Multimodal currency recognition system, method, storage medium and terminal

Abstract

The application provides a multimodal currency identification system, method, storage medium and terminal, and belongs to the technical field of currency identification. The system and method of the application adopt texture perception and open set identification technology, improve the identification capability of deep features through multimodal feature fusion, and improve the identification performance of the texture perception multi-currency identification algorithm. The open set identification algorithm can realize the identification of new currencies and new denomination banknotes through a small number of samples, and has low overall development cost and high identification precision.

Description

Technical Field

[0001] This invention belongs to the field of currency recognition technology, specifically relating to a multimodal currency recognition system, method, storage medium, and terminal based on texture perception and open set recognition. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art or art.

[0003] With the development of economic globalization, multiple currencies often circulate in the markets of many countries, regions, and organizations. However, the variety of currencies across countries and regions is vast, and some currencies even share certain similarities. This makes it impossible to effectively identify a large number of currencies and their denominations manually. Therefore, how to efficiently distinguish between multiple currencies, denominations, and counterfeits is a pressing problem for businesses and financial organizations such as banks. However, current currency identification systems still have some issues in use, for the following reasons:

[0004] 1. Performance in recognizing multiple currencies and denominations needs improvement: See [link / reference] Figure 1 Market-ready banknote recognition algorithms typically employ deep convolutional neural networks (CNNs) with numerous parameters, such as ResNet and VGGNet. However, the convolution operations used in deep CNNs obtain global features through weight sharing. Therefore, during currency sample recognition, as the network depth increases, the texture features (local features) of the currency are gradually lost. This leads to a high false recognition rate when currencies have a certain degree of similarity (especially when color features are highly similar). Furthermore, conventional banknote recognition methods often employ closed-set recognition, meaning the currency ID and denomination are known, and the corresponding ID and denomination are used during training. However, this closed-set recognition method cannot handle new currency IDs and denominations.

[0005] 2. Long Development Time and High Development Costs: To achieve high recognition rates, the market often uses one algorithm per currency. This leads to the need to develop multiple algorithms in the multi-currency recognition market. Developing a multi-algorithm recognition system not only requires significant human resources, but also necessitates the retraining of large datasets and even algorithm redesign when new currencies or denominations emerge. This results in long development times and high costs. Therefore, a technical solution is needed that eliminates the need for algorithm retraining when faced with new currencies or denominations.

[0006] 3. Currency data collection and storage issues. Conventional convolutional neural network algorithms often adopt a data-driven approach, that is, training the model by collecting a large amount of sample data. However, currency data collection often faces many problems, such as exchange rate issues and security issues when storing large amounts of currency. Therefore, a technical solution is needed that can identify new currencies and new denominations of currency using only a small number of samples.

[0007] In conclusion, current multi-currency recognition technology still has many shortcomings in its application. Summary of the Invention

[0008] In order to overcome the shortcomings of the prior art, the present invention aims to provide a multimodal currency recognition system, method, storage medium and terminal that can solve the above-mentioned problems.

[0009] The specific plan is as follows:

[0010] A multimodal currency recognition system based on texture perception and open set recognition is disclosed. The system includes an image acquisition module, an image processing module, a currency recognition module, and a result output module connected by telecommunications. The image acquisition module employs a multimodal sensor to generate a multimodal feature map of the currency to be recognized. The image processing module includes an image preprocessing unit and a feature extraction unit. The image preprocessing unit performs unified preprocessing on the received multimodal feature map, while the feature extraction unit acquires global features of the image through the backbone network of the image processing algorithm and local features through the texture perception module of the image processing algorithm. The currency recognition module includes a storage unit and a comparison and determination unit. The storage unit stores a template library and a test library. The comparison and determination unit compares the received feature information with the currency features in the template library to determine the classification and denomination of the currency to be tested. The result output module visualizes the recognition results of the currency recognition module through a display unit.

