Few-sample ISAR target identification method based on self-supporting scattering feature coding
Through self-supported scattering feature encoding and convolutional neural networks, high-dimensional scattering features are generated in ISAR images, and iterative training is carried out in combination with self-support and interaction loss, the low signal-to-noise ratio and edge blur problems of target component recognition in ISAR images are solved, improving the Identification effect and segmentation performance.
Patent Information
- Application Number
- CN202410924456.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-11
- Publication Date
- 2024-11-12
AI Technical Summary
The prior art is difficult to accurately identify target components in ISAR images with low signal-to-noise ratio and edge blur, and the semantic segmentation algorithm has limited in-class diversity and inter-class differences, making it difficult to improve the recognition effect.
Using a method based on self-supported scattering feature encoding, high-dimensional scattering features are generated through convolutional neural networks, combined with self-support and interaction loss for iterative training, and generate model weights of component segmentation algorithms, which are used to identify ISAR target components under the condition of few samples. .
It improves the identification accuracy and segmentation performance of ISAR target components, provides clearer boundaries under low signal-to-noise ratio and edge blur, and enhances the analysis ability of target attitude and operating state.
Smart Images

Figure CN118941886A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of target recognition, and in particular to a few-sample ISAR target recognition method based on self-supporting scattering feature coding. Background Art
[0002] The target in the Inverse Synthetic Aperture Radar (ISAR) image is composed of discontinuous scattering points. The intensity information of the scattering points is affected by many factors such as the target shape, material and radar bandwidth. In addition, the change of the relative observation angle of the radar to the target may also cause a large difference in the target ISAR image.
[0003] The continuous development of imaging technology has promoted the improvement of ISAR imaging resolution. The precise interpretation technology of radar targets has also attracted more and more attention from researchers. Existing target component recognition tasks mostly use target detection schemes, that is, using rectangular detection boxes to select different target components. The components obtained by box selection can only obtain their positioning information, and it is difficult to obtain the shape and structural characteristics of the components, which is difficult to meet the requirements of accurate ISAR target recognition.
[0004] Semantic segmentation methods based on optical images include clustering, threshold judgment and edge detection. These methods are based on traditional feature designs such as pixel grayscale and gradient, and are difficult to adapt to ISAR images with low signal-to-noise ratio. Unlike optical imaging, target ISAR images often present low signal-to-noise ratio, blurred edges, and serious sidelobe effects of some strong scattering points, which poses a severe challenge to the precise segmentation of ISAR target components. Deep neural networks have shown great advantages in nonlinear modeling and feature mining, and have developed rapidly in the fields of space environment detection and medical image segmentation.
[0005] Existing few-shot semantic segmentation methods have achieved different degrees of performance improvement by improving the matching strategy between support samples and query samples. However, the relatively scarce data conditions can only represent limited target differences, and intra-class diversity and inter-class differences are still important factors restricting the performance improvement of few-shot semantic segmentation algorithms. Summary of the invention
[0006] The technical problem to be solved by the present invention is how to enhance the recognition effect and improve the segmentation performance; in view of this, the present invention provides a few-sample ISAR target recognition method based on self-supporting scattering feature coding.
[0007] The technical solution adopted by the present invention is a few-sample ISAR target recognition method based on self-supporting scattering feature coding, comprising:
[0008] Step S1, preprocessing the target ISAR image;
[0009] Step S2, dividing the current ISAR image into a training data set and a test data set according to a preset ratio, and constructing a feature embedding network based on a convolutional neural network to generate high-dimensional query features and support features;
[0010] Step S3, under the experimental setting of few samples N way-K shot, the training data set is divided into query samples and support samples, and the samples are input into the feature embedding network to obtain the corresponding high-dimensional feature tensor, and the high-dimensional scattering features are generated by using the high-dimensional feature tensor based on the processing method of scattering texture similarity to determine the segmentation result of the target component;
[0011] Step S4, performing iterative training under the supervision of self-support and interaction losses; after the iterative training is completed, the model weight of the component segmentation algorithm is obtained;
[0012] Step S5, loading the model weights, and using the test data set to obtain a few-sample ISAR target component recognition result.
[0013] In one embodiment, step S1 includes:
[0014] Motion compensation is performed on the defocused ISAR image, and filtering and gamma transformation are used to improve the contrast of weak components in the image, making the detailed information in the ISAR image richer.
