SAR image feature extraction method, device, equipment, medium and program product

Abstract

The present disclosure provides a SAR image feature extraction method, which can be applied to the field of synthetic aperture radar image processing technology. The method comprises: acquiring a SAR image; inputting the SAR image into a deep neural network for feature extraction to determine the scattering center of the SAR image; wherein the deep neural network comprises a plurality of neurons, the weight parameters of the neurons are determined by the distance from the non-center point of the neurons to the center of the neurons, and the weight parameters of the neurons are fixed. The present disclosure also provides a SAR image feature extraction device, equipment, storage medium and program product.

Description

Technical Field

[0001] This disclosure relates to the field of synthetic aperture radar image processing technology, and specifically to a SAR image feature extraction method, apparatus, device, medium, and program product. Background Technology

[0002] With the development of Synthetic Aperture Radar (SAR) technology, the number of SAR satellites in orbit has increased significantly, and the spatial resolution of SAR images has continued to improve, exhibiting high-definition geometric features similar to satellite optical images. However, the unique imaging mechanism of SAR images makes their interpretation difficult, affecting the extraction of various types of information and the industrial application of SAR images.

[0003] In recent years, artificial intelligence processing methods have begun to be promoted and applied to SAR images, especially deep learning methods, which have achieved certain applications in information extraction and ground feature interpretation of SAR images. However, deep learning methods rely heavily on a large amount of manually labeled sample image data. It is precisely because of a large number of labeled sample images that the weight parameters of artificial neurons in deep networks can be effectively learned.

[0004] Because of the large number of neurons and the numerous weight parameters for each neuron, a large number of samples are needed to effectively determine the weight parameters of these artificial neurons. For SAR images, because SAR images are greatly affected by radar azimuth and incident angle, the number of sample images that need to be manually labeled for the same ground feature will increase exponentially, and the difficulty of sample collection will also increase significantly. Therefore, this seriously restricts the in-depth application of deep learning methods in SAR images. Summary of the Invention

[0005] In view of the above problems, this disclosure provides SAR image feature extraction methods, apparatus, devices, media and program products to improve the efficiency of SAR image feature extraction, and to at least partially solve the above technical problems.

[0006] According to a first aspect of this disclosure, a SAR image feature extraction method is provided, comprising: acquiring a SAR image; inputting the SAR image into a deep neural network for feature extraction to determine the scattering center of the SAR image; wherein the deep neural network includes multiple neurons, the weight parameters of the neurons are determined by the distance from the non-center point of the neuron to the center of the neuron, and the weight parameters of the neurons are fixed.

[0007] According to an embodiment of this disclosure, the neuron includes a first neuron for determining the circular scattering center of a SAR image. Inputting the SAR image into a deep neural network for feature extraction to determine the scattering center of the SAR image includes: determining a first weight based on a first distance from a non-center point of the first neuron to the center of the first neuron; and performing feature extraction on the SAR image based on the first weight to determine the circular scattering center; wherein the first weight is inversely proportional to the first distance in a non-linear manner.

[0008] According to embodiments of this disclosure, inputting a SAR image into a deep neural network for feature extraction to determine the scattering center of the SAR image further includes: determining the size of a first neuron; determining the number of non-center points of the first neuron based on the size of the first neuron; and determining the feature extraction scale of the SAR image based on the number of non-center points of the first neuron.

[0009] According to embodiments of this disclosure, the neuron further includes a second neuron for determining the elliptical scattering center of the SAR image. Inputting the SAR image into the deep neural network for feature extraction to determine the scattering center of the SAR image further includes: determining a second weight based on a second distance from the non-center point of the second neuron to the center of the second neuron and the elliptical direction; and performing feature extraction on the SAR image based on the second weight to determine the elliptical scattering center; wherein the second weight is inversely proportional to the second distance in a non-linear manner, and the elliptical direction includes 0°, 45°, 90°, and 135°.

[0010] According to embodiments of this disclosure, inputting a SAR image into a deep neural network for feature extraction to determine the scattering center of the SAR image further includes: adjusting a second weight based on the elliptic flattening.