[0011] Furthermore, the multimodal sensor includes a visible light sensor, an infrared light sensor, and an ultraviolet light sensor, used to acquire multimodal feature maps of both sides of the coin.

[0012] Furthermore, the backbone network of the image processing module adopts a 10-channel recognition neural network, and the received image features include visible light RGB three-channel image features, infrared light image features, and ultraviolet light image features corresponding to the front and back of the currency, totaling 10 channels of image feature information.

[0013] Furthermore, the texture perception module of the image processing module is configured as follows:

[0014] f1(x, n) = BConv size = (1, 2n-1)(x) (1)

[0015] f2(x, n) = BConv size = (2n-1, 1)(x) (2)

[0016] f3(x, n) = BConv dilate＝2n -1(x) (3)

[0017] x1, x2, x3, x4=Splited(X) (4)

[0018]

[0019] RF = f1(X, 1) + Cat n＝1，2，3，4 F n (6)

[0020] In the formula, n∈[1,2,3,4] represents the branch number.

[0021] Formula (1) represents the convolution operation and normalization operation with a kernel size of (1, 2n-1);

[0022] Formula (2) represents the convolution operation with kernel size (2n-1, 1) and the normalization operation;

[0023] Formula (3) represents the convolution operation with kernel size (3, 3) and dilation rate of 2n-1, as well as the normalization operation;

[0024] Formula (4) indicates that the input feature X is cut into x n ;

[0025] Formula (5) represents the branch network in the texture-aware module;

[0026] Formula (6) represents the texture-aware module that introduces a residual mechanism.

[0027] Furthermore, the algorithm model parameters corresponding to the image processing module include the number of iterations, learning rate, loss function, and training optimization function. The loss function adopts the Tripletloss function, the training optimization function adopts the SGD method, and the parameters inside the model are updated through backpropagation.

[0028] Furthermore, the currency recognition module employs an open set recognition algorithm comprising a cosine similarity calculation unit, a comparison unit, and a threshold determination unit. The cosine similarity calculation unit is used to calculate and sort the cosine similarity of the features of the currency to be tested. The comparison unit is used to extract the highest cosine similarity score and match it with the template label in the template library. The threshold determination unit is used to set a threshold and extract valid matching results.

[0029] The present invention also provides a currency identification method, the method comprising:

[0030] S1. Data acquisition: Collect a dataset of currencies with different denominations and values, including both genuine and counterfeit currencies, using a multimodal sensor. Therefore, each currency corresponds to a multimodal image under visible light, ultraviolet light, and red light.

[0031] S2. Dataset labeling: Label the currency ID, face value, and authenticity of the collected currency images, so that each currency image contains three labels: currency ID, face value, and authenticity.

[0032] S3. Model training: A currency recognition model with 10 input channels and a texture perception module is trained by data input, model construction, and model parameter setting.

[0033] S4. Model testing: The model is tested and evaluated through data partitioning, feature extraction, and feature comparison.

[0034] S5. Model Application: Data is collected on the currency to be tested, and the collected multimodal features are imported into the currency recognition model. The recognition results are then visualized.

[0035] Furthermore, the model application in step S5 includes:

[0036] S51. Embed the currency recognition model with algorithm, algorithm parameters, and model weights into the image processing device;

[0037] S52. Establish a template library containing a large number of currency IDs and different denominations, including genuine and counterfeit images. Input the images in the template library into the processing device and extract the features of the images in the template library, and store the features and their labels.

[0038] S53. The currency to be detected is subjected to multimodal sensor to collect multimodal features, and the multimodal features are input into the algorithm to obtain identification features;

[0039] S54. The identification feature is compared with the feature in the template library by score comparison. The template label in the template library corresponding to the highest score is the prediction result of the test currency.

[0040] S55. When new currencies or new denominations of currency appear, collect a small number of currency samples, label the currencies, and add the features of the new currencies to the template library.

[0041] S56. Repeat steps S53-S55 to update and manage the currency recognition module.

