A lightweight breast microwave image reconstruction method, device, medium and product

Abstract

The application discloses a lightweight breast microwave image reconstruction method and device, medium and product, and relates to the field of microwave imaging. The method comprises the following steps: collecting breast microwave scattering field data to obtain an original microwave breast image; constructing a lightweight multi-scale convergent convolution based on different groups of convolution modules; constructing a microwave image reconstruction network based on the lightweight multi-scale convergent convolution and in combination with an encoder-decoder architecture; and obtaining a reconstructed microwave breast image based on the original microwave breast image by using the microwave image reconstruction network. The application can reduce the model parameter quantity, reduce the calculation complexity, and improve the inference speed, thereby meeting the application requirements of clinical portability and light weight.

Description

Technical Field

[0001] This application relates to the field of microwave imaging, and in particular to a lightweight method, device, medium, and product for microwave image reconstruction of breast tissue. Background Technology

[0002] Microwave imaging technology, with its advantages of being non-contact, radiation-free, and miniaturizable, has significant application value in fields such as medical imaging (e.g., breast imaging) and security detection. Its core lies in reconstructing the dielectric property distribution of the target by solving the electromagnetic inverse scattering problem. In microwave breast imaging, the high dielectric constant contrast between adipose tissue and fibroglandular tissue poses a significant challenge to microwave inverse scattering imaging of breast tissue. Traditional electromagnetic inverse scattering solutions (such as Contrast Source Inversion (CSI)) suffer from low imaging accuracy and a tendency to get trapped in local optima. In recent years, with the rapid development of deep learning, researchers have achieved better microwave image results than traditional inverse scattering imaging methods by inputting electromagnetic data into neural networks to reconstruct microwave images. While deep learning-based microwave image reconstruction methods can significantly improve image quality, existing general-purpose neural network architectures generally suffer from large model parameters, high computational complexity, and slow inference speed in microwave image reconstruction, making it difficult to meet the needs of portable and lightweight clinical applications, seriously hindering the clinical translation and implementation of microwave technology. Summary of the Invention

[0003] To address the aforementioned problems in the existing technology, this application provides a lightweight method, device, medium, and product for breast microwave image reconstruction.

[0004] To achieve the above objectives, this application provides the following solution: In a first aspect, this application provides a lightweight method for breast microwave image reconstruction, comprising: Microwave scattering field data of the breast were collected to obtain raw microwave breast images; Lightweight multiscale-converge convolution (MCC) is constructed based on convolution modules with different numbers of groups. A microwave image reconstruction network is constructed based on the aforementioned lightweight multi-scale convergent convolution and combined with an encoder-decoder architecture. The microwave image reconstruction network is used to obtain a reconstructed microwave breast image based on the original microwave breast image.

[0005] Secondly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the lightweight breast microwave image reconstruction method provided above.

[0006] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the lightweight breast microwave image reconstruction method described above.

[0007] Fourthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the lightweight breast microwave image reconstruction method described above.

[0008] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a lightweight method, device, medium, and product for breast microwave image reconstruction. By constructing lightweight multi-scale convergent convolutions based on convolutional modules with varying numbers of groups, it enables fine-grained extraction of scattering features from different breast tissues (such as skin, fat, and fibroglandular tissues), and achieves global correlation and efficient fusion of cross-channel features. By constructing a microwave image reconstruction network based on lightweight multi-scale convergent convolutions and an encoder-decoder architecture, the network's learning ability for nonlinear mappings is enhanced, while maintaining sensitivity to multi-scale scattering features at each stage of feature compression and expansion. This reduces the number of parameters, computational complexity, and inference speed of the microwave image reconstruction network. Using the constructed microwave image reconstruction network, reconstructed microwave breast images are obtained from the original microwave breast images, improving the reconstruction effect and efficiency. Furthermore, the implantability of the microwave image reconstruction network (a deep learning network) meets the clinical needs for portability and lightweight applications. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 A schematic flowchart of a lightweight breast microwave image reconstruction method provided in an embodiment of this application; Figure 2 A schematic diagram of a lightweight multi-scale convergent convolution structure provided in an embodiment of this application; Figure 3 A schematic diagram of the structure of an initial microwave image reconstruction network provided in an embodiment of this application; Figure 4 An example image showing the reconstruction result of the original microwave breast image provided in an embodiment of this application; Figure 5 This is a schematic diagram illustrating the reconstruction performance comparison results provided in an embodiment of this application; Figure 6 This is a schematic diagram showing the comparison results of network parameter quantities provided in an embodiment of this application; Figure 7 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0011] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0012] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0013] In one exemplary embodiment, this application provides a lightweight breast microwave image reconstruction method. This method is executed by a computer device, specifically a terminal or server, or both. In this embodiment, the method is described using a server as an example. Figure 1 As shown, the method includes: Step 100: Collect microwave scattering field data of the breast to obtain the raw microwave breast image; Step 101: Construct lightweight multi-scale convergent convolutions based on convolutional modules with different numbers of groups; Step 102: Construct a microwave image reconstruction network based on lightweight multi-scale convergent convolution and combined with an encoder-decoder architecture; Step 103: Using a microwave image reconstruction network, a reconstructed microwave breast image is obtained based on the original microwave breast image.