[0011] According to embodiments of this disclosure, inputting a SAR image into a deep neural network for feature extraction to determine the scattering center of the SAR image further includes: thresholding the pixel values ​​of the SAR image according to a first weight and / or a second weight to obtain an activation input; and using a regular linear rectified function to activate the activation input to obtain a scattering center feature map of the SAR image.

[0012] A second aspect of this disclosure provides a SAR image feature extraction apparatus, comprising: an acquisition module for acquiring a SAR image; and an extraction module for inputting the SAR image into a deep neural network for feature extraction to determine the scattering center of the SAR image; wherein the deep neural network comprises multiple neurons, the weight parameters of the neurons are determined by the distance from the non-center point of the neuron to the center of the neuron, and the weight parameters of the neurons are fixed.

[0013] A third aspect of this disclosure provides an electronic device comprising: one or more processors; and 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 perform the methods of any of the above embodiments.

[0014] A fourth aspect of this disclosure also provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, cause the processor to perform the methods of any of the above embodiments.

[0015] The fifth aspect of this disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements the methods of any of the above embodiments.

[0016] Compared with existing technologies, the SAR image feature extraction method, apparatus, electronic device, storage medium, and program product provided in this disclosure have at least the following beneficial effects:

[0017] (1) The SAR image feature extraction method disclosed herein directly determines the weight parameters of the neuron by the distance relationship between the non-center point of the neuron and the center of the neuron, so that the deep neural network containing the neurons of this disclosure does not need to perform a large number of sample learnings to accurately extract the scattering center features in the SAR image, which greatly improves the feature extraction efficiency of the SAR image.

[0018] (2) The SAR image feature extraction method disclosed herein is simple and fast. It determines the extraction weight of the circular scattering center based on the nonlinear inverse relationship between the first distance from the non-center point of the neuron to the center of the neuron and the weight.

[0019] (3) The SAR image feature extraction method disclosed herein can also extract features from elliptical scattering centers in SAR images. By increasing the elliptical direction, the feature extraction direction of neurons is determined, thereby improving the accuracy of feature extraction from SAR images. Attached Figure Description

[0020] The foregoing contents, as well as other objects, features, and advantages of this disclosure, will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0021] Figure 1 The illustration schematically depicts application scenarios of SAR image feature extraction methods, apparatus, devices, media, and program products according to embodiments of the present disclosure;

[0022] Figure 2 A flowchart illustrating a SAR image feature extraction method according to an embodiment of the present disclosure is shown schematically.

[0023] Figure 3 A schematic block diagram of a SAR image feature extraction apparatus according to an embodiment of the present disclosure is shown; and

[0024] Figure 4 A block diagram schematically illustrates an electronic device suitable for implementing a SAR image feature extraction method according to an embodiment of the present disclosure. Detailed Implementation

[0025] The embodiments of the present disclosure will now 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 disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of the present disclosure.

[0026] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0027] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0028] When using expressions such as "at least one of A, B, and C", they should generally be interpreted in accordance with the meaning that is commonly understood by a person skilled in the art (e.g., "a system having at least one of A, B, and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B, and C, etc.).

[0029] In the technical solution of this invention, the user information (including but not limited to user personal information, user image information, user device information, such as location information) and data (including but not limited to data used for analysis, stored data, and displayed data) involved are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with the relevant laws, regulations, and standards of the relevant countries and regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0030] Artificial neurons are the basic units that make up neural networks. They mimic the structure and characteristics of biological neurons, receiving a set of input signals and producing an output signal. An artificial neuron can be viewed as a mathematical function or model that transforms input signals into output signals. Each input signal is individually weighted, and the sum is passed through a nonlinear function (such as the sigmoid function) to produce the output signal. In convolutional neural networks, artificial neurons perform convolution operations, that is, they calculate the inner product of the input signals and then input the result into an activation function. The activation function is used to simulate the nonlinear activation function of biological neurons, restricting the neuron's output signal to a specific range. Common activation functions include sigmoid and ReLU (Rectified Linear Unit).