[0042] The present invention also provides a computer-readable storage medium having computer instructions stored thereon, which, when executed, perform the steps of the aforementioned method.

[0043] The present invention also provides a terminal, including a memory and a processor, wherein the memory stores a template library and computer instructions that can be executed on the processor, and the processor executes the steps of the aforementioned method when executing the computer instructions.

[0044] Compared with the prior art, the beneficial effects of the present invention are as follows: The system and method of this application adopt texture perception and open set recognition technology, improve the recognition ability of deep features through multimodal feature fusion, improve the recognition performance of multi-currency recognition algorithm of texture perception, and realize the recognition of new currency and new denomination banknotes with a small number of samples. The overall development cost is low and the recognition accuracy is high. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of an existing recognition algorithm;

[0046] Figure 2 This is a schematic diagram of a multimodal currency recognition system.

[0047] Figure 3 This is a schematic diagram of multimodal feature fusion.

[0048] Figure 4 This is a diagram illustrating the texture-aware algorithm process.

[0049] Figure 5 Here is a structural diagram of the texture perception module;

[0050] Figure 6 This is a flowchart illustrating the process of an open set recognition algorithm based on texture awareness.

[0051] Figure 7 This is a flowchart of a currency identification method. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] Flowcharts are used in this specification to illustrate the operations performed by the system according to embodiments of this specification. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, the steps can be processed in reverse order or simultaneously. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0054] First Embodiment

[0055] A multimodal currency recognition system based on texture perception and open set recognition, see [link to relevant documentation]. Figure 2 The system includes an image acquisition module, an image processing module, a currency recognition module, and a result output module connected by telecommunications.

[0056] The image acquisition module employs a multimodal sensor to generate multimodal feature maps of the currency to be identified. Specifically, the multimodal sensor includes a visible light sensor, an infrared light sensor, and an ultraviolet light sensor, used to acquire multimodal feature maps of both sides of the currency.

[0057] The image processing module includes an image preprocessing unit and a feature extraction unit. The image preprocessing unit performs unified preprocessing on the received multimodal feature maps. The feature extraction unit collects global features of the image through the backbone network of the image processing algorithm and collects local features of the image through the texture perception module of the image processing algorithm.

[0058] Furthermore, the backbone network of the image processing module adopts a 10-channel recognition neural network, and the received image features include visible light RGB three-channel image features, infrared light image features, and ultraviolet light image features corresponding to the front and back of the currency, totaling 10 channels of image feature information.

[0059] Specifically, in multimodal feature fusion, after the currency passes through multimodal sensors (including visible light sensors, infrared light sensors, and ultraviolet light sensors), multimodal feature maps of the visible light, ultraviolet light, and infrared light regions can be obtained (currency exhibits fluorescence in response to ultraviolet and infrared light, therefore, the multimodal features of ultraviolet and infrared light possess a certain ability to distinguish between genuine and counterfeit currency). The multimodal feature maps are fused using a feature cascading (splicing) method, such as... Figure 3 A schematic diagram of multimodal feature fusion.

[0060] Specifically, in currency recognition, the fusion of multimodal feature maps from ultraviolet, visible, and infrared light can generate a 10-channel multimodal feature map (five channels each for the front and back sides, including a visible light RGB three-channel image, one infrared image, and one ultraviolet image). After the multimodal feature map is input into the network, depth features are extracted using common convolution methods. Simultaneously, the convolutional kernels extract features from different channels within the same receptive field region through weight sharing. Subsequently, feature integration is performed, mapping the ultraviolet and infrared multimodal features to the depth features. The introduction of ultraviolet and infrared multimodal feature maps can improve the algorithm's ability to distinguish genuine from counterfeit currency.