[0015] In one embodiment, step S2 includes:
[0016] The training data set and the test data set are divided into a ratio of 7:3, and the training data set is trained. At the same time, a feature embedding network based on convolutional neural network is constructed to generate high-dimensional query features and support features.
[0017] In one embodiment, step S3 includes:
[0018] A high-dimensional scattering feature tensor having both shallow texture information and high-level semantic information is obtained through a multi-layer convolutional neural network, wherein the high-dimensional scattering feature tensor includes: a query feature tensor and a support feature tensor;
[0019] Determining a query sample prototype using the high-dimensional scattering feature tensor;
[0020] Based on the sample prototype, determining the cosine similarity between the query feature tensor and the supporting feature tensor;
[0021] Based on the cosine similarity, a high-dimensional scattering feature is generated using the ISAR image of the query sample;
[0022] Performing feature weighting processing on the high-dimensional scattering features and the query sample prototype;
[0023] The target component segmentation result is obtained by using the weighted high-dimensional scattering features and the query feature tensor.
[0024] In one embodiment, step S4 includes:
[0025] The target recognition model was trained from scratch using the ISAR dataset, and the model weights were iteratively optimized using the stochastic gradient descent method.
[0026] Another aspect of the present invention further provides a few-sample ISAR target recognition device based on self-supporting scattering feature coding, comprising:
[0027] A preprocessing unit configured to preprocess the target ISAR image;
[0028] A partitioning unit is configured to partition the current ISAR image into a training data set and a test data set according to a preset ratio, and to construct a feature embedding network based on a convolutional neural network for generating high-dimensional query features and supporting features;
[0029] A computing unit is configured to divide the training data set into query samples and support samples under the experimental setting of few samples N way-K shot, and input the training data set into a feature embedding network to obtain a corresponding high-dimensional feature tensor, and generate a high-dimensional scattering feature based on a scattering texture similarity processing method by using the high-dimensional feature tensor to determine the target component segmentation result;
[0030] The iterative unit is configured to perform iterative training under the supervision of self-support and interaction losses; after the iterative training is completed, the model weight of the component segmentation algorithm is obtained;
[0031] The output unit is configured to load the model weights and obtain a few-sample ISAR target component recognition result using the test data set.
[0032] Another aspect of the present invention provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the few-sample ISAR target recognition method based on self-supporting scattering feature coding as described in any of the above items.
[0033] Another aspect of the present invention further provides a computer storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the few-sample ISAR target recognition method based on self-supporting scattering feature coding as described in any of the above items are implemented.
[0034] Compared with the prior art, the present invention has at least the following advantages:
[0035] The present invention belongs to a method based on metric learning. It uses a feature embedding network to obtain target scattering characteristics that have both shallow texture and high-level semantic information, combines intra-class and inter-class similarity metrics to obtain better segmentation boundaries, and considers the component recognition problem from the perspective of few-sample semantic segmentation. Therefore, while the present invention accurately identifies components, the segmentation results are helpful for analyzing the target posture and operating status. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] Figure 1 A schematic flow chart of a few-sample ISAR target recognition method based on self-supporting scattering feature coding according to an embodiment of the present invention;
[0037] Figure 2 A logic schematic diagram of a few-sample ISAR target recognition method based on self-supporting scattering feature coding according to an embodiment of the present invention;
[0038] Figure 3 A schematic diagram of an ISAR target imaging result according to an embodiment of the present invention;
[0039] Figure 4 A schematic diagram of a scattering feature encoding network according to an embodiment of the present invention;
[0040] Figure 5 A schematic diagram of a component segmentation process based on a scattering matching strategy according to an embodiment of the present invention;
[0041] Figure 6 is a schematic diagram of component segmentation results according to an embodiment of the present invention;
[0042] Figure 7 It is a schematic diagram of the structure of a few-sample ISAR target recognition device based on self-supporting scattering feature coding according to an embodiment of the present invention;
[0043] Figure 8 The figure is a schematic diagram of the composition of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION
[0044] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined purpose, the present invention is described in detail below in conjunction with the accompanying drawings and preferred embodiments.
[0045] Unless otherwise defined, all terms (including technical terms and scientific terms) used in this article have the same meaning as commonly understood by ordinary technicians in the field to which this application belongs. It should also be understood that terms (such as terms defined in commonly used dictionaries) should be interpreted as having the same meaning as their meaning in the context of the relevant technology, and will not be interpreted in an idealized or overly formal sense unless explicitly defined in this article.