[0031] Figure 1 The illustration shows an application scenario of the SAR image feature extraction method, apparatus, device, medium, and program product according to embodiments of the present disclosure.

[0032] like Figure 1 As shown, application scenario 100 according to this embodiment may include terminal devices 101, 102, and 103, a network 104, and a server 105. Network 104 serves as a medium for providing a communication link between terminal devices 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0033] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, and 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platforms, etc. (for example only). For instance, SAR images to be processed can be provided to server 105 through terminal devices 101, 102, and 103.

[0034] Terminal devices 101, 102, and 103 can be various electronic devices with displays and web browsing capabilities, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0035] Server 105 can be a server providing various services, such as a backend management server supporting websites browsed by users using terminal devices 101, 102, and 103 (this is just an example). The backend management server can analyze and process received user requests and other data, and feed back the processing results (e.g., web pages, information, or data obtained or generated according to user requests) to the terminal devices. In particular, server 105 is equipped with a deep neural network containing neurons from any embodiment of this disclosure, which can provide SAR image feature extraction services.

[0036] It should be noted that the SAR image feature extraction method provided in this disclosure embodiment can generally be executed by server 105. Correspondingly, the SAR image feature extraction device provided in this disclosure embodiment can generally be located in server 105. The SAR image feature extraction method provided in this disclosure embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with terminal devices 101, 102, 103 and / or server 105. Correspondingly, the SAR image feature extraction device provided in this disclosure embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with terminal devices 101, 102, 103 and / or server 105.

[0037] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0038] The following will be based on Figure 1 The described scene, through Figure 2 The SAR image feature extraction method of the disclosed embodiments is described in detail.

[0039] Figure 2 A flowchart illustrating a SAR image feature extraction method according to an embodiment of the present disclosure is shown schematically.

[0040] like Figure 2 As shown, the SAR image feature extraction method of this embodiment includes, for example, operations S210 to S220, and the SAR image feature extraction method can be executed by a computer program on corresponding computer hardware.

[0041] Operate S210 to acquire SAR images.

[0042] In operation S220, the SAR image is input into a deep neural network for feature extraction to determine the scattering center of the SAR image. The deep neural network consists of multiple neurons, and the weight parameters of each neuron are determined by the distance from a non-center point of the neuron to its center, and these weight parameters are fixed.

[0043] For example, since there are a large number of ground object scattering centers on high-resolution SAR images, and the scattering centers of different ground objects will have different distributions on the image, the scattering center is an important feature of the ground object. Therefore, corresponding artificial neurons can be designed according to the scattering centers on the SAR image to perceive the various scattering centers of the ground object.

[0044] For example, a convolutional neural network (CNN) consists of multiple convolutional layers, each containing some neurons.

[0045] To determine the weight parameters of each neuron, an optimization-based iterative calculation method can be employed. This method directly calculates the weight parameters of each neuron based on the statistical characteristics of the SAR image and the scattering properties of ground objects. Specifically, the SAR image is first preprocessed to extract key features such as the shape, size, and orientation of ground objects. Then, based on these features and the scattering properties of the ground objects, mathematical models and algorithms are used to calculate the weight parameters of each neuron.

[0046] After determining the weight parameters for each neuron, this untrained neural network can be used for feature extraction from SAR images. Specifically, the SAR image is input into the CNN, and each convolutional layer uses its internal neurons to sense and respond to various scattering centers in the image. The output is then used as a feature representation of the image for subsequent land cover classification and recognition tasks.

[0047] For example, determining weight parameters based on the distance from the neuron's non-center point to its center point is an intuitive and effective method. The idea behind this method is that different locations of a neuron correspond to different receptive fields in the image; the closer a location is to the neuron's center, the stronger its sensitivity to the image, and therefore, it receives a larger weight when calculating the weights.

[0048] In practical computation, the size and shape of neurons can be determined as needed, and then weights can be calculated based on the distance of each location from the center of the neuron. Locations closer to the center receive a larger weight, while locations farther away receive a smaller weight. This yields a set of distance-based weight parameters for use in convolution operations.