[0014] By implementing steps 100-103 above, this application has excellent image detail and structure, and can also significantly reduce the number of network parameters and computational complexity, enabling rapid breast imaging and achieving a significant lightweight effect. It has great practical value in clinical real-time imaging or portable devices.

[0015] In an exemplary embodiment of this application, in order to enhance the representation ability of local features of different scattering regions of the breast, to achieve multi-scale feature extraction of different scattering regions within the breast, and to achieve effective global correlation and efficient fusion of multi-scale refined features obtained from multiple sets of convolutional modules, and to help reconstruct a breast image composed of different scattering targets (skin, fat, and fibrous glands, etc.), the implementation process of step 101 described above may include: Step 101-1: Construct multiple sets of convolutional modules to achieve fine-grained extraction of scattering features from different breast tissues (skin, fat, fibrogland). Within these multiple convolutional modules, the input feature map... Perform a convolution channel transformation of a set size (e.g., [1, 1]), and divide the convolution channels and filters into a first preset number of groups to obtain a convolution module with the first preset number of groups. The corresponding output features are denoted as... , The first preset number of groups is (e.g., 16 groups); after each group of convolutional modules performs independent convolution operations according to the set convolutional kernel (e.g., [3, 3]), after processing by batch normalization (BN) and modified linear units (e.g., ReLU activation function), the output feature maps of each group of convolutional modules are concatenated to obtain the output feature maps of multiple groups of convolutional modules; The design of step 101-1 enhances the ability to characterize the local features of different scattering regions of the breast by increasing the number of groups, thereby enabling multi-scale feature extraction of different scattering regions within the breast. Step 101-2: Construct a few-group convolutional modules to achieve global correlation and efficient fusion of cross-channel features; in the few-group convolutional modules, the output feature maps and filters of the multi-group convolutional modules are divided into a second preset number of groups to obtain multiple feature sub-maps, denoted as... (For example, the number of groups can be determined according to the number of breast tissue types. The main breast tissues include fat, glands, and skin. For ease of calculation, the second preset number of groups can be 4 in practical applications.) After batch standardization and modified linear unit processing on multiple groups of feature sub-maps according to the set convolution kernel (e.g., [3, 3]), the output feature maps of a smaller number of convolution modules are obtained. The second preset number of groups is less than the first preset number of groups, i.e. ; The settings in steps 101-2 above can effectively correlate and efficiently fuse the multi-scale refinement features obtained from multiple convolutional modules globally, which helps to reconstruct breast images composed of different scattering targets (skin, fat, and fibrous glands, etc.). Step 101-3: Stack multiple sets of convolutional modules and a few sets of convolutional modules according to a set number of times, and perform residual processing on the input feature maps of the multiple sets of convolutional modules and the output feature maps of the few sets of convolutional modules to obtain the output of the multi-scale convergent convolution, thus completing the construction of the multi-scale convergent convolution. The residual processing refers to processing the input feature maps before the multiple convolutions. Output feature maps of a few convolutional modules Element-wise addition yields: .