[0046] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.
[0047] For ease of understanding, necessary explanations are given below on the relevant technologies involved in the present invention.
[0048] The embodiment of the present invention is a few-sample ISAR target recognition method based on self-supporting scattering feature coding, such as Figure 1 As shown, including:
[0049] Step S1, preprocessing the target ISAR image;
[0050] Step S2, dividing the current ISAR image into a training data set and a test data set according to a preset ratio, and constructing a feature embedding network based on a convolutional neural network to generate high-dimensional query features and support features;
[0051] Step S3, under the experimental setting of few samples N way-K shot, the training data set is divided into query samples and support samples, and the samples are input into the feature embedding network to obtain the corresponding high-dimensional feature tensor, and the high-dimensional scattering features are generated by using the high-dimensional feature tensor based on the processing method of scattering texture similarity to determine the segmentation result of the target component;
[0052] Step S4, performing iterative training under the supervision of self-support and interaction losses; after the iterative training is completed, the model weight of the component segmentation algorithm is obtained;
[0053] Step S5, loading the model weights, and using the test data set to obtain a few-sample ISAR target component recognition result.
[0054] It should be noted that the target recognition method proposed in this article may specifically refer to a method for recognizing ISAR target components.
[0055] Can be combined Figure 2 , the method provided in this embodiment will be described in detail step by step below.
[0056] Step S1, preprocessing the target ISAR image.
[0057] refer to Figure 3 In this embodiment, the target ISAR imaging results often show low signal-to-noise ratio and blurred edges. The sidelobe effect of some strong scattering points affects the scattering information representation of the components in the ISAR image. The above factors may have a significant impact on the accuracy of component recognition. Therefore, it is necessary to first perform motion compensation on the defocused ISAR image, and at the same time use filtering and gamma transformation to improve the contrast of weak components in the image, so that the detailed information in the ISAR image is richer.
[0058] Step S2, dividing the current ISAR image into a training data set and a test data set according to a preset ratio, and constructing a feature embedding network based on a convolutional neural network to generate high-dimensional query features and support features.
[0059] Specifically, the few-sample ISAR target component recognition architecture proposed in the present invention can divide the data into a training set and a test set according to a preset ratio of 7:3. As a feasible data reference, the training set contains 360 pairs of samples and the test set contains 252 sample pairs, all of which are composed of ISAR images and complex-valued echoes of four types of targets.
[0060] Step S3, under the experimental setting of few samples N way-K shot, the training data set is divided into query samples and support samples, and the samples are input into the feature embedding network to obtain the corresponding high-dimensional feature tensor. The high-dimensional feature tensor is used to generate high-dimensional scattering features based on the scattering texture similarity processing method to determine the target component segmentation result.
[0061] The few-shot episodic learning paradigm is adopted, that is, the training data is divided into support samples and query samples. Support samples are represented by superscript s, and query samples are represented by superscript q. The training process consists of multiple iterations, each of which contains the support set S = (I s ,M s ) and the query set Q = (I q ,M q ). represents the ISAR image, is the corresponding mask label. In each round of training, the few-shot semantic segmentation learns (I q ,I s ,M s ) to M q 's mapping.
[0062] The feature embedding network is used to process the target ISAR image. Through multi-layer convolutional neural network operations, a high-dimensional scattering feature tensor F with both shallow texture information and high-level semantic information is obtained. s ,F q As a pixel-level classification task, semantic segmentation loses more detailed texture information when downsampling more. Therefore, when designing the encoding network, avoid using a network structure with more downsampling or deeper layers. Figure 4 shown.
[0063] The present invention generates a scattering code query tensor by continuously collecting and updating high-confidence query features, and uses the above encoding to obtain component features with higher similarity. Figure 5 .
[0064] Generate query feature tensor F q With the support feature tensor F s After that, calculate the query sample prototype: P s =MAP(M s ,F s ). Among them, M s is the true mask label. s The support feature tensor generated for the support samples. MAP stands for masked average pooling.
[0065] Secondly, calculate P s ,F s The cosine similarity sim(P s ,F q ), generating an initial scatter prediction mask M with high similarity to the query sample initial .
