an embodiment of the present disclosure, a computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, an apparatus or a device. For example, according to an embodiment of the present disclosure, a computer-readable storage medium may include the ROM 602 and / or the RAM 603 described above and / or one or more memories other than the ROM 602 and the RAM 603.

[0073] The embodiment of the present disclosure also includes a computer program product, which includes a computer program, and the computer program contains program code for executing the method shown in the flowchart. When the computer program product is run in a computer system, the program code is used to enable the computer system to implement the processing method provided by the embodiment of the present disclosure.

[0074] The above functions defined in the system / device of the embodiment of the present disclosure are performed when the computer program is executed by the processor 601. According to the embodiment of the present disclosure, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0075] In one embodiment, the computer program may rely on tangible storage media such as optical storage devices, magnetic storage devices, etc. In another embodiment, the computer program may also be transmitted and distributed in the form of signals on a network medium, and downloaded and installed through the communication part 609, and / or installed from a removable medium 611. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0076] In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 609, and / or installed from the removable medium 611. When the computer program is executed by the processor 601, the above functions defined in the system of the embodiment of the present disclosure are performed. According to the embodiment of the present disclosure, the system, device, apparatus, module, unit, etc. described above can be implemented by a computer program module.

[0077] According to an embodiment of the present disclosure, the program code for executing the computer program provided by the embodiment of the present disclosure can be written in any combination of one or more programming languages. Specifically, these computing programs can be implemented using high-level process and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, Java, C++, python, "C" language or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on the remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using an Internet service provider to connect through the Internet).

[0078] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a 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 flow chart, and the combination of the boxes in the block diagram or flow chart 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.

[0079] It will be appreciated by those skilled in the art that the features described in the various embodiments of the present disclosure may be combined and / or combined in a variety of ways, even if such combinations and / or combinations are not explicitly described in the present disclosure. In particular, without departing from the spirit and teachings of the present disclosure, the features described in the various embodiments of the present disclosure may be combined and / or combined in a variety of ways. All of these combinations and / or combinations fall within the scope of the present disclosure.

[0080] The embodiments of the present disclosure are described above. However, these embodiments are only for illustrative purposes and are not intended to limit the scope of the present disclosure. Although the embodiments are described above, this does not mean that the measures in the various embodiments cannot be used in combination to advantage. The scope of the present disclosure is limited by the attached embodiments and their equivalents. Without departing from the scope of the present disclosure, those skilled in the art may make a variety of substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. A lightweight SAR image segmentation network system, characterized in that: include: A plurality of cascaded downsampling modules are configured to perform multiple downsampling on the SAR image to obtain first feature maps of different scales; A dilated spatial pyramid pooling module is configured to perform dilated convolution on the first feature map output by the last level of downsampling and perform pooling at different scales to obtain a second feature map; A feature fusion module is configured to perform adaptive weighted fusion on the first feature map and the second feature map to obtain a third feature map; A lightweight channel attention module is configured to adaptively adjust the weights of each channel in the third feature map based on a lightweight channel attention mechanism to obtain a fourth feature map; An upsampling module is used to upsample the fourth feature map to obtain a segmented image, wherein the segmented image includes a segmentation result of the target object.

2. The system according to claim 1, characterized in that The atrous spatial pyramid pooling module comprises: a parallel atrous convolution branch and a pyramid pooling branch, a first fusion unit and a first convolution unit; The atrous convolution branch is configured to obtain contextual semantic information of the first feature map output by the last level of downsampling to obtain a first fused feature map; The pyramid pooling branch is configured to obtain information of different regions of the first feature map output by the last level of downsampling to obtain a second fused feature map; A first fusion unit is configured to fuse the first fused feature map, the first feature map output by the last level of downsampling, and the second fused feature map to obtain a fifth feature map; The first convolution unit is configured to perform convolution processing on the fifth feature map to compress the number of channels of the fifth feature map to obtain the second feature map.

3. The system according to claim 2, characterized in that The atrous convolution branch includes a plurality of parallel atrous convolution units with different dilation rates, a second fusion unit, and a second convolution unit; the plurality of parallel atrous convolution units with different dilation rates are configured to capture contextual semantic information of the first feature map output by the last level of downsampling based on different receptive fields; The first fusion unit is configured to fuse the captured contextual semantic information; the second convolution unit is configured to perform convolution processing on the fused contextual semantic information to obtain a first fused feature map; The pyramid pooling branch includes multiple parallel adaptive pooling layers of different scales, a third fusion unit and a third convolution unit; the multiple parallel adaptive pooling layers of different scales are configured to pool the first feature map output by the last level of downsampling at different scales to obtain information of different regions; the third fusion unit is configured to fuse information of different regions; the third convolution unit is configured to perform convolution processing on the fused information of different regions to obtain a second fused feature map.

4. The system according to claim 1, characterized in that The feature fusion module includes: A first upsampling unit is configured to upsample the second feature map to match a resolution of the first feature map; a fourth fusion unit, configured to perform element-wise addition of the first feature map and the upsampled second feature map to obtain a third fused feature map; A first branch is configured to obtain a global feature from the third fused feature map; A second branch is configured to obtain local features from the third fused feature map; A first activation layer is configured to determine a first weight corresponding to the first feature map and a second weight corresponding to the second feature map based on the fused global features and local features; The fourth fusion unit is configured to perform weighted fusion on the first feature map and the second feature map based on the first weight and the second weight to obtain the third feature map.

5. The system according to claim 1, characterized in that The lightweight channel attention module includes a global average pooling unit, a 1×1 convolution layer, a second activation layer and a fusion layer connected in sequence, and the input of the fusion layer is the third feature map and the output of the second activation layer; The global average pooling unit, the 1×1 convolutional layer, and the second activation layer adaptively adjust the weights of each channel in the third feature map based on an overly lightweight channel attention mechanism.

6. The system according to any one of claims 1 to 5, characterized in that: The convolution processing of the multiple cascaded downsampling modules is depth-wise separable convolution, and the depth-wise separable convolution operation includes depth-wise convolution and point-wise convolution.

7. The system according to claim 6, characterized in that In the depth convolution, a convolution kernel is applied independently to each input channel; the point-by-point convolution uses a 1×1 convolution kernel to convolve the output of the depth convolution to fuse the features of different channels.

8. A lightweight SAR image segmentation method, implemented based on the lightweight SAR image segmentation network system according to any one of claims 1 to 7, characterized in that: include: The SAR image is downsampled multiple times to obtain the first feature maps of different scales; Perform hole convolution on the first feature map output by the last level of downsampling and perform pooling at different scales to obtain the second feature map; Performing adaptive weighted fusion on the first feature map and the second feature map to obtain a third feature map; Based on an overly lightweight channel attention mechanism, adaptively adjust the weights of each channel in the third feature map to obtain a fourth feature map; The fourth feature map is upsampled to obtain a segmented image, where the segmented image includes a segmentation result of the target object.

9. An electronic device, characterized in that: include: one or more processors; a memory for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method of claim 8.

10. A computer-readable storage medium, characterized in that: Executable instructions are stored thereon, and when the instructions are executed by a processor, the processor implements the method of claim 8.

Images