[0016] In the formula, This is the output of a multi-scale convergent convolution.

[0017] Based on this, by fusing residual connections in multi-scale convolution, the gradient vanishing and feature degradation problems of deep networks (i.e. microwave image reconstruction networks) can be alleviated, ensuring the complete transmission of key features of microwave images.

[0018] Furthermore, such as Figure 2 As shown, the number of stacking operations can be set to 2, but in practical applications, the number of stacking operations can be flexibly set according to the complexity of the image reconstruction task, thereby further enhancing the network's learning ability for nonlinear mappings.

[0019] In an exemplary embodiment of this application, in order to ensure that the microwave image reconstruction network maintains its sensitivity to multi-scale scattering features at each stage of feature compression and expansion, the implementation process of step 102 provided above includes: Step 102-1: Construct a sample dataset and divide it into a training set and a test set according to a set ratio; For example, in practical applications, the implementation process of step 102-1 can be described as follows: Step (11): Obtain breast microwave scattering field sample data, use the Born iteration method to solve the breast microwave scattering field sample data by inverse scattering, and introduce the Tikhonov regularization term in the solution process to obtain the initial breast reconstruction image set; For example, a Gaussian band pulse with a bandwidth of 1 GHz was used as the transmitted signal. An array of 18 antennas was formed around the surface of the breast (45 cases in total). Echo data (i.e., breast microwave scattering field sample data) at multiple frequency points was collected in a cyclic single-transmitter, multiple-receiver mode. In this embodiment, the scattering field at 15 frequency points between 300 MHz and 1.15 GHz can be used for inverse scattering solution using the Born iterative method, in which a Tikhonov regularization term is introduced. (Regularization weight) =0.01), therefore, the objective function of the dielectric contrast image (i.e., the initial reconstructed image of the breast) obtained by inverse scattering is as follows: ; ; ; In the formula, This is an initial reconstructed image of the breast. As slack variables, In the first Scattered fields collected at each antenna location The number of antenna positions. For the scattered field, For the incident field, It is the identity matrix. For the Green's function of the probe region, For the Green's function of the target region, For Tikhonov regularization terms, The initial reconstructed image of the breast to be optimized. For regularization weights, The initial reconstructed image of the breast when the objective function is minimized. The slack variables are those that minimize the objective function. For complex fields, Indicates a condition.

[0020] By solving the above objective function, we can obtain 45 initial breast reconstruction images with an image size of 256×256.

[0021] Step (12): The images in the initial breast reconstruction image set are augmented using a hybrid data augmentation strategy to obtain a sample dataset.

[0022] To further expand the dataset size and simulate various imaging noises in real clinical scenarios, and to overcome the overfitting problem caused by small clinical sample training, in this step, the 45 initial breast reconstruction images obtained in the above step (11) can be expanded to 3000 images through a mixed data augmentation strategy of axis flipping, Gaussian blur, additive noise, multiplicative noise, etc. Then, they are divided into training set, validation set (used to verify the performance of the initial microwave image reconstruction network after training, which is a conventional technique and will not be elaborated here), and test set in a 7:2:1 ratio. Step 102-2: In the encoder-decoder architecture, lightweight multi-scale convergent convolutions are embedded in the encoder's downsampling process and the decoder's upsampling process to construct the initial microwave image reconstruction network, and the loss function of the initial microwave image reconstruction network is set to a dynamically weighted composite loss function. In practical applications, the downsampling process of the initial microwave image reconstruction network increases the number of feature channels from 64 to 128, 256, 512, and 1024 sequentially, while the upsampling process halves the number of channels in the reverse direction. Skip connections are used to fuse the feature maps of the encoder and decoder. MCC (Multi-Channel Convolution) is embedded during both the encoder's downsampling and decoder's upsampling processes, ensuring the network maintains sensitivity to multi-scale scattering features at each stage of feature compression and expansion. This completes the construction of the multi-scale convergent convolutional neural network architecture (i.e., the initial microwave image reconstruction network). Figure 3 As shown in the figure. In actual training and application, the upsampling and downsampling processes of the microwave image reconstruction network are similar to those of the initial microwave image reconstruction network.

