Palmprint image recognition method based on complex-valued convolutional neural network

Abstract

This invention provides a palmprint image recognition method based on a complex-valued convolutional neural network, relating to the technical field of biometric feature recognition. The method involves converting an initial palmprint image into an HSV image and generating a complex-valued palmprint image in iHSV format using a complex-valued image conversion model. A fully complex-valued palmprint image recognition network is constructed, and the network parameters of the palmprint image recognition network are optimized using a complex-valued dynamic activation function and a complex-valued dynamic loss function. The complex-valued palmprint image is input into the optimized palmprint image recognition network to obtain an output result. The output result is matched with palmprint data in a database to output a recognition result. This method addresses the issues of palmprint recognition with small samples and improves palmprint recognition accuracy.

Description

Technical Field

[0001] The present application relates to the field of biometric recognition technology, and in particular to a palmprint image recognition method based on a complex-valued convolutional neural network. Background Art

[0002] Palmprint recognition is a human identification technology that uses the right or left palm in a manner similar to fingerprints. Compared to more common biometric features such as faces and fingerprints, palmprints offer advantages such as a large acquisition area, rich texture information, low acquisition equipment requirements, and high security. In particular, the data matching process can be performed using low-resolution images. Therefore, palmprints have attracted widespread attention in the field of biometric recognition. Furthermore, users have greater initiative during the palmprint acquisition process and can decide whether to show their palmprints. Furthermore, palmprints are less likely to be secretly captured by surveillance cameras, thus reducing the risk of privacy violations. Therefore, palmprints have shown great potential in identification applications. Early palmprint recognition mainly relied on manual annotation of unique palmprint feature points, such as main lines, wrinkles, fine textures, ridge ends, and bifurcations. The locations of these feature points and the surrounding texture information were calculated and then compared with a palmprint database to produce recognition results. However, this method relies heavily on manual intervention and is subject to high subjectivity. For example, different operators may have different understandings of the locations of palmprint feature points, which inevitably leads to errors in the annotation process. Even for the same palmprint, different operators may produce different annotation results. In addition, this method is time-consuming, but the recognition efficiency and accuracy do not achieve ideal results.

[0003] Therefore, how to improve the accuracy of palmprint recognition in small sample scenarios has become a technical problem that needs to be solved urgently. Summary of the Invention

[0004] In order to improve the accuracy of palmprint recognition in small sample scenarios, the present application provides a palmprint image recognition method and related devices based on a complex-valued convolutional neural network.

[0005] The present application provides a palmprint image recognition method based on a complex-valued convolutional neural network using the following technical solutions:

[0006] A palmprint image recognition method based on a complex-valued convolutional neural network, comprising:

[0007] Convert the acquired initial palmprint image into an HSV image, and generate a complex-valued palmprint image in iHSV format through a complex-valued image conversion model;

[0008] Constructing a palmprint image recognition network with full complex-valued operations, the network includes a complex-valued convolution module, a complex-valued pooling module, a complex-valued normalization module, a complex-valued residual block, and a complex-valued dense block, wherein the complex-valued residual block and the complex-valued dense block are connected by a conversion layer, and the conversion layer is composed of complex-valued convolution and complex-valued pooling;

[0009] Using a complex-valued dynamic activation function and a complex-valued dynamic loss function to optimize the network parameters of the palmprint image recognition network;

[0010] Inputting the complex-valued palmprint image into the optimized palmprint image recognition network and obtaining an output result;

[0011] The output result is matched with the palmprint data in the database to output the recognition result.

[0012] Optionally, the step of converting the acquired initial palmprint image into an HSV image and generating a complex-valued palmprint image in iHSV format through a complex-valued image conversion model includes:

[0013] Get the initial palmprint image and convert the RGB format palmprint image into HSV format image , where h represents hue, S represents saturation, and V represents lightness;

[0014] Convert HSV format image to iHSV complex value format image , achieved through the following channel settings:

[0015] Hue Fixed Channel , with saturation S as the real axis and lightness V as the imaginary axis;

[0016] Saturation fixed channel , with lightness V as the real axis and the product of saturation and hue as the imaginary axis;

[0017] Luminance fixed channel , generated by polar coordinate transformation;

[0018] The real and imaginary parts of the three channels are separated and recombined to generate a complete complex-valued image.