[0061] Furthermore, to improve the performance of multi-currency recognition algorithms, this application designs a texture-aware recognition algorithm, specifically by embedding the aforementioned texture-aware module. In everyday life, conventional recognition algorithms often employ deep neural networks with a large number of parameters, such as VGG and ResNet, to achieve higher recognition rates. However, Convolutional Neural Networks (CNNs) typically acquire global features, thus lacking the ability to mine local features. Currency often contains a large number of texture features (a type of local feature). Therefore, this application designs a texture-aware module to mine the texture features of currencies, thereby improving the performance of multi-currency recognition. See the texture-aware algorithm process diagram below. Figure 4 .

[0062] Furthermore, regarding the texture perception module, in the human visual system, a set of receptive fields of different sizes helps to highlight the region near the fovea of ​​the retina. This region is highly sensitive to subtle spatial changes (such as texture changes), which is beneficial for target recognition. Therefore, this application designs a texture perception module by simulating human visual perception through convolution with different kernel sizes. Simultaneously, since high-dimensional mapping is not conducive to feature representation, and although there are subtle differences in the features carried by different channels within the same feature layer, the statistical properties between multi-channel features are consistent. Therefore, to alleviate the feature loss problem during the transition from high-dimensional features to low-dimensional features, this paper proposes a texture perception module based on channel segmentation. The texture perception module is structured as follows:

[0063] f1(x, n) = BConv size＝(1，2n-1) (x) (1)

[0064] f2(x, n) = BConv size＝(2n-1，1) (x) (2)

[0065] f3(x, n) = BConv dilate＝2n-1 (x) (3)

[0066] In the formula, n∈[1,2,3,4] represents the branch number, formula (1) represents the convolution operation and normalization operation with a kernel size of (1,2n-1), formula (2) represents the convolution operation and normalization operation with a kernel size of (2n-1,1), and formula (3) represents the convolution operation and normalization operation with a kernel size of (3,3) and an inflation rate of 2n-1.

[0067] The input feature X can be segmented into x. n It can be expressed by formula (4), while the branch network in the texture-aware module can be expressed by formula (5):

[0068] x1, x2, x3, x4=Splited(X) (4)

[0069]

[0070] In addition, to optimize the texture perception module, a residual mechanism can be introduced. Overall, the texture perception module can be represented as:

[0071] RF = f1(X, 1) + Cat n＝1，2，3，4 F n (6)

[0072] The texture-aware module corresponding to the above algorithm process can also be found in [reference needed]. Figure 5 .

[0073] Furthermore, the algorithm model parameters corresponding to the image processing module include the number of iterations, learning rate, loss function, and training optimization function. The loss function adopts the Triplet loss function, the training optimization function adopts the SGD method, and the parameters inside the model are updated through backpropagation.

[0074] The currency recognition module includes a storage unit and a comparison and determination unit. The storage unit stores a template library and a test library. The comparison and determination unit compares the received feature information with the currency features in the template library to determine the classification and denomination of the currency to be tested.

[0075] Furthermore, the currency recognition module employs an open set recognition algorithm comprising a cosine similarity calculation unit, a comparison unit, and a threshold determination unit. The cosine similarity calculation unit is used to calculate and sort the cosine similarity of the features of the currency to be tested. The comparison unit is used to extract the highest cosine similarity score and match it with the template label in the template library. The threshold determination unit is used to set a threshold and extract valid matching results.

[0076] Specifically, the currency recognition module adopts an open-set recognition solution. Traditional banknote recognition methods often use closed-set recognition, where the currency ID and denomination are known and used during training. However, this closed-set approach cannot handle new currency IDs and denominations. Therefore, this paper adopts an open-set recognition method, allowing for unseen IDs or denominations of the currency to be recognized. Specifically, during the recognition process, the trained neural network algorithm serves only as a feature extraction network and does not participate in classification. See [link to relevant documentation]. Figure 6 Each currency in the test set is compared with the currencies in the template library, and the template label in the template library corresponding to the highest score in the comparison is the prediction result of the currency to be identified.

[0077] The result output module visualizes the recognition results of the currency recognition module through the display unit.

[0078] Specifically, the display unit uses displays such as LCD screens or LED screens to receive and display the prediction results from the currency recognition module.