[0066] M initial =softmax(sim(P s ,F q ))
[0067] As mentioned above, the scattering pixel textures between different parts of the same target are closer than the scattering pixel textures between the same parts of different targets. Therefore, the ISAR image of the query sample is used to generate high-dimensional features. When the confidence level is set to τ, M initial The prediction mask retained when τ > τ. The confidence is used to select the prediction features with more intra-class similarity and closer to the real mask. In the experiment, τ = 0.6 is set. The initial scatter prediction mask of the generated query prototype contains more feature tensors that characterize the characteristics of the target component.
[0068] Next, feature weighting is performed on the scatter matching prototypes and the supporting prototypes. Where γ represents the prototype weight, which can be updated adaptively and iteratively as the gradient changes during training. The initial value is set to γ = 0.5.
[0069] The final component segmentation result is obtained by using the weighted high-dimensional scattering feature encoding and query sample features:
[0070]
[0071] In order to make the initial scatter prediction mask generated by the query sample closer to the real mask, cover more areas that are conducive to improving segmentation accuracy, and make the features generated by the query sample contain clearer division boundaries, the following loss function is used to optimize the target component segmentation framework:
[0072]
[0073] Among them, L1 represents the self-support loss, L ce Represents the calculation of pixel-by-pixel cross entropy loss.
[0074] In this article, the improved algorithm based on self-support loss and interaction loss can be referred to as follows:
[0075] The interactive loss uses pixel points and their neighborhood information to jointly describe the current spatial pixel points. The encoded feature vector is projected into a high-dimensional feature distribution. By measuring the distribution between the predicted value and the true value, the predicted vector is made to approach the direction of high-order consistency with the true value. The present invention uses the interactive loss to achieve continuous performance optimization of the ISAR target component segmentation algorithm.
[0076] From the above analysis, we can see that the scattering textures between components in the target ISAR image have a strong correlation. In addition, the ISAR target imaging results contain a large number of discontinuous scattering points. Different ISAR targets may have components with similar geometric structures. The correlation between pixels of different classes also contains important structural information.
[0077] Therefore, the present invention designs an interaction loss. By modeling the correlation between pixels, the scattered features are mapped into a high-dimensional tensor to maximize the feature similarity, thereby improving the component segmentation effect.
[0078] Assume that in the query sample, the feature region R' with a scale of A×A is used to segment the target component pixel by pixel using its scattering feature encoding vector. The prediction probability output by the algorithm is B=[b1,b2,...,b A×A ] T The truth mask of the corresponding region of R' is M = [m1,m2,...,m A×A ] T .
[0079] It is known from the image registration theory in the field of medical segmentation that by maximizing the correlation of the high-dimensional distribution between the predicted value and the true value mask, the predicted value can be made closer to the true value. The present invention measures the intrinsic correlation by calculating the covariance and entropy value of the above two high-dimensional distributions. According to Bayesian theory, the following formula can be used as an approximation of the high-correlation covariance matrix.
[0080]
[0081] Define the interaction loss L mum (b,m) is as follows:
[0082]
[0083] Among them, λ∈[0,1] represents the weight factor. E is the amount of data in the current training batch. C represents the number of feature channels.
[0084] The pixel-by-pixel cross entropy loss is used to evaluate the similarity of the scattering texture intensity between tensors, while the interaction loss and self-support loss focus on the structural similarity between the two types of samples. The overall loss function of the component recognition algorithm described in the present invention is composed of the self-support loss L1 and the interaction loss L2. mum Two parts, L = L1 + L mum (b,m).
[0085] Step S4, performing iterative training under the supervision of self-support and interaction losses; after the iterative training is completed, the model weights of the component segmentation algorithm are obtained.
[0086] Specifically, mean Intersection-over-Union (mIoU), the most common indicator in the field of few-sample semantic segmentation, is used as the evaluation index. At the same time, the component recognition algorithm proposed in the present invention is evaluated under the 1-shot / 5-shot settings. Taking into account the differences between optical data and ISAR imaging principles, the present invention does not load a pre-trained model, and uses the ISAR data set to train the component recognition model from scratch. The model is iteratively optimized using the stochastic gradient descent method. The initial learning rate is set to 0.001. The total number of training rounds is 3000 times, and the learning rate decays to 0.1 times the previous one after 1000 rounds of iteration. During the training process, the initial size of the ISAR image and its mask is retained for input feature embedding network training, that is, the height and width of the input data are both 256 pixels, which is consistent with the number of sampling points in the construction process of the ISAR data set. After the training is completed, the weight model of the segmentation algorithm is saved. It is used to test the segmentation accuracy of the algorithm.