[0019] Optionally, the complex-valued residual block includes at least two residual structures:

[0020] Residual structure 1: Through complex-valued convolution and complex-valued normalization, the number of input and output channels are made consistent, and weighted fusion is performed;

[0021] Residual structure 2: The input data is processed by a complex-valued activation function and then weightedly fused with the original input.

[0022] Optionally, each layer in the complex-valued dense block receives the outputs of all previous layers as input and is processed by a nonlinear transformation module, which consists of complex-valued normalization, complex-valued activation function and complex-valued convolution.

[0023] Optionally, the complex-valued dynamic activation function is defined as:

[0024] ;

[0025] Among them, the plural ; ;

[0026] ;

[0027] Indicates the restricted area for dynamic adjustment of complex phase, Indicates dynamic adjustment of the ReLU activation threshold.

[0028] Optionally, the complex-valued dynamic loss function By complex cross entropy loss function and the complex-valued weighted triplet loss function composition:

[0029] ;

[0030] in:

[0031] ;

[0032] ;

[0033] represents the complex output of the model;

[0034] ;

[0035] in, represents the anchor point sample, represents a positive sample, represents negative samples, and Represent the weights of positive and negative samples respectively.

[0036] Optionally, the distance metric in the complex-valued weighted triple loss function is calculated by the following formula:

[0037] .

[0038] In summary, this application addresses the challenges of small-sample palmprint recognition by converting an initial palmprint image into an HSV image and generating a complex-valued palmprint image in iHSV format using a complex-valued image conversion model. Furthermore, a fully complex-valued palmprint image recognition network is constructed, and its parameters are optimized using a complex-valued dynamic activation function and a complex-valued dynamic loss function. The complex-valued palmprint image is input into the optimized palmprint image recognition network to obtain an output result. Finally, the output result is matched with palmprint data in a database to output a recognition result. This approach addresses the challenges of small-sample palmprint recognition and improves palmprint recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a flowchart of the first embodiment of the palmprint image recognition method based on a complex-valued convolutional neural network of the present application;

[0040] Figure 2 is a flow chart of the complex value conversion model of this application;

[0041] Figure 3 This is the overall identification model diagram of this application;

[0042] Figure 4 This is the structure diagram of the four complex-valued residual blocks in the complex-valued convolutional neural network model of this application, where Figure 4 (a) in the figure is the complex-valued residual block 1; Figure 4 (b) in the figure is the complex-valued residual block 2; Figure 4 (c) in the figure is the complex-valued residual block 3; Figure 4 (d) in the figure is the complex-valued residual block 4;

[0043] Figure 5 These are the two major residual structure diagrams of this application, where Figure 5 (a) in the equation represents the residual structure 1. Figure 5 (b) in the figure is residual structure 2;

[0044] Figure 6 is the complex-valued dense block graph of this application. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solutions and advantages of this application more clearly understood, the present application is further described in detail below through the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0046] The present invention provides a palmprint image recognition method based on a complex-valued convolutional neural network. Figure 1 , Figure 1 This is a flowchart of the first embodiment of the palmprint image recognition method based on complex-valued convolutional neural network of this application.

[0047] In this embodiment, the palmprint image recognition method based on the complex-valued convolutional neural network includes the following steps:

[0048] Step S10: converting the acquired initial palmprint image into an HSV image, and generating a complex-valued palmprint image in iHSV format through a complex-valued image conversion model.

[0049] The step of converting the acquired initial palmprint image into an HSV image and generating a complex-valued palmprint image in iHSV format through a complex-valued image conversion model includes:

[0050] Get the initial palmprint image and convert the RGB format palmprint image into HSV format image , where h represents hue, S represents saturation, and V represents lightness;

[0051] Convert HSV format image to iHSV complex value format image , achieved through the following channel settings:

[0052] Hue Fixed Channel , with saturation S as the real axis and lightness V as the imaginary axis;

[0053] Saturation fixed channel , with lightness V as the real axis and the product of saturation and hue as the imaginary axis;

[0054] Luminance fixed channel , generated by polar coordinate transformation;

[0055] The real and imaginary parts of the three channels are separated and recombined to generate a complete complex-valued image.