[0079] Second Embodiment

[0080] A currency recognition method based on the system described in the first embodiment, the method comprising:

[0081] S1, Data Acquisition

[0082] The dataset of currencies with different denominations and values, including both genuine and counterfeit currencies, is collected using multimodal sensors. Therefore, each currency corresponds to a multimodal image under visible light, ultraviolet light, and red light.

[0083] S2, Dataset Labeling

[0084] The currency images are labeled with the currency ID, face value, and authenticity, so that each currency image contains three tags: currency ID, face value, and authenticity.

[0085] S3, Model Training

[0086] A currency recognition model with 10 input channels and a texture-aware module is trained by inputting data, building the model, and setting model parameters.

[0087] S31, Data Input

[0088] During the image data input process, the network input needs to be adjusted first. As mentioned earlier, this paper uses multimodal image input. Therefore, when inputting images, not only are currency images in the visible light region input into the network, but also currency images in the infrared and ultraviolet light regions are used as input. That is, when inputting, the RGB three channels of the currency image are combined with one channel of the infrared light image and one channel of the ultraviolet light image (a total of 10 channels of image features are included when both sides of the currency are included) and then input into the network.

[0089] S32, Model Construction

[0090] The network model mainly consists of two parts: the backbone network and the texture-aware module. Of course, image preprocessing is also included before feature extraction.

[0091] (a) Backbone network: This paper uses commonly used recognition networks as the backbone network (such as VGG, ResNet, etc.) for initial feature extraction. During the training process, a pre-trained model trained on the ImageNet dataset is loaded. In this paper, the input channels of the network need to be adjusted and set to 10.

[0092] (b) Texture perception module: To further improve the model's recognition rate, this paper adopts a texture perception module, which assists in currency recognition by obtaining texture features in the currency. The specific design process of the module is described in equations (1)-(6) above. Figure 5 .

[0093] S33. Setting and updating model parameters

[0094] The model parameters include: number of iterations, learning rate, loss function, and training optimization function. In one example, the number of update iterations is 100, the learning rate is set to 1e-3, and the learning rate is planned to be multiplied by 0.1 at iterations of 80 and 90 to facilitate model convergence; the triplet loss function is used as the loss function; and SGD (stochastic gradient descent) is used as the training optimization function. The internal parameters of the model are updated through backpropagation, and the model is saved every 20 iterations.

[0095] S4. Model testing: The model is tested and evaluated through data partitioning, feature extraction, and feature comparison.

[0096] (a) Data Partitioning: Test data can be currencies with known authenticity and denominations from the training set, or currencies with known authenticity and denominations not from the training set (open set recognition method). The dataset is divided into two parts: a test set and a template set. Each image in both the test set and the template set should have a correct label.

[0097] (b) Feature extraction: The test image and template first need to be processed by a neural network to extract features. The specific process of feature extraction is as follows: after the deep features are processed by the texture perception module, they are expanded into 1-dimensional features, and then normalized by the 1D BatchNormalization operation.

[0098] (c) Feature Comparison: The features of the test image and the features in the template library are compared using cosine similarity to obtain a score. The scores are then ranked, and the template label in the template library corresponding to the highest score is the predicted currency. Simultaneously, to obtain better results during application, a threshold can be set, such as considering a comparison correct only when the comparison score is greater than 0.9. Furthermore, the algorithm's performance can be evaluated by calculating metrics such as EER, FAR, and TAR. Once all algorithm performance metrics meet the requirements, the algorithm is applied.

[0099] S5. Model Application: Data is collected on the currency to be tested, and the collected multimodal features are imported into the currency recognition model. The recognition results are then visualized. Specifically, model application includes the following steps:

[0100] S51. Embed the currency recognition model with algorithm, algorithm parameters, and model weights into the image processing device;

[0101] S52. Establish a template library containing a large number of currency IDs and different denominations, including genuine and counterfeit images. Input the images in the template library into the processing device and extract the features of the images in the template library, and store the features and their labels.

