Multi-region double-stage RMB true and false identification method and electronic equipment

Abstract

The invention discloses a multi-region double-stage RMB true and false identification method and electronic equipment, and the method comprises the steps: obtaining the image information of a ticket sample, and carrying out the correction and registration of the image information, and obtaining the image data of the ticket sample; extracting a characteristic value of at least one anti-counterfeiting identification area in the image data and judging whether the characteristic value is normal or not; if the characteristic value is abnormal, inputting the characteristic value of the anti-counterfeiting identification area and the image of the functional area of the ticket sample into a classification network model for training to obtain a classification identification result of the ticket sample, a serial number identification result of the ticket sample and a version identification result of the ticket sample; and analyzing the version identification result of the ticket sample, the serial number identification result of the ticket sample and the classification identification result of the ticket sample, and judging whether the ticket sample is true or false. Through the technical scheme of the invention, the accuracy of algorithm authenticity judgment can be effectively improved, the feature analysis result and the network model design are combined, the banknote authenticity identification algorithm is realized, and the accuracy of identification is further improved.

Description

Technical Field

[0001] The invention relates to the technical field of ticket sample identification, and in particular to a multi-region dual-stage RMB authenticity identification method and electronic equipment. Background Art

[0002] Banknote authenticity identification technology is a very important technology with a wide range of applications, including banks, shopping malls, supermarkets, ATMs, etc. With the development of science and technology, banknote anti-counterfeiting technology is also constantly being updated, from the initial simple optical feature recognition to the current deep learning technology, which continuously improves the anti-counterfeiting ability of banknotes.

[0003] In recent years, with the rapid development of deep learning technology, more and more banknote anti-counterfeiting technologies have begun to use deep learning technology for anti-counterfeiting detection. For example, "A Banknote Anti-Counterfeiting Detection Method Based on Deep Learning" released on October 28, 2022 is a banknote anti-counterfeiting technology based on deep learning. This technology uses a deep learning method to normalize the collected banknote images, and intercepts multiple feature areas on the normalized banknote images, and then divides the data set into a training set and a test set according to a preset ratio, and inputs the training set into the feature extraction network to obtain a deep feature map, and inputs the deep feature map into the classification model for classification training to obtain an initial detection model, and finally obtain an anti-counterfeiting detection model.

[0004] The inventors found that existing anti-counterfeiting detection technologies require a large amount of data sets and computing resources while improving accuracy and efficiency, and model training and optimization also require a lot of time and effort. Therefore, existing detection technologies cannot meet actual production needs. Summary of the invention

[0005] The embodiments of the present invention aim to solve at least one of the above technical problems.

[0006] In the first aspect, an embodiment of the present invention provides a multi-region dual-stage method for distinguishing the authenticity of RMB, comprising: acquiring image information of the ticket sample and correcting and aligning the image information to obtain image data of the ticket sample; extracting feature values ​​of at least one anti-counterfeiting identification area in the image data and judging whether the feature values ​​are normal; if the feature values ​​are abnormal, inputting the feature values ​​of the anti-counterfeiting identification area and the image of the functional area of ​​the ticket sample into a classification network model training to obtain a classification recognition result of the ticket sample and a serial number recognition result and a version recognition result of the ticket sample; analyzing the version recognition result of the ticket sample, the serial number recognition result of the ticket sample and the classification recognition result of the ticket sample, and judging the authenticity of the ticket sample.

[0007] In a second aspect, an embodiment of the present invention provides an electronic device comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute any one of the multi-region two-stage RMB authenticity identification methods described above in the present invention.

[0008] In a third aspect, an embodiment of the present invention provides a storage medium, in which one or more programs including execution instructions are stored, and the execution instructions can be read and executed by an electronic device (including but not limited to a computer, a server, or a network device, etc.) to execute any of the above-mentioned multi-region two-stage RMB authenticity identification methods of the present invention.

[0009] In a fourth aspect, an embodiment of the present invention further provides a computer program product, comprising a computer program stored on a storage medium, wherein the computer program comprises program instructions, and when the program instructions are executed by a computer, the computer executes any one of the above-mentioned multi-region two-stage RMB authenticity identification methods.

[0010] The method of the embodiment of the present invention obtains a preliminary regional authenticity determination result by performing feature analysis on the extracted ticket sample image, and then performs secondary confirmation judgment through a deep learning model, which can effectively improve the accuracy of the algorithm's authenticity determination. The feature analysis results are combined with the network model design to achieve the specific implementation of the banknote authenticity identification algorithm, further improving the accuracy of multi-region dual-stage RMB authenticity identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying any creative work.