[0056] It's important to note that the HSV model describes color using hue, saturation, and value, which better aligns with human intuitive color perception. Color comparison and matching in the HSV model is generally more intuitive and accurate than in the RGB model because changes in hue, saturation, and value are independent and linear. Converting raw palmprint images to HSV format facilitates subsequent recognition because it makes it easier to distinguish and isolate objects with similar colors. HSV is used to express the three primary attributes of color: hue, saturation, and value. Hue, saturation, and value represent the angular, radial, and vertical axes of the cylindrical HSV color model.

[0057] In the specific implementation, the image conversion process is carried out around three channels, namely the hue fixed channel , saturation fixed channel and lightness fixed hue .

[0058] Hue Fixed Channel : Fix the hue element to form a rectangular plane. Take saturation and lightness as the coordinate axes, with saturation as the real axis and lightness as the imaginary axis, and then derive its position in the coordinate axis as shown in the following formula:

[0059] ;

[0060] Saturation fixed channel : Fix the saturation element to form a hollow cylinder, which is then expanded to form a rectangular plane with lightness as the real axis and the product of saturation and hue as the imaginary axis. Then, its position in the coordinate axis is derived as shown in the following formula:

[0061] ;

[0062] Brightness fixed hue : Fix the saturation element to form a circular plane, so the polar coordinates are established on this plane, and the center angle is As the unit, its position in the coordinate axis is derived as follows:

[0063] ;

[0064] The real and imaginary parts of the above three channels are recombined to form the real and imaginary parts corresponding to the complex-valued image. The process of the entire complex-valued conversion model is as follows: Figure 2 shown.

[0065] Step S20: constructing a palmprint image recognition network with full complex-valued operations, the network including a complex-valued convolution module, a complex-valued pooling module, a complex-valued normalization module, a complex-valued residual block and a complex-valued dense block, wherein the complex-valued residual block and the complex-valued dense block are connected through a conversion layer, and the conversion layer is composed of complex-valued convolution and complex-valued pooling.

[0066] The complex-valued residual block includes at least two residual structures:

[0067] Residual structure 1: Through complex-valued convolution and complex-valued normalization, the number of input and output channels are made consistent, and weighted fusion is performed;

[0068] Residual structure 2: The input data is processed by a complex-valued activation function and then weightedly fused with the original input.

[0069] Each layer in the complex-valued dense block receives the output of all previous layers as input and is processed by a nonlinear transformation module, which consists of complex-valued normalization, complex-valued activation function and complex-valued convolution.

[0070] In the specific implementation, the main body of the overall recognition model is composed of complex-valued residual blocks and complex-valued dense blocks. The combination of the two modules can better extract the corresponding features of the image, while solving the problems of gradient disappearance and gradient explosion. The two modules are connected by a conversion layer, which is composed of complex-valued convolution and complex-valued pooling. The conversion layer can effectively reduce the size of the feature map and make the feature map data consistent during transmission. The introduction of the designed complex-valued activation function and complex-valued loss function further improves the efficiency of the model. The overall recognition model is as follows Figure 3 Each complex-valued residual block consists of two residual structures, but different residual blocks have different numbers of residual structures. The four complex-valued residual block structures are as follows: Figure 4 shown. Figure 4 Represents the four complex-valued residual block structures in the complex-valued convolutional neural network model, Figure 4 (a) in FIG1 is a complex-valued residual block 1, which consists of a residual structure 1 and two residual structures 2. Figure 4 (b) in FIG is a complex-valued residual block 2, which consists of a residual block structure 1 and three residual structures 2. Figure 4 (c) in FIG. 3 is a complex-valued residual block 3, which consists of one residual block structure 1 and five residual structures 2. Figure 4 (d) in FIG. 4 is a complex-valued residual block 4, which is composed of a residual block structure 1 and two residual structures 2.

[0071] Residual structure 1 adds a complex-valued convolution and complex-valued regularization process to make the number of output channels at both ends consistent, and then performs a weighted sum. Residual structure 2 directly performs a weighted sum of the input data and the processed score data through a complex-valued activation function. Figure 5 shown. Figure 5 represents the two residual structures in the complex-valued residual block, where Figure 5 (a) in the figure represents the residual structure 1, which adds a complex-valued convolution and a complex-valued regularization process to make the number of output channels at both ends consistent, and then performs a weighted sum. Figure 5 (b) in the figure is the residual structure 2, which directly performs a weighted sum of the input data and the processed score data through a complex-valued activation function.