[0102] S53. The currency to be detected is subjected to multimodal sensor to collect multimodal features, and the multimodal features are input into the algorithm to obtain identification features;

[0103] S54. The identification feature is compared with the feature in the template library by score comparison. The template label in the template library corresponding to the highest score is the prediction result of the test currency.

[0104] S55. When new currencies or new denominations of currency appear, collect a small number of currency samples, label the currencies, and add the features of the new currencies to the template library.

[0105] S56. Repeat steps S53-S55 to update and manage the currency recognition module.

[0106] The currency in the above proposal is more suited to banknotes.

[0107] Third Embodiment

[0108] The present invention also provides a computer-readable storage medium storing computer instructions thereon, which, when executed, perform the steps of the aforementioned method. The method is described in detail in the foregoing section and will not be repeated here.

[0109] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. This program can be stored in a computer-readable storage medium, including permanent and non-permanent, removable and non-removable media. Information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0110] The computer program code required for the operation of each part of this application can be written in any one or more programming languages, including object-oriented programming languages ​​such as Java, Scala, Smalltalk, Eiffel, JADE, Emerald, C++, C#, VB.NET, Python, etc.; conventional procedural programming languages ​​such as C, Visual Basic, Fortran2003, Perl, COBOL2002, PHP, ABAP; dynamic programming languages ​​such as Python, Ruby, and Groovy; or other programming languages. This program code can run entirely on the user's computer, or as a standalone software package on the user's computer, or partially on the user's computer and partially on a remote computer, or entirely on a remote computer or processing device. In the latter case, the remote computer can be connected to the user's computer through any network, such as a local area network (LAN) or wide area network (WAN), or connected to an external computer (e.g., via the Internet), or in a cloud computing environment, or used as a service such as Software as a Service (SaaS).

[0111] Fourth embodiment

[0112] The present invention also provides a terminal, including a memory and a processor. The memory stores a template library and computer instructions executable on the processor. When the processor executes the computer instructions, it performs the steps of the aforementioned method. The method is described in detail in the foregoing section and will not be repeated here.

[0113] In summary, this application adopts a multimodal feature fusion approach, which integrates image features from different modalities to improve the recognition and representation capabilities of deep features. Secondly, this paper proposes a multi-currency recognition algorithm based on texture perception, which extracts a large amount of texture information from banknotes through a texture perception module to compensate for the local features lost by the convolutional neural network when extracting deep features, thereby improving recognition performance. Finally, this paper adopts an open-set recognition mode, which can achieve the recognition of new currencies and new denominations with a small number of samples.