[0087] Step S5, loading the model weights, and using the test data set to obtain a few-sample ISAR target component recognition result.
[0088] Load the weight model obtained in the algorithm training phase, input the constructed test data set into the few-sample component segmentation model, and output the component segmentation results as follows Figure 6 .
[0089] A large number of comparative experiments are used to verify the advancedness of the few-sample ISAR target component recognition algorithm described in the present invention. Further analysis shows that the recognition scheme described in the present invention does not expand the network structure, but improves the component segmentation performance based on the non-parametric method, so the best ISAR target component recognition performance is achieved with the least parameters.
[0090] Compared with the prior art, the method provided by the embodiment of the present invention has at least the following advantages:
[0091] 1) The present invention solves the problem of blurred boundary of component segmentation caused by factors such as low signal-to-noise ratio and sidelobe effect by adaptively aggregating the characterization prototypes with high confidence of each component of the ISAR target. The experimental data contains many low signal-to-noise ratio ISAR images that are greatly affected by noise, blurred boundaries and discontinuous distribution of scattering points, which pose a higher challenge to the accurate segmentation of target components. Due to the use of a mask generation scheme based on scattering texture matching, the present invention is less affected by scattering noise and has good recognition robustness for low-quality ISAR data.
[0092] 2) In order to solve the problem of blurred edge and low feature utilization of target ISAR image under few sample conditions, the present invention uses the inherent scattering texture feature similarity of the component presented in the ISAR target imaging results to obtain the initial scattering prediction code closer to the true value of the query sample mask, and at the same time generates a high-resolution scattering coding feature tensor covering global semantics and local texture. By establishing a high correlation between the query sample and the support sample, the segmentation problem caused by feature differences is alleviated. The self-support and cross-correlation losses are designed to optimize the scattering feature measurement while obtaining a better ISAR target component recognition boundary.
[0093] The second embodiment of the present invention corresponds to the first embodiment. This embodiment introduces a small sample ISAR target recognition device based on self-supporting scattering feature coding, referring to Figure 7 , including the following components:
[0094] A preprocessing unit configured to preprocess the target ISAR image;
[0095] A partitioning unit is configured to partition the current ISAR image into a training data set and a test data set according to a preset ratio, and to construct a feature embedding network based on a convolutional neural network for generating high-dimensional query features and supporting features;
[0096] A computing unit is configured to divide the training data set into query samples and support samples under the experimental setting of few samples N way-K shot, and input the training data set into a feature embedding network to obtain a corresponding high-dimensional feature tensor, and generate a high-dimensional scattering feature based on a scattering texture similarity processing method by using the high-dimensional feature tensor to determine the target component segmentation result;
[0097] The iterative unit is configured to perform iterative training under the supervision of self-support and interaction losses; after the iterative training is completed, the model weight of the component segmentation algorithm is obtained;
[0098] The output unit is configured to load the model weights and obtain a few-sample ISAR target component recognition result using the test data set.
[0099] A third embodiment of the present invention is an electronic device, referring to Figure 8, can be understood as a physical device, including a processor and a memory storing instructions executable by the processor. When the instructions are executed by the processor, the following operations are performed:
[0100] Step S1, preprocessing the target ISAR image;
[0101] Step S2, dividing the current ISAR image into a training data set and a test data set according to a preset ratio, and constructing a feature embedding network based on a convolutional neural network to generate high-dimensional query features and support features;
[0102] Step S3, under the experimental setting of few samples N way-K shot, the training data set is divided into query samples and support samples, and the samples are input into the feature embedding network to obtain the corresponding high-dimensional feature tensor, and the high-dimensional scattering features are generated by using the high-dimensional feature tensor based on the processing method of scattering texture similarity to determine the segmentation result of the target component;
[0103] Step S4, performing iterative training under the supervision of self-support and interaction losses; after the iterative training is completed, the model weight of the component segmentation algorithm is obtained;
[0104] Step S5, loading the model weights, and using the test data set to obtain a few-sample ISAR target component recognition result.