[0012] Figure 1 A flow chart of an embodiment of a multi-region dual-stage RMB authenticity identification method of the present invention; Figure 2 A flow chart of another embodiment of the multi-region dual-stage RMB authenticity identification method of the present invention; Figure 3 A schematic diagram of the classification network model structure of the multi-region dual-stage RMB authenticity identification method of the present invention; Figure 4 It is a schematic diagram of the local convolutional layer structure of the classification network model of the multi-region dual-stage RMB authenticity identification method of the present invention; Figure 5 A schematic diagram of the process of a multi-region dual-stage RMB authenticity identification method provided by an embodiment of the present invention; Figure 6 The present invention is a block diagram showing the structure of an electronic device according to an embodiment of the present invention.

[0013] in, Figure 6 The corresponding relationship between the reference numerals and component names in the figure is: 610: processor; 620: memory; 630: input device; 640: output device. DETAILED DESCRIPTION

[0014] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0015] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application may be combined with each other.

[0016] The present invention may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present invention may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communications network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0017] In the present invention, "module", "device", "system" and the like refer to related entities applied to computers, such as hardware, a combination of hardware and software, software or software in execution, etc. In detail, for example, an element can be, but is not limited to, a process, a processor, an object, an executable element, an execution thread, a program and / or a computer running on a processor. In addition, an application program or a script program, a server running on a server can be an element. One or more elements can be in an execution process and / or thread, and an element can be localized on a computer and / or distributed between two or more computers, and can be operated by various computer-readable media. An element can also communicate through a local and / or remote process according to a signal with one or more data packets, for example, a signal from a data that interacts with another element in a local system, a distributed system, and / or a network on the Internet through a signal and interacts with other systems.

[0018] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include" and "comprise" include not only those elements, but also other elements that are not explicitly listed, or also include elements inherent to such processes, methods, articles or equipment. In the absence of further restrictions, the elements defined by the statement "include..." do not exclude the existence of other identical elements in the process, method, article or equipment that includes the elements.

[0019] The embodiment of the present invention provides a multi-region dual-stage RMB authenticity identification method, which can be applied to electronic devices. The electronic devices can be computers, servers or other electronic products, etc., and the present invention does not limit this.

[0020] Please refer to Figure 1 , which shows a multi-region dual-stage RMB authenticity identification method provided by an embodiment of the present invention.

[0021] like Figure 1 As shown, in step 101, the image information of the ticket sample is obtained and the image information is corrected and registered to obtain the image data of the ticket sample; In step 102, a feature value of at least one anti-counterfeiting identification area in the image data is extracted and it is determined whether the feature value is normal; In step 103, if the feature value is abnormal, the feature value of the anti-counterfeiting identification area and the image of the functional area of ​​the ticket sample are input into the classification network model training to obtain the classification recognition result of the ticket sample and the serial number recognition result and version recognition result of the ticket sample; In step 104, the version recognition result of the ticket sample, the serial number recognition result of the ticket sample and the classification recognition result of the ticket sample are analyzed to determine the authenticity of the ticket sample.

[0022] In this embodiment, for step 101, various types of information of the ticket sample are obtained by using various types of camera devices and sensor devices, wherein the ticket sample is not limited to different types of banknotes, card varieties or other bills with a certain value. In the process of collecting ticket samples, all versions and denominations of banknotes are covered as much as possible. The various types of information of the ticket sample can be the front view, back view, perspective view, test image, fluorescent image, thickness information, magnetic signal and electrical signal of the ticket sample. The various types of image information of the ticket sample are corrected and registered to obtain the image data of the ticket sample, and the image data is a unified viewpoint to ensure the consistency and comparability of the data. In order to improve the accuracy and robustness of the banknote authenticity identification algorithm, a large amount of banknote image data needs to be collected. Among them, the front view, back view, perspective view, test image, fluorescent image, thickness information, magnetic signal and electrical signal are all very important data. These data sources can include scenes such as banknote printing factories, printing technology research institutes, various self-service banks, ATM machines, supermarket cash registers, etc.