[0114] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multimodal currency recognition system based on texture perception and open set recognition, characterized in that: The system includes an image acquisition module, an image processing module, a currency recognition module, and a result output module connected via telecommunications. The image acquisition module employs a multimodal sensor to generate a multimodal feature map of the currency to be identified. The image processing module includes an image preprocessing unit and a feature extraction unit. The image preprocessing unit performs unified preprocessing on the received multimodal feature maps. The feature extraction unit collects global features of the image through the backbone network of the image processing algorithm and collects local features of the image through the texture perception module of the image processing algorithm. The currency recognition module includes a storage unit and a comparison and determination unit. The storage unit stores a template library and a test library. The comparison and determination unit compares the received feature information with the currency features in the template library to determine the classification and denomination of the currency to be tested. The currency recognition module adopts an open set recognition method, that is, the currency ID to be identified can include unseen IDs or denominations. The result output module visualizes the recognition results of the currency recognition module through the display unit; The multimodal sensor includes a visible light sensor, an infrared light sensor, and an ultraviolet light sensor, used to acquire multimodal feature maps of both sides of the coin; The backbone network of the image processing module adopts a 10-channel recognition neural network, and the received image features include visible light RGB three-channel image features, infrared light image features, and ultraviolet light image features corresponding to the front and back of the currency, totaling 10 channels of image feature information; The texture perception module of the image processing module is configured as follows: f1(x, n) = B Conv size=(1,2n-1) (x)(1) f2(x, n) = B Conv size=(2n-1,1) (x)(2) f3(x, n) = B Conv dilate=2n-1 (x) (3) x1, x2, x3, x4 = Splited(X)(4) (5) RF=f1(X, 1) +Cat n=1,2,3,4 F n (6) In the formula, n ∈ [1,2,3,4] represents the branch number. Formula (1) represents a convolution kernel size of (1, 2n) 1) Convolution and normalization operations; Formula (2) indicates that the convolution kernel size is (2n) Convolution operations and normalization operations of 1,1); Formula (3) represents the convolution operation with kernel size (3,3) and dilation rate of 2n-1, as well as the normalization operation; Formula (4) indicates that the input feature X is segmented into ; Formula (5) represents the branch network in the texture-aware module; Formula (6) represents the texture-aware module that introduces a residual mechanism; The currency recognition module employs an open-set recognition algorithm comprising a cosine similarity calculation unit, a comparison unit, and a threshold determination unit. The cosine similarity calculation unit is used to calculate and sort the cosine similarity of the features of the currency to be tested. The comparison unit is used to extract the highest cosine similarity score and match it with the template label in the template library. The threshold determination unit is used to set a threshold and extract valid matching results. The currency recognition module adopts open set recognition, where the currency ID to be recognized is allowed to contain unseen IDs or denominations; during the recognition process, the trained neural network algorithm is only used as a feature extraction network and does not participate in classification recognition; each currency in the test set is compared with the currencies in the template library, and the template label in the template library corresponding to the highest score in the comparison is the prediction result of the currency to be recognized.

2. The system according to claim 1, characterized in that: The algorithm model parameters corresponding to the image processing module include the number of iterations, learning rate, loss function, and training optimization function. The loss function adopts the Triplet loss function, the training optimization function adopts the SGD method, and the parameters inside the model are updated through backpropagation.

3. A currency identification method based on the system described in claim 1 or 2, characterized in that, The methods include: S1. Data acquisition: Collect a dataset of currencies with different denominations and values, including both genuine and counterfeit currencies, using a multimodal sensor. Therefore, each currency corresponds to a multimodal image under visible light, ultraviolet light, and red light. S2. Dataset labeling: Label the currency ID, face value, and authenticity of the collected currency images, so that each currency image contains three labels: currency ID, face value, and authenticity. S3. Model training: A currency recognition model with 10 input channels and a texture perception module is trained by data input, model construction and model parameter setting. S4. Model testing: The model is tested and evaluated through data partitioning, feature extraction, and feature comparison. S5. Model Application: Data is collected on the currency to be tested, and the collected multimodal features are imported into the currency recognition model. The recognition results are then visualized.

4. The method according to claim 3, characterized in that, Model applications include: S51. Embed the currency recognition model with algorithm, algorithm parameters, and model weights into the image processing device; S52. Establish a template library containing a large number of currency IDs and different denominations, including genuine and counterfeit images. Input the images in the template library into the processing device and extract the features of the images in the template library, and store the features and their labels. S53. The currency to be detected is subjected to multimodal sensor to collect multimodal features, and the multimodal features are input into the algorithm to obtain identification features; S54. The identification feature is compared with the feature in the template library by score comparison. The template label in the template library corresponding to the highest score is the prediction result of the test currency. S55. When new currencies or new denominations of currency appear, collect a small number of currency samples, label the currencies, and add the features of the new currencies to the template library. S56. Repeat steps S53-S55 to update and manage the currency recognition module.

5. A computer-readable storage medium storing computer instructions thereon, characterized in that: The computer instructions, when executed, perform the steps of the method described in claim 3 or 4.

6. A terminal, comprising a memory and a processor, wherein the memory stores a template library and computer instructions executable on the processor, characterized in that: When the processor executes the computer instructions, it performs the steps of the method of claim 3 or 4.

Images