[0023] Furthermore, considering the complex scattering characteristics of breast microwave images, the constructed network architecture (i.e., the initial microwave image reconstruction network) needs to be guided to have good reconstruction capabilities at both the pixel level and internal tissue type in complex breast images during the parameter learning process. Therefore, this application defines the aforementioned dynamically weighted composite loss function, expressed as: ; In the formula, It is a dynamically weighted composite loss function. For L1 loss function, The weight parameters of the L1 loss function, The structural similarity loss function is... These are the weight parameters of the structural similarity loss function. For edge loss function, These are the weight parameters of the edge loss function. For the perceptual loss function, These are the weight parameters for the perceptual loss function.

[0024] Among them, (1) loss function This ensures the pixel accuracy of breast images, providing a basic convergence target for the initial microwave image reconstruction network, defined as follows: .

[0025] In the formula, and These represent the initial breast reconstruction image (i.e., the true dielectric label of the breast) and the microwave breast image reconstructed by the network, respectively. This represents the number of samples.

[0026] (2) Structural similarity loss function This ensures the overall structure and visual perception quality of the image. By measuring the brightness, contrast, and structural information within a local window, it guides the network to focus on the consistency of the reconstructed image with the real label in terms of macroscopic structure. Its definition is as follows: .

[0027] In the formula, and These represent the first and second images in the initial breast reconstruction image, respectively. Local windows (total) The mean and variance of (a window), The covariance represents the difference between the initial reconstructed breast image and the microwave reconstructed breast image. and represents the constants used to maintain the numerical stability of the initial reconstructed mammary image and the network-reconstructed microwave mammary image, respectively. and The images of microwave breast tissue reconstructed by the network represent the first and second parts, respectively. Local windows (total) The mean and variance of (a window), and The values ​​represent the covariance of the initial breast reconstruction image and the microwave breast reconstruction image, respectively. express and Structural similarity.

[0028] (3) Edge loss function The aim is to specifically enhance the clarity of different tissue boundaries in microwave breast images reconstructed by a network, addressing the edge blurring problem that is easily caused by lightweight networks. Its definition is as follows: .

[0029] In the formula, This represents the Laplacian image operator used to extract edges.

[0030] (4) Perceptual loss function This ensures the semantic consistency of the reconstructed microwave breast images and guides the initial microwave image reconstruction network to generate more accurate images in terms of tissue type. Its definition is as follows: ; In the formula, This indicates that the feature extraction network uses the first 16 layers of the network as the perceptual loss function to avoid redundant computation caused by deep networks.

[0031] Step 102-3: Using training and testing sets, combined with a dynamically weighted composite loss function, train and test the initial microwave image reconstruction network until the trained initial microwave image reconstruction network meets the set conditions, and obtain the trained initial microwave image reconstruction network. During training, the learning rate, number of training epochs, optimizer, and batch size can be set according to actual needs. For example, the sample data of breast microwave scattering field (i.e., microwave breast image data) and the initial reconstructed breast image (label) in the training set can be used as input to train the parameters of the initial microwave image reconstruction network. The batch size of the training parameters of the initial microwave image reconstruction network is set to 8, the initial learning rate is 1e-3, and the learning rate is decayed to 1e-5 using a cosine annealing learning rate scheduling strategy. The number of training epochs is 100 (i.e., the set conditions for training to be met), and the initial microwave image reconstruction network is trained using Adam optimization. The training hardware is an NVIDIA GeForce RTX 4090 GPU.

[0032] The initial breast reconstruction images from the test set are input into the trained initial microwave image reconstruction network to output high-fidelity breast microwave images, and the image reconstruction metrics and network lightweight metrics are evaluated. Figure 4 This paper demonstrates a breast image reconstructed from a raw microwave breast image using an initial microwave image reconstruction network trained according to the present application. Based on... Figure 4 It is evident that the method provided in this application can effectively reconstruct the internal tissues of the breast, with very clear details and structure.

[0033] Step 102-4: Use the trained initial microwave image reconstruction network as the microwave image reconstruction network.