[0049] This method of calculating weights is flexible and adjustable, and can be adjusted and optimized according to specific tasks and data characteristics. At the same time, it has a sound theoretical basis and mathematical interpretability, and can be easily combined with other deep learning techniques to achieve more complex image analysis and processing tasks.

[0050] For example, the method disclosed herein can use neural network models such as U-net and DeepLap to process SAR images.

[0051] U-Net is a commonly used convolutional neural network model consisting of a compression path and an expansion path. It can effectively extract features from images and is used for tasks such as image segmentation and object detection. When processing SAR images, U-Net can be used to extract features of target regions such as scattering centers and perform classification and localization.

[0052] DeepLap is a graph convolutional neural network (GCN) based model that can perform hierarchical feature extraction on images and is used for tasks such as image classification and object detection. When processing SAR images, DeepLap can be used to extract high-level features of target regions such as scattering centers, and then classify and locate them.

[0053] Besides U-Net and DeepLap, there are many other similar neural network models, such as ResNet and DenseNet. ResNet is a residual network that improves network depth and performance by introducing residual blocks. DenseNet is a dense network that improves feature transfer and utilization efficiency by introducing more connections. These models can also be used for tasks such as feature extraction, classification, and localization in SAR image processing. The specific model chosen depends on the application requirements and data characteristics.

[0054] According to embodiments of this disclosure, the neuron includes, for example, a first neuron, for determining the circular scattering center of the SAR image. For example, the scattering center of the SAR image is determined by operations S321 to S322.

[0055] In operation S321, the first weight is determined based on the first distance from the non-center point of the first neuron to the center of the first neuron.

[0056] In operation S322, feature extraction is performed on the SAR image based on a first weight to determine the circular scattering center. The first weight is non-linearly inversely proportional to a first distance.

[0057] For example, consider a SAR image containing a circular scattering center. To extract this circular scattering center, a special first neuron is designed with a shape and size corresponding to the circular scattering center.

[0058] In the feature extraction stage, the SAR image is input into a deep neural network, and a convolution operation is performed using the first neuron. The distance from the non-center point of the first neuron to its center determines the value of the first weight. Specifically, a non-linear inverse weight can be calculated based on this distance, and this weight corresponds to the importance of the circular scattering center.

[0059] For example, calculating the neuron weight parameter w1 for sensing the circular SAR scattering center. i The specific calculation formula is as follows:

[0060]

[0061] Where i is the index number, which takes the integers 1, 2, 3, ..., 9, and the parameter u... i and v i For parameters that appear in pairs, nine values ​​are taken respectively: (u1, v1) takes (-1, 1), (u2, v2) takes (0, 1), (u3, v3) takes (1, 1), (u4, v4) takes (-1, 0), (u5, v5) takes (0, 0), (u6, v6) takes (1, 0), (u7, v7) takes (-1, -1), (u8, v8) takes (0, -1), and (u9, v9) takes (1, -1).

[0062] After calculating the first weight, this weight is used to extract features from the SAR image to determine the location and shape of the circular scattering center. This feature extraction process may include some complex calculations and operations, such as convolution, activation functions, and pooling.

[0063] In this way, the features of circular scattering centers can be extracted from SAR images and weighted in a non-linear manner. This example uses the concept of a first neuron and a first weight, but in practice, multiple neurons and weights can be designed as needed to handle different types of scattering centers and image features.

[0064] It should be noted that this example is just a simplified demonstration; real-world scenarios can be much more complex. For instance, the circular scattering center may not be perfectly circular, but rather exhibit some degree of deformation or distortion. In such cases, more complex neurons and corresponding weight calculation methods are needed to accurately extract these features.

[0065] According to embodiments of this disclosure, the scattering center of a SAR image can also be determined by operating S421 to S423.

[0066] In operation S421, the size of the first neuron is determined.

[0067] In operation S422, the number of non-central points of the first neuron is determined based on the size of the first neuron.

[0068] In operation S423, the feature extraction scale of the SAR image is determined based on the number of non-center points of the first neuron.