[0105] The fourth embodiment of the present invention, the process of the few-sample ISAR target recognition method based on self-supporting scattering feature coding in this embodiment is the same as that of the first, second or third embodiment, except that, in engineering implementation, this embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the method of the present invention can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a device to execute the method described in the embodiment of the present invention.
[0106] Through the description of the specific implementation methods, a deeper and more specific understanding of the technical means and effects adopted by the present invention to achieve the predetermined purpose should be obtained. However, the accompanying drawings are only for reference and illustration purposes and are not intended to limit the present invention.
Claims
1. A few-sample ISAR target recognition method based on self-supporting scattering feature coding, characterized in that: include: Step S1, preprocessing the target ISAR image; Step S2, dividing the current ISAR image into a training data set and a test data set according to a preset ratio, and constructing a feature embedding network based on a convolutional neural network to generate high-dimensional query features and support features; Step S3, under the experimental setting of few samples N way-K shot, the training data set is divided into query samples and support samples, and the samples are input into the feature embedding network to obtain the corresponding high-dimensional feature tensor, and the high-dimensional scattering features are generated by using the high-dimensional feature tensor based on the processing method of scattering texture similarity to determine the segmentation result of the target component; Step S4, performing iterative training under the supervision of self-support and interaction losses; After the iterative training is completed, the model weights of the component segmentation algorithm are obtained; Step S5, loading the model weights, and using the test data set to obtain a few-sample ISAR target component recognition result.
2. The few-sample ISAR target recognition method based on self-supporting scattering feature coding according to claim 1 is characterized in that: The step S1 comprises: Motion compensation is performed on the defocused ISAR image, and filtering and gamma transformation are used to improve the contrast of weak components in the image, making the detailed information in the ISAR image richer.
3. The method for identifying target components using few-sample ISAR based on self-supporting scattering feature coding according to claim 2, characterized in that: The step S2 comprises: The training data set and the test data set are divided into a ratio of 7:3, and the training data set is trained. At the same time, a feature embedding network based on convolutional neural network is constructed to generate high-dimensional query features and support features.
4. The few-sample ISAR target recognition method based on self-supporting scattering feature coding according to claim 3 is characterized in that: The step S3 comprises: A high-dimensional scattering feature tensor having both shallow texture information and high-level semantic information is obtained through a multi-layer convolutional neural network, wherein the high-dimensional scattering feature tensor includes: a query feature tensor and a support feature tensor; Determining a query sample prototype using the high-dimensional scattering feature tensor; Based on the sample prototype, determining the cosine similarity between the query feature tensor and the supporting feature tensor; Based on the cosine similarity, a high-dimensional scattering feature is generated using the ISAR image of the query sample; Performing feature weighting processing on the high-dimensional scattering features and the query sample prototype; The target component segmentation result is obtained by using the weighted high-dimensional scattering features and the query feature tensor.
5. The few-sample ISAR target recognition method based on self-supporting scattering feature coding according to claim 4 is characterized in that: The step S4 comprises: The target recognition model was trained from scratch using the ISAR dataset, and the model weights were iteratively optimized using the stochastic gradient descent method.
6. A few-sample ISAR target recognition device based on self-supporting scattering feature coding, characterized in that: include: A preprocessing unit configured to preprocess the target ISAR image; A partitioning unit is configured to partition the current ISAR image into a training data set and a test data set according to a preset ratio, and to construct a feature embedding network based on a convolutional neural network for generating high-dimensional query features and supporting features; A computing unit is configured to divide the training data set into query samples and support samples under the experimental setting of few samples N way-K shot, and input the training data set into a feature embedding network to obtain a corresponding high-dimensional feature tensor, and generate a high-dimensional scattering feature based on a scattering texture similarity processing method by using the high-dimensional feature tensor to determine the target component segmentation result; The iterative unit is configured to perform iterative training under the supervision of self-support and interaction losses; After the iterative training is completed, the model weights of the component segmentation algorithm are obtained; The output unit is configured to load the model weights and obtain a few-sample ISAR target component recognition result using the test data set.
7. An electronic device, characterized in that: The electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, the steps of the few-sample ISAR target recognition method based on self-supporting scattering feature coding as described in any one of claims 1 to 5 are implemented.
8. A computer storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps of the few-sample ISAR target recognition method based on self-supporting scattering feature coding as claimed in any one of claims 1 to 5 are implemented.