[0023] Afterwards, for step 102, after obtaining the image data of the ticket sample, extract the characteristic value of at least one anti-counterfeiting identification area in the image data and judge whether the characteristic value is normal. Each ticket sample contains multiple anti-counterfeiting identification areas, and only the characteristic value of one anti-counterfeiting identification area needs to be obtained for judgment, and the characteristic values ​​of each different anti-counterfeiting identification area are different. After extracting the characteristic value of an anti-counterfeiting identification area, perform characteristic analysis on the characteristic value to obtain a preliminary result of the authenticity judgment of the anti-counterfeiting area. This judgment result is not necessarily normal or abnormal, but only a preliminary judgment result. The characteristic analysis of the anti-counterfeiting identification area is performed through the mean, variance, the distance between the sample histogram and the true coin histogram, and the cosine distance between the sample and the existing true coin sample, and the authenticity of the area is preliminarily judged in combination with the existing threshold and the positive sample characteristic value range. Extracting the characteristic value of at least one anti-counterfeiting identification area in the image data is to extract the ROI area for different anti-counterfeiting identification areas according to the characteristics of the ticket sample (banknote). At the same time, ROI area extraction is performed on some specific functional areas (such as version, serial number, etc.) to obtain the corresponding area image. After obtaining the corresponding regional image, the image data needs to be preprocessed, and the preprocessed data needs to be normalized, Gaussian smoothed and denoised to improve the accuracy and efficiency of subsequent algorithms.

[0024] Then, for step 103, when it is determined that the characteristic value of the anti-counterfeiting identification area is abnormal, it is necessary to input the characteristic value of the anti-counterfeiting identification area and the image of the functional area of ​​the ticket sample into the classification network model training. The functional area is input into PaddleOCR and the classification network for training, and the classification network adopts the cross entropy loss function. The trained PaddleOCR network can recognize the serial number in the functional area, and the trained classification network can recognize the version of the ticket sample. After inputting the characteristic value of the anti-counterfeiting identification area and the image of the functional area of ​​the ticket sample into the classification network model training, the serial number recognition result and version recognition result of the ticket sample can be obtained.

[0025] Finally, for step 104, the version recognition results of the ticket sample, the serial number recognition results of the ticket sample, and the classification recognition results of the ticket sample are analyzed, and the authenticity of the ticket sample is determined. For example, multiple regions of the banknote are detected and the results are analyzed. If there is an area in all the regions that is determined to be a fake sample, the sample banknote is identified as a counterfeit banknote at this stage, and the second authenticity judgment result of the sample banknote is obtained. According to the authenticity identification result output by the network model, combined with the preliminary authenticity judgment result obtained by analyzing the first stage features (the feature value of the anti-counterfeiting identification area), the version recognition result and the serial number recognition result, the authenticity of the final result of the banknote is determined.

[0026] The method of the embodiment of the present application performs feature analysis on the extracted ticket sample images to obtain a preliminary regional authenticity determination result, and then performs secondary confirmation and judgment through a deep learning model, which can effectively improve the accuracy of the algorithm's authenticity determination. The feature analysis results are combined with the network model design to achieve the specific implementation of the banknote authenticity identification algorithm, further improving the accuracy of multi-region dual-stage RMB authenticity identification.

[0027] Please refer to Figure 2 , which shows another multi-region dual-stage RMB authenticity identification method provided by an embodiment of the present invention. Figure 1 A flowchart of the steps further defined in step 101 of "correcting and registering the image information".

[0028] like Figure 2 As shown, in step 201, the image of the ticket sample is subjected to image geometric correction using a calibration dot matrix and a perspective calibration method, wherein the image geometric correction can make the angle information, size information, position information and illumination information of the ticket sample image consistent; In step 202, when there is uneven grayscale distribution in the image of the ticket sample, bright field correction is performed on the image of the ticket sample; In step 203, according to a registration algorithm for feature point matching, feature points in the image of the ticket sample are extracted and matched with feature points in a preset ticket sample image.