[0069] For example, SAR images can be input into a deep neural network for feature extraction to determine the scattering center of the SAR image. In this process, factors such as the size of the first neuron, the number of non-center points, and the feature extraction scale can be further considered.

[0070] First, determine the size of the first neuron. This size can affect the ability to perceive and extract scattering centers in an image. If the neuron size is too large, the scattering centers may be too smooth, losing detail; if the neuron size is too small, the presence of scattering centers may not be fully perceived. Therefore, an appropriate neuron size needs to be selected based on the specific task and data characteristics.

[0071] Next, the number of non-center points is determined based on the size of the first neuron. The number of non-center points refers to the number of pixels outside the neuron's center point. These pixels play a crucial role in the convolution operation, capturing the shape and detail information of the scattering center in the image. Therefore, the ability to perceive and extract image features is determined based on the neuron's size and the number of non-center points.

[0072] Finally, the feature extraction scale of the SAR image is determined based on the number of non-center points of the first neuron. Increasing the number of non-center points enhances the perception and extraction capabilities of image features, but also requires more computational resources and time. Therefore, it is necessary to select an appropriate number of non-center points based on specific needs and computational resources to ensure the feature extraction scale while avoiding over-computation and resource waste.

[0073] In this way, the scattering center of SAR images can be determined more accurately, and features can be extracted from them.

[0074] Understandably, this example only mentions factors such as the size of the first neuron, the number of non-center points, and the feature extraction scale. In reality, other factors can also be considered, such as the combination and interaction between different neurons, to achieve more complex and refined feature extraction tasks.

[0075] According to embodiments of this disclosure, the neuron may further include a second neuron for determining the elliptical scattering center of the SAR image. For example, the scattering center of the SAR image can also be determined by operations S521 to S522.

[0076] In operation S521, the second weight is determined based on the second distance from the non-center point of the second neuron to the center of the second neuron, and the direction of the ellipse.

[0077] In operation S522, features are extracted from the SAR image based on a second weight to determine the elliptical scattering center. The second weight is non-linearly inversely proportional to the second distance, and the elliptical direction includes 0°, 45°, 90°, and 135°.

[0078] For example, first, a second neuron is designed with a shape and size corresponding to the elliptical scattering center. This neuron is an ellipse, either horizontally or vertically, rather than a circle. Therefore, the orientation of the ellipse needs to be considered, including directions such as 0°, 45°, 90°, and 135°.

[0079] Next, the SAR image is input into the deep neural network, and a convolution operation is performed using a second neuron. The distance from the non-center point of the second neuron to its center, as well as the direction of the ellipse, determines the value of the second weight. Specifically, a non-linear inverse weight is calculated based on this distance and the direction of the ellipse, and this weight corresponds to the importance of the elliptical scattering center.

[0080] For example, the weight parameters of four neurons for sensing the scattering center of an elliptical SAR are calculated. These four neurons correspond to four directional values ​​θ, namely θ values ​​of 0°, 45°, 90°, and 135°. The corresponding weight parameter w2 is calculated based on the directional value. i The calculation formula is:

[0081]

[0082] Where i is the index number and u is the parameter. i and v i As with the above rules, θ is a directional value, for example, 0°, 45°, 90°, and 135° respectively. This way, we can obtain the weight parameters of the neurons at the elliptical scattering centers corresponding to the four different directions.

[0083] After calculating the second weight, this weight is used to extract features from the SAR image to determine the location and shape of the elliptical scattering center. This feature extraction process can also include some complex calculations and operations, such as convolution, activation functions, pooling, etc.

[0084] In this way, features of elliptical scattering centers can be extracted from SAR images and weighted in a non-linear manner. This example uses the concept of a second neuron and a second weight, but in practice, multiple neurons and weights can be designed as needed to handle different types of scattering centers and image features.

[0085] It should be noted that this example is merely a simplified demonstration, and real-world scenarios may be far more complex. For instance, an elliptical scattering center may not be a perfectly horizontal or vertical ellipse, but rather exhibit some degree of deformation or distortion. In such cases, more complex neurons and corresponding weight calculation methods are required to accurately extract these features. Furthermore, for images containing both circular and elliptical scattering centers, even more complex neural network structures and weight calculation methods are needed to handle both types of scattering centers simultaneously.