[0034] In one exemplary embodiment of this application, in order to demonstrate the beneficial effect of the method provided in this application on breast image reconstruction, this embodiment is illustrated by comparing the results of ablation experiments.

[0035] Table 1 presents the ablation experiment results comparing the performance of the MCC and dynamically weighted composite loss function proposed in this application for breast image reconstruction. As shown in Table 1, when using MCC (Experiment B) and the dynamically weighted composite loss function (Experiment C) respectively, compared to the baseline network (Experiment A), the Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM) of the breast image are improved to approximately 36 dB and 0.98, respectively. When using both the MCC and dynamically weighted composite loss function proposed in this application (Experiment D), PSNR and SSIM are significantly increased to 39.05 dB and 0.9900, respectively. This ablation experiment confirms the significant synergistic effect among the various features proposed in this application, fully demonstrating the beneficial effect of the proposed method for breast image reconstruction.

[0036] Table 1 Ablation experimental results of MCC module and dynamically weighted composite loss function

[0037] further, Figure 5 The performance comparison of microwave image reconstruction using the method provided in this application and without using the method is shown. The image reconstruction metrics PSNR and SSIM of the method provided in this application reach 39.05 dB and 0.9900, respectively, while the number of network parameters and floating-point operations (FLOPs) are only 13.34 M and 30.71 G (e.g., ...). Figure 6 (As shown). In contrast, the PSNR and SSIM of the UNet network without the method of this application are only 32.34 dB and 0.9557, respectively, but the number of parameters and FLOPs are as high as 31.04 M and 54.75 G. The above comparative data can illustrate that the method provided by this application, when implemented, achieves a significant lightweight effect by reducing the number of parameters by 56.9% and the computational cost by 43.9%, while improving the peak signal-to-noise ratio of the reconstructed image (i.e., the reconstructed microwave breast image) by 20.7% and the structural similarity by 3.6%. It effectively solves the technical contradiction of difficulty in balancing model complexity and reconstruction accuracy in the prior art, and has great practical value in clinical real-time imaging or portable devices. Figure 5 In the figure, (a) represents the peak signal-to-noise ratio results with and without using the method provided in this application, and (b) represents the structural similarity results with and without using the method provided in this application; Figure 6 In the text, (a) represents the parameter quantity results when the method provided in this application is not used and when the method provided in this application is used, and (b) represents the floating-point operation quantity results when the method provided in this application is not used and when the method provided in this application is used.

[0038] In summary, the method provided in this application not only significantly improves the peak signal-to-noise ratio and structural similarity of microwave breast image reconstruction, resulting in excellent image details and structure, but also greatly reduces the number of network parameters and computational complexity. It can quickly achieve breast imaging and achieve a significant lightweight effect. It can solve the problems of large number of parameters and high computational complexity in existing neural network breast image reconstruction models, thereby meeting the clinical application requirements for portability and lightweight design.

[0039] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 7As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores lightweight breast microwave image reconstruction data. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a lightweight breast microwave image reconstruction method.

[0040] Those skilled in the art will understand that Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer equipment to which the present application is applied. Specific computer equipment may include, for example, [the following is a list of possible additional structures]. Figure 7 The diagram shows more or fewer components, or combinations of certain components, or different component arrangements.

[0041] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0042] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0043] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0044] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (RRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0045] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0046] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0047] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A lightweight method for breast microwave image reconstruction, characterized in that, include: Microwave scattering field data of the breast were collected to obtain raw microwave breast images; Lightweight multi-scale convergent convolutions are constructed based on convolution modules with different numbers of groups; A microwave image reconstruction network is constructed based on the aforementioned lightweight multi-scale convergent convolution and combined with an encoder-decoder architecture. The microwave image reconstruction network is used to obtain a reconstructed microwave breast image based on the original microwave breast image.