[0029] In this embodiment, for step 201, the calibration dot matrix and perspective calibration method are used to perform geometric correction on the image to eliminate perspective distortion, radial distortion, and tangential distortion, etc., generated by the camera during the process of photographing banknotes. Since there may be problems such as angle deviation and distortion during the acquisition process, the image needs to be corrected so that the angle, size, position, lighting, and other information of the banknote image are consistent. The geometric correction mainly includes 1. Determine the area to be corrected in the image: It is usually necessary to select the area to be corrected in the image, such as text, table, or image. Selecting the correct area is crucial to the effect of perspective correction. 2. Extract the quadrilateral in the area: The perspective correction algorithm needs to extract the quadrilateral in the area to be corrected and then convert it into a rectangle. Edge detection and corner point detection can be used to extract quadrilaterals. 3. Calculate the perspective transformation matrix: The perspective correction algorithm corrects the image by calculating the perspective transformation matrix. The perspective transformation matrix can be obtained by selecting the coordinates of four vertices and the matrix of the target plane. Apply the perspective transformation matrix: The quadrilateral can be converted into a rectangle by applying the perspective transformation matrix to the image. In this process, interpolation technology can be used to fill in pixel values. 4. Check the corrected image: Finally, check whether the corrected image meets the expected effect. If the image is still distorted or deformed, you can optimize it by adjusting the four vertex coordinates and the perspective transformation matrix.

[0030] Then, for step 202, when there is uneven grayscale distribution in the image of the ticket sample, it is necessary to perform bright field correction on the image of the ticket sample. Uneven grayscale distribution refers to the situation where the grayscale in the image is not evenly distributed when the light source cannot provide uniform illumination, the camera lens has vignetting, the camera sensor has impurities, or the surface of the measured target is nonlinear. For such distortion, it is necessary to perform bright field correction on the image, and the process is calculated based on the following formula:

[0031] Among them, Median is an operator for finding the median of the difference between the bright field and the dark field, and the correction coefficient is used to correct the grayscale of the corrected image.

[0032] Finally, for step 203, according to the registration algorithm of feature point matching, feature points in the image of the ticket sample are extracted and matched with feature points in the preset ticket sample image. This process is also called banknote image registration. The banknote registration step is one of the important links in data processing. It can make the position, size, rotation and other information of the banknote image consistent, so as to facilitate subsequent processing and analysis.

[0033] The method of the embodiment of the present application uses a registration algorithm based on feature point matching to achieve image registration by extracting feature points in an image and matching them with feature points in a reference image.

[0034] It should be noted that the registration algorithm for feature point matching provided in this application can be divided into the following steps: a) Extract feature points: Extract feature points from the image by using feature point detection algorithms such as SIFT (Scale-invariant feature transform), SURF (Speeded Up Robust Features), and ORB (Oriented FAST and Rotated BRIEF).

[0035] b) Feature point matching: Match the feature points in the image to be registered with the feature points in the reference image to obtain a set of feature point pairs.

[0036] c) Calculate the transformation matrix: By using algorithms such as RANSAC (Random Sample Consensus), calculate the transformation matrix between feature point pairs, such as the affine transformation matrix and perspective transformation matrix.

[0037] d) Applying the transformation matrix: transforming the image to be registered according to the calculated transformation matrix so that the image to be registered is aligned with the reference image as closely as possible.

[0038] e) Check the registration results: Finally, check whether the registration results meet the expected results. If the registration is not ideal, you can optimize it by adjusting parameters or reselecting reference images.

[0039] In some optional embodiments, extracting the feature value of at least one anti-counterfeiting identification area in the image data and judging whether the feature value is normal is to perform feature analysis on the anti-counterfeiting identification area of ​​the ticket sample through the mean, variance, the distance between the ticket sample histogram and the preset ticket sample histogram, and the cosine distance between the ticket sample and the preset ticket sample, and judge the authenticity of the anti-counterfeiting identification area of ​​the ticket sample. Combined with the existing threshold and the range of positive sample feature values, a preliminary judgment is made on the authenticity of the area.

[0040] In some optional embodiments, different versions of ticket samples and images of different anti-counterfeiting identification areas are modeled separately and trained using a classification network model, wherein the classification network model is a partial convolution classification network model and uses a cross entropy loss function. The classification network model is a network model with an eigenvalue embedding layer. The core convolution layer of the network uses partial convolution, which has lower latency and higher throughput, can effectively reduce the amount of calculation and memory usage, and can also greatly improve the efficiency and accuracy of the algorithm. After the eigenvalues ​​obtained by inputting the image ROI area and the CPU (Central Processing Unit) stage on the GPU (graphics processing unit) device, the network model can obtain the authenticity identification result of this stage.