[0086] According to embodiments of this disclosure, the scattering center of a SAR image can also be determined, for example, by operating S621.

[0087] In operation S621, the second weight is adjusted according to the elliptic flattening.

[0088] For example, the adjustment of the elliptic flattening to the second weight can also be considered. The flattening of an ellipse is an important parameter describing the shape of an ellipse, reflecting the proportional relationship between the major and minor axes. When the flattening of an ellipse is close to 1, it means that the ellipse is closer to a circle; while when the flattening of an ellipse is far from 1, it means that the ellipse is more slender or flatter.

[0089] To better capture the characteristics of the elliptical scattering center, the second weight can be adjusted based on the flattening of the ellipse. Specifically, the flattening of the ellipse corresponding to the second neuron can be calculated, and the second weight can be increased or decreased accordingly. If the flattening of the ellipse is close to 1, it indicates that the ellipse is relatively rounded, and the second weight can be appropriately increased to emphasize the features of the round part; if the flattening of the ellipse is far from 1, it indicates that the ellipse is relatively slender or flat, and the second weight can be appropriately decreased to weaken the features of the round part and enhance the performance of the elliptical part.

[0090] In this way, the location and shape of the elliptical scattering center can be determined more accurately, and its features can be extracted. In this example, the second weight is adjusted according to the flattening of the ellipse, but in practice, more complex neural network structures and weight adjustment methods can be designed as needed to handle more complex and diverse scattering center types and image features.

[0091] For example, to facilitate calculations and reduce the amount of computation, the flattening ratio of the ellipse can be set to 1:2.

[0092] According to embodiments of this disclosure, determining the scattering center of a SAR image by operating steps S721-S722 may further include:

[0093] In operation S721, the pixel values ​​of the SAR image are thresholded according to the first weight and / or the second weight to obtain the activation input.

[0094] When operating the S722, a regular linear rectified function is used to activate the input and output to obtain the scattering center feature map of the SAR image.

[0095] For example, a first weight and a second weight are used to extract features from the SAR image to determine the scattering center. Then, a thresholding process and a regular linear rectified function are further introduced to obtain a clearer and more accurate scattering center feature map.

[0096] First, the pixel values ​​of the SAR image are thresholded based on a first weight and / or a second weight. This process can be understood as filtering the image, suppressing the parts with lower pixel values ​​(parts with lower weights) and retaining the parts with higher pixel values ​​(parts with higher weights) as the selection result. This filtering process can be implemented using simple threshold judgment or complex image processing techniques.

[0097] Next, a regular linear rectified function (LRC) is used to activate the filtered results, yielding the scattering center feature map of the SAR image. The LRC is a commonly used activation function that maps negative inputs to 0 and positive inputs to themselves, thus achieving non-linear activation. This process can be understood as performing a non-linear transformation on the filtered results to enhance the characteristic representation of scattering centers in the image.

[0098] In this way, a clearer and more accurate scattering center feature map can be obtained, thus better extracting scattering center information from SAR images. This example uses thresholding and a regular linear rectified function to enhance the representation of scattering center features, but in practice, more complex neural network structures and activation functions can be designed as needed to achieve more refined and complex feature extraction tasks.

[0099] For example, the activation output of an artificial neuron is calculated based on a regular linear rectified function. Using the regular linear rectified function as the activation function and setting 0.5 as the activation threshold, the activation output formula for the neuron sensing the SAR scattering center is:

[0100]

[0101] Where i is the index number, which takes the values ​​1, 2, ..., 9, and w... i The neuron weight parameters (including w1) for the circles and four types of ellipses calculated above are... i and w2 i ), I i This represents the pixel value of the SAR image input. `max` indicates that the summation calculated within the parentheses is compared to 0, and the maximum value is taken as the neuron's output. This indicates that the summation calculation is performed for index numbers i from 1 to 9.

[0102] Based on the above-described SAR image feature extraction method, this disclosure also provides a SAR image feature extraction device. The following will be combined with... Figure 3 The SAR image feature extraction device is described in detail.