[0072] Complex-valued dense blocks such as Figure 6 As shown in , each layer Xi in the block receives all previous layers as input. Where C represents the nonlinear transformation process, which consists of complex-valued normalization, complex-valued activation function and complex-valued convolution.

[0073] Step S30: Optimizing the network parameters of the palmprint image recognition network using a complex-valued dynamic activation function and a complex-valued dynamic loss function.

[0074] It's understandable that activation functions can introduce nonlinearity into models, allowing neural networks to learn complex patterns. However, traditional activation functions can only process real values, not complex values. Therefore, for fully complex-valued networks, a complex-valued activation function that can efficiently process complex values is needed. Existing complex-valued activation functions have some flaws. For example, while the ZReLU activation function only retains positive real and imaginary parts, the CReLU activation function processes the real and imaginary parts of a complex number separately through the ReLU activation function. Such complex-valued activation functions can lead to problems such as information fragmentation and destroy the internal correlation of complex numbers. Therefore, a dynamic complex-valued activation function, DCZReLU, is designed to address the shortcomings of existing activation functions by introducing dynamic parameter adjustment.

[0075] It should be noted that the complex-valued dynamic activation function is defined as:

[0076] ;

[0077] Among them, the plural ; ;

[0078] ;

[0079] Indicates the restricted area for dynamic adjustment of complex phase, Indicates dynamic adjustment of the ReLU activation threshold.

[0080] In practice, the loss function quantifies the difference between the model's predicted value and the true value, reflecting the model's performance. By minimizing the loss function, the model can gradually adjust its parameters to improve prediction accuracy. The loss function can also provide direction for optimization algorithms (such as gradient descent), helping the model to continuously improve during training. The cross-entropy loss function focuses on optimizing the probability distribution of classification tasks, ensuring that the model output is consistent with the true label. The triple loss function, on the other hand, enhances feature differentiation through contrastive learning, bringing similar samples closer together and heterogeneous samples further apart. This embodiment designs a complex-valued dynamic triple loss function and dynamically combines it with the complex-valued cross-entropy loss function. Combining the advantages of both, the model can both optimize classification probabilities and learn more discriminative features. At the same time, it can balance the classification task and feature learning, reduce the risk of overfitting, and improve the model's generalization performance. The dynamic combination can further optimize feature learning and classification performance for complex-valued data.

[0081] It is understandable that the complex-valued dynamic loss function By complex-valued cross entropy loss function and the complex-valued weighted triplet loss function composition:

[0082] ;

[0083] in:

[0084] ;

[0085] ;

[0086] represents the complex output of the model;

[0087] ;

[0088] in, represents the anchor point sample, represents a positive sample, represents negative samples, and Represent the weights of positive and negative samples respectively.

[0089] In a specific implementation, the distance metric in the complex-valued weighted triple loss function is calculated by the following formula:

[0090] .

[0091] Step S40: inputting the complex-valued palmprint image into the optimized palmprint image recognition network and obtaining an output result.

[0092] Step S50: matching the output result with the palmprint data in the database and outputting the recognition result.