2. The lightweight breast microwave image reconstruction method according to claim 1, characterized in that, Lightweight multi-scale convergent convolutions are constructed based on convolution modules with different numbers of groups, including: Construct multiple sets of convolutional modules; in the multiple sets of convolutional modules, perform convolutional channel transformation of a set size on the input feature map, and divide the convolutional channels and filters into a first preset number of groups to obtain the first preset number of convolutional modules; after performing independent convolutional operations on each group of convolutional modules according to a set convolutional kernel, batch normalization and modified linear unit processing are performed, and the output feature maps of each group of convolutional modules are spliced ​​together to obtain the output feature maps of multiple sets of convolutional modules; Construct a few-group convolutional module; in the few-group convolutional module, the output feature maps of the filter and the multiple-group convolutional module are divided into a second preset number of groups to obtain multiple feature sub-maps; according to the set convolutional kernel, the multiple feature sub-maps are batch normalized and modified linear unit processed to obtain the output feature map of the few-group convolutional module; the second preset number of groups is less than the first preset number of groups; The multiple sets of convolutional modules and the few sets of convolutional modules are stacked according to a set number of times, and the input feature maps of the multiple sets of convolutional modules and the output feature maps of the few sets of convolutional modules are subjected to residual processing to obtain the output of the multi-scale convergent convolution, thereby completing the construction of the multi-scale convergent convolution.

3. The lightweight breast microwave image reconstruction method according to claim 1, characterized in that, Based on the aforementioned lightweight multi-scale convergent convolution, a microwave image reconstruction network is constructed using an encoder-decoder architecture, including: Construct a sample dataset and divide it into a training set and a test set according to a set ratio; In the encoder-decoder architecture, the lightweight multi-scale convergent convolution is embedded in the downsampling process of the encoder and the upsampling process of the decoder to construct an initial microwave image reconstruction network, and the loss function of the initial microwave image reconstruction network is set as a dynamically weighted composite loss function. The initial microwave image reconstruction network is trained and tested using the training set and the test set, combined with the dynamically weighted composite loss function, until the trained initial microwave image reconstruction network meets the set conditions, thus obtaining the trained initial microwave image reconstruction network. The trained initial microwave image reconstruction network is used as the microwave image reconstruction network.

4. The lightweight breast microwave image reconstruction method according to claim 2, characterized in that, Construct a sample dataset, including: Microwave scattering field sample data of breast tissue was acquired, and the inverse scattering solution of the breast microwave scattering field sample data was performed using the Born iterative method. Tikhonov regularization term was introduced in the solution process to obtain the initial reconstructed image set of breast tissue. The images in the initial breast reconstruction image set were augmented using a hybrid data augmentation strategy to obtain the sample dataset.

5. The lightweight breast microwave image reconstruction method according to claim 4, characterized in that, The hybrid data augmentation strategy includes at least one of axis flipping processing, Gaussian blur processing, additive noise processing, and multiplicative noise processing.

6. The lightweight breast microwave image reconstruction method according to claim 4, characterized in that, The breast microwave scattering field sample data was inversely scattered using the Born iterative method, and the objective function of the initial reconstructed breast image obtained by introducing a Tikhonov regularization term during the solution process is expressed as follows: ； ； ； In the formula, This is an initial reconstructed image of the breast. As slack variables, In the first Scattered fields collected at each antenna location The number of antenna positions. For the scattered field, For the incident field, It is the identity matrix. For the Green's function of the probe region, For the Green's function of the target region, For Tikhonov regularization terms, The initial reconstructed image of the breast to be optimized. For regularization weights, The initial reconstructed image of the breast when the objective function is minimized. The slack variables are those that minimize the objective function. For complex fields, Indicates a condition.

7. The lightweight breast microwave image reconstruction method according to claim 3, characterized in that, The dynamically weighted composite loss function is expressed as: ； In the formula, It is a dynamically weighted composite loss function. For L1 loss function, The weight parameters of the L1 loss function, The structural similarity loss function is... These are the weight parameters of the structural similarity loss function. For edge loss function, These are the weight parameters of the edge loss function. For the perceptual loss function, These are the weight parameters for the perceptual loss function.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that the processor executes the computer program to implement the lightweight breast microwave image reconstruction method according to any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the lightweight breast microwave image reconstruction method according to any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the lightweight breast microwave image reconstruction method according to any one of claims 1-7.

Images