[0103] Figure 3 A schematic block diagram of a SAR image feature extraction apparatus according to an embodiment of the present disclosure is shown.

[0104] like Figure 3 As shown, the SAR image feature extraction device 300 of this embodiment includes, for example, an acquisition module 310 and an extraction module 320.

[0105] The acquisition module 310 is used to acquire SAR images. In one embodiment, the acquisition module 310 can be used to perform the operation S210 described above, which will not be repeated here.

[0106] The extraction module 320 is used to input the SAR image into a deep neural network for feature extraction to determine the scattering center of the SAR image. The deep neural network includes multiple neurons, and the weight parameters of each neuron are determined by the distance from a non-center point of the neuron to the center of the neuron, and the weight parameters of the neurons are fixed. In one embodiment, the extraction module 320 can be used to perform the operation S220 described above, which will not be repeated here.

[0107] According to embodiments of this disclosure, any plurality of modules in the acquisition module 310 and extraction module 320 may be combined into one module, or any one of these modules may be split into multiple modules. Alternatively, at least a portion of the functionality of one or more of these modules may be combined with at least a portion of the functionality of other modules and implemented in one module. According to embodiments of this disclosure, at least one of the acquisition module 310 and extraction module 320 may be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, at least one of the acquisition module 310 and extraction module 320 may be at least partially implemented as a computer program module, which, when run, can perform corresponding functions.

[0108] Figure 4 A block diagram schematically illustrates an electronic device suitable for implementing a SAR image feature extraction method according to an embodiment of the present disclosure.

[0109] like Figure 4 As shown, an electronic device 400 according to an embodiment of the present disclosure includes a processor 401, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 402 or a program loaded from a storage portion 408 into a random access memory (RAM) 403. The processor 401 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 401 may also include onboard memory for caching purposes. The processor 401 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.

[0110] RAM 403 stores various programs and data required for the operation of electronic device 400. Processor 401, ROM 402, and RAM 403 are interconnected via bus 404. Processor 401 performs various operations of the method flow according to embodiments of the present disclosure by executing programs in ROM 402 and / or RAM 403. It should be noted that the programs may also be stored in one or more memories other than ROM 402 and RAM 403. Processor 401 may also perform various operations of the method flow according to embodiments of the present disclosure by executing programs stored in said one or more memories.

[0111] According to embodiments of this disclosure, the electronic device 400 may further include an input / output (I / O) interface 405, which is also connected to a bus 404. The electronic device 400 may also include one or more of the following components connected to the I / O interface 405: an input section 406 including a keyboard, mouse, etc.; an output section 407 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 408 including a hard disk, etc.; and a communication section 409 including a network interface card such as a LAN card, modem, etc. The communication section 409 performs communication processing via a network such as the Internet. A drive 410 is also connected to the I / O interface 405 as needed. A removable medium 411, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 410 as needed so that computer programs read from it can be installed into the storage section 408 as needed.

[0112] This disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the SAR image feature extraction method according to the embodiments of this disclosure.

[0113] According to embodiments of this disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium, such as, but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this disclosure, the computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this disclosure, the computer-readable storage medium may include ROM 402 and / or RAM 403 and / or one or more memories other than ROM 402 and RAM 403 described above.

[0114] Embodiments of this disclosure also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code enables the computer system to implement the SAR image feature extraction method provided in the embodiments of this disclosure.

[0115] When the computer program is executed by the processor 401, it performs the functions defined in the system / apparatus of this disclosure embodiments. According to embodiments of this disclosure, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0116] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via communication section 409, and / or installed from removable medium 411. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0117] In such an embodiment, the computer program can be downloaded and installed from a network via communication section 409, and / or installed from removable medium 411. When the computer program is executed by processor 401, it performs the functions defined in the system of this disclosure embodiment. According to embodiments of this disclosure, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0118] According to embodiments of this disclosure, program code for executing the computer programs provided in embodiments of this disclosure can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can execute entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0119] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0120] Those skilled in the art will understand that the features described in the various embodiments and / or claims of this disclosure can be combined or combined in various ways, even if such combinations or combinations are not explicitly described in this disclosure. In particular, the features described in the various embodiments and / or claims of this disclosure can be combined or combined in various ways without departing from the spirit and teachings of this disclosure. All such combinations and / or combinations fall within the scope of this disclosure.