[0093] It's worth noting that the emergence of convolutional neural networks has provided a practical solution for palmprint recognition based on deep learning algorithms. This field has brought new directions to biometric recognition. Convolutional neural networks typically extract features from regions of interest in an image, perform convolution operations, and then use the extracted texture features for linear classification, significantly reducing labor costs. Compared with traditional palmprint recognition methods, deep learning-based palmprint recognition reduces human intervention and avoids the influence of subjective factors. It can also extract effective features from low-resolution palmprint images, offering advantages such as high recognition accuracy and strong generalization. Complex-valued neural networks have also been a research hotspot in recent years. In fields such as speech enhancement, image, and signal processing, complex-valued neural networks have demonstrated superior performance compared to real networks. Currently, the vast majority of deep learning building blocks, techniques, and architectures are based on real-valued operations and representations. However, recent research on recurrent neural networks and earlier fundamental theoretical analysis suggest that complex numbers may have richer representational capabilities and may also promote noise-resistant memory retrieval mechanisms. Currently, mainstream convolutional neural networks are also largely real-valued. While real-valued networks have achieved significant results in computer vision, complex-valued networks still have certain advantages. For example, complex numbers possess a range of advantageous properties that real numbers lack, such as those for expressing vector aggregation, rotation, and exponential operations. Complex numbers can simultaneously process both amplitude and phase information, making the model more flexible in capturing data complexity. This dual information helps simplify model representation. Complex-valued functions can operate in higher-dimensional feature spaces, enabling the model to better separate data from different categories. Higher dimensions provide more degrees of freedom for decision boundaries. Orthogonal decision boundaries in complex-valued networks can effectively reduce interference between categories, making the model more stable on new data and improving generalization. Because complex-valued neural networks can effectively express complex patterns in higher dimensions, they can often achieve the same performance with fewer parameters, which helps reduce the risk of overfitting. This embodiment, based on practical application scenarios, reduces the cost of palmprint data acquisition and recognition model training, improving palmprint recognition accuracy in small sample scenarios. Furthermore, by operating directly on the image in the complex domain, without converting the data into separate real and imaginary components, computational complexity is significantly reduced, while also improving the model's ability to resist interference.

[0094] It should be noted that the effects achieved by the method of this embodiment include:

[0095] Improving recognition performance in small sample scenarios: By combining a complex-valued convolutional neural network (CV-CNN) architecture with complex-valued image transformation and a dynamic loss function, the accuracy and efficiency of palmprint recognition with small samples are significantly improved. Complex-valued networks can fully utilize phase and amplitude information in the complex domain, reducing reliance on large-scale training data and addressing the recognition bottleneck caused by traditional real-valued networks that ignore phase information.

[0096] Enhanced feature representation capabilities:

[0097] The complex-valued image conversion model (iHSV format) combines the intuitive perception advantages of the HSV color model with the multidimensional representation capabilities of complex numbers to map the hue, saturation, and brightness of the palmprint into complex domain features, effectively preserving the texture details and phase information of the palmprint;

[0098] The combination of complex-valued residual blocks and dense blocks enables the extraction and reuse of multi-level semantic features. Through gradient optimization and feature aggregation mechanisms, the model's sensitivity to subtle differences in palmprints is improved.

[0099] Optimize training stability and model depth:

[0100] The complex-valued residual block alleviates the gradient vanishing and exploding problems by weightedly fusing input and output features, supporting the training of deeper networks.

[0101] The complex-valued dense block enhances the nonlinear expression capability of the model by reusing multi-scale features while reducing parameter redundancy.

[0102] Dynamic adaptive mechanism improves generalization ability:

[0103] The dynamic complex-valued activation function (DCZReLU) avoids information fragmentation of complex features and maintains internal correlation of complex numbers by adjusting the phase limit parameter (α) and the activation threshold (β);

[0104] The dynamic combined loss function (cross entropy + triple loss) adaptively balances the classification task and feature discriminability through weight parameters (α, β), reduces the risk of overfitting, and enhances the model's robustness to noise and interference.

[0105] Efficient end-to-end complex stream processing:

[0106] The fully complex-valued operation process does not require data to be split into real and imaginary parts for separate processing, reducing computational complexity and information loss.

[0107] The end-to-end architecture simplifies the palmprint recognition process and improves computing efficiency, making it suitable for real-time or resource-constrained scenarios.

[0108] Reduce application costs:

[0109] Efficient training in small sample scenarios reduces the demand for labeled data and computing resources;

[0110] The high-precision recognition capability reduces the dependence on high-resolution acquisition equipment and broadens the application prospects of palmprint recognition in low-cost terminal devices.

[0111] This embodiment significantly improves the accuracy, efficiency, and robustness of palmprint recognition through the combination of complex-valued image conversion, dynamic complex-valued network architecture, and adaptive optimization mechanism. It has outstanding practical value, especially in small sample and low-resolution scenarios, and provides an efficient and low-cost solution for the field of biometric recognition.

[0112] This embodiment converts the acquired initial palmprint image into an HSV image and generates a complex-valued palmprint image in iHSV format using a complex-valued image conversion model. A fully complex-valued palmprint image recognition network is constructed, and the network parameters of the network are optimized using a complex-valued dynamic activation function and a complex-valued dynamic loss function. The complex-valued palmprint image is input into the optimized palmprint image recognition network to obtain an output result. The output result is matched with palmprint data in a database to output a recognition result. This solves the problem of small-sample palmprint recognition and improves the accuracy of palmprint recognition.

[0113] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of this application. In actual applications, technicians in this field can select part or all of it according to actual needs to achieve the purpose of this embodiment scheme, and no restrictions are imposed here.

[0114] In addition, for technical details not fully described in this embodiment, please refer to the palmprint image recognition method based on complex-valued convolutional neural network provided in any embodiment of the present application, which will not be repeated here.

[0115] In addition, it should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or system comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or system. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or system comprising the element.

[0116] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0117] Through the above description of the embodiments, those skilled in the art will clearly understand that the above-mentioned embodiments and methods can be implemented using software plus the necessary general-purpose hardware platform. Of course, hardware can also be used, but in many cases the former is a more preferred embodiment. Based on this understanding, the technical solution of this application, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory (ROM) / RAM, a magnetic disk, or an optical disk) and includes several instructions for enabling a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods of each embodiment of this application. The above are only preferred embodiments of this application and do not limit the scope of the patent application. Any equivalent structure or equivalent process transformation made using the contents of this application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the scope of patent protection of this application.

Claims

1. A palmprint image recognition method based on complex-valued convolutional neural network, characterized in that: include: Convert the acquired initial palmprint image into an HSV image, and generate a complex-valued palmprint image in iHSV format through a complex-valued image conversion model; Constructing a palmprint image recognition network with full complex-valued operations, the network includes a complex-valued convolution module, a complex-valued pooling module, a complex-valued normalization module, a complex-valued residual block, and a complex-valued dense block, wherein the complex-valued residual block and the complex-valued dense block are connected by a conversion layer, and the conversion layer is composed of complex-valued convolution and complex-valued pooling; Using a complex-valued dynamic activation function and a complex-valued dynamic loss function to optimize the network parameters of the palmprint image recognition network; Inputting the complex-valued palmprint image into the optimized palmprint image recognition network and obtaining an output result; Matching the output result with the palmprint data in the database and outputting a recognition result; The complex-valued dynamic activation function is defined as: ； Among them, the plural ; ; ； Indicates the restricted area for dynamic adjustment of complex phase, Indicates dynamic adjustment of the ReLU activation threshold; Among them, the complex-valued dynamic loss function By complex cross entropy loss function and the complex-valued weighted triplet loss function composition: ； in: ； ； represents the complex output of the model; ； in, represents the anchor point sample, represents a positive sample, represents negative samples, and Represent the weights of positive and negative samples respectively.

2. The method according to claim 1, characterized in that The step of converting the acquired initial palmprint image into an HSV image and generating a complex-valued palmprint image in iHSV format through a complex-valued image conversion model includes: Get the initial palmprint image and convert the RGB format palmprint image into HSV format image , where h represents hue, S represents saturation, and V represents lightness; Convert HSV format image to iHSV complex value format image , achieved through the following three channel settings: Hue Fixed Channel , with saturation S as the real axis and lightness V as the imaginary axis; Saturation fixed channel , with lightness V as the real axis and the product of saturation and hue as the imaginary axis; Luminance fixed channel , generated by polar coordinate transformation; The real and imaginary parts of the three channels are separated and recombined to generate a complete complex-valued image.

3. The method according to claim 1, characterized in that The complex-valued residual block includes at least two residual structures: Residual structure 1: Through complex-valued convolution and complex-valued normalization, the number of input and output channels are made consistent, and weighted fusion is performed; Residual structure 2: The input data is processed by a complex-valued activation function and then weightedly fused with the original input.

4. The method according to claim 1, wherein Each layer in the complex-valued dense block receives the output of all previous layers as input and is processed by a nonlinear transformation module, which consists of complex-valued normalization, complex-valued activation function and complex-valued convolution.

5. The method according to claim 1, wherein The distance metric in the complex-valued weighted triplet loss function is calculated by the following formula: 。

Images