[0121] The embodiments of this disclosure have been described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of this disclosure. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. The scope of this disclosure is defined by the appended claims and their equivalents. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of this disclosure, and all such substitutions and modifications should fall within the scope of this disclosure.

Claims

1. A SAR image feature extraction method, characterized in that, include: Acquire SAR images; The SAR image is input into a deep neural network for feature extraction to determine the scattering center of the SAR image; The deep neural network includes multiple neurons, and the weight parameters of each neuron are determined by the distance from the non-center point of the neuron to the center of the neuron, and the weight parameters of each neuron are fixed. The neuron includes a first neuron for determining the circular scattering center of the SAR image. The step of inputting the SAR image into a deep neural network for feature extraction to determine the scattering center of the SAR image includes: The first weight is determined based on the first distance from the non-center point of the first neuron to the center of the first neuron; Based on the first weight, feature extraction is performed on the SAR image to determine the circular scattering center; Wherein, the first weight is inversely proportional to the first distance in a non-linear manner; The formula for calculating the first weight is: Where i is the index number, u i and v i These are parameters that appear in pairs.

2. The method according to claim 1, characterized in that, The step of inputting the SAR image into a deep neural network for feature extraction to determine the scattering center of the SAR image further includes: Determine the size of the first neuron; Based on the size of the first neuron, determine the number of non-central points of the first neuron; and The feature extraction scale of the SAR image is determined based on the number of non-center points of the first neuron.

3. The method according to claim 1, characterized in that, The neuron further includes a second neuron for determining the elliptical scattering center of the SAR image. The step of inputting the SAR image into a deep neural network for feature extraction to determine the scattering center of the SAR image further includes: The second weight is determined based on the second distance from the non-center point of the second neuron to the center of the second neuron, and the direction of the ellipse; Based on the second weight, feature extraction is performed on the SAR image to determine the elliptical scattering center; The second weight is inversely proportional to the second distance in a non-linear manner, and the elliptical direction includes 0°, 45°, 90° and 135°.

4. The method according to claim 3, characterized in that, The step of inputting the SAR image into a deep neural network for feature extraction to determine the scattering center of the SAR image further includes: The second weight is adjusted based on the elliptic flattening.

5. The method according to claim 3, characterized in that, The step of inputting the SAR image into a deep neural network for feature extraction to determine the scattering center of the SAR image further includes: Based on the first weight and / or the second weight, the pixel values ​​of the SAR image are thresholded to obtain the activation input; The scattering center feature map of the SAR image is obtained by activating the input using a regular linear rectified function.

6. A SAR image feature extraction device, characterized in that, include: The acquisition module is used to acquire SAR images; as well as An extraction module is used to input the SAR image into a deep neural network for feature extraction in order to determine the scattering center of the SAR image; The deep neural network includes multiple neurons, and the weight parameters of each neuron are determined by the distance from the non-center point of the neuron to the center of the neuron, and the weight parameters of each neuron are fixed. The neuron includes a first neuron for determining the circular scattering center of the SAR image. The step of inputting the SAR image into a deep neural network for feature extraction to determine the scattering center of the SAR image includes: The first weight is determined based on the first distance from the non-center point of the first neuron to the center of the first neuron; Based on the first weight, feature extraction is performed on the SAR image to determine the circular scattering center; Wherein, the first weight is inversely proportional to the first distance in a non-linear manner; The formula for calculating the first weight is: Where i is the index number, u i and v i These are parameters that appear in pairs.

7. An electronic device, characterized in that, include: One or more processors; Storage device 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 perform the method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, It stores executable instructions that, when executed by a processor, cause the processor to perform the method according to any one of claims 1 to 5.

9. A computer program product, characterized in that, Includes a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 5.

Images