[0041] The model for further classification of counterfeit currency is the counterfeit currency classification model. The specific implementation path is as follows: The first step is to train a classification network using known counterfeit banknote types; The second step is to extract the fc layer of the classification network as the feature vector after feature dimensionality reduction; The third step is to cluster each type of feature vector and calculate the average cluster radius and cluster center of known types; The fourth step is to calculate the distance between the feature vector to be detected and each class; Step 5: If the distance between the feature vector of the feature to be detected and each category exceeds 5 times the cluster radius, the user is prompted whether to add a new category; Step 6: When the user determines that a new category needs to be added, a new category is added and the classification network is retrained; In the seventh step, when the user determines that no new categories need to be added, the features to be detected are added to the known categories and the classification network is retrained.

[0042] This method can effectively identify counterfeit currency by extracting feature vectors and performing cluster analysis, and can be dynamically updated as new types of counterfeit currency emerge, improving the accuracy and robustness of detection.

[0043] It should be noted that the method of the present application uses a version recognition model, a number recognition model, a feature positioning model and an anti-counterfeiting classification model. After a banknote to be detected passes through the machine, a set of images is collected. First, this set of images is sent to the version recognition model to identify the version of the altered banknote. After identifying the version, the feature detection model of the version is read. The feature detection model decomposes the set of images into several features, and identifies the type corresponding to each feature, and cuts out the feature small image from the set of images. If the cut feature image is a number feature, the feature image is sent to the number recognition model to identify the number of the altered banknote. If the cut feature is an anti-counterfeiting feature, the image is subjected to anti-counterfeiting recognition, and the feature is identified as a genuine banknote or a counterfeit banknote. If the feature is identified as a counterfeit banknote, the type of the altered banknote is further identified through the counterfeit banknote classification model.

[0044] In some optional embodiments, the anti-counterfeiting identification area is tested and the results are analyzed. If there is a sample determined to be a fake ticket in the anti-counterfeiting identification area, the sample is identified as a counterfeit currency at this stage, and the second authenticity determination result of the sample is obtained. The authenticity of the sample is determined based on the characteristic value of the anti-counterfeiting identification area of ​​the sample and the version recognition result and the serial number recognition result output by the classification network model. For example, multiple areas of a banknote are tested and the results are analyzed. If there is an area in all areas that is determined to be a fake sample, the sample banknote is identified as a counterfeit currency at this stage, and the second authenticity determination result of the sample banknote is obtained. According to the authenticity identification result output by the network model, combined with the preliminary authenticity determination result, version recognition result and serial number recognition result obtained by the first stage feature analysis, the authenticity of the final result of the banknote is determined.

[0045] It should be noted that the technical solution of the present application combines feature analysis and deep learning classification algorithms, which can effectively solve the problems of inaccurate identification results of a single deep learning algorithm or feature analysis algorithm in the banknote authenticity identification algorithm, insufficient robustness in multiple scenarios, and inaccurate identification of the overall authenticity of banknotes based on a single-region feature analysis algorithm. At the same time, the technical solution of the present application uses a variety of sensors and cameras to obtain various information about banknotes, and the model aggregates multi-modal inputs, which can effectively improve the accuracy and recall rate of banknote authenticity identification.

[0046] It should be noted that the technical solution of this application utilizes a variety of information such as the front view, back view, perspective view, test image, fluorescent image, thickness information, magnetic signal and electrical signal of the banknote to analyze and identify the banknote from multiple angles, thereby improving the accuracy and robustness of the algorithm. This application combines feature analysis and deep learning, which can not only utilize the knowledge and experience of human experts, but also give full play to the advantages of deep learning, thereby improving the accuracy and efficiency of the algorithm. This application also designs a network model based on local convolution, which has lower latency and higher throughput, and can greatly improve the efficiency and accuracy of the algorithm. The method involved in this application takes into account the real-time and scalability of the algorithm, and can be applied to various self-service banks, ATMs, supermarket cash registers and other scenarios to meet the needs of different scenarios.

[0047] It should be noted that in a large amount of experimental data, the multi-region dual-stage RMB authenticity identification method provided by this application has a high accuracy and recall rate, can effectively distinguish the authenticity of banknotes, and ensure the security and credibility of banknotes. At the same time, the algorithm also has good real-time and scalability, and can be applied to various self-service banks, ATM machines, supermarket cash registers and other scenarios.

[0048] Please refer to Figure 3 , Figure 4 , which shows a schematic diagram of the classification network model structure of a multi-region two-stage RMB authenticity identification method provided by an embodiment of the present invention.

[0049] The network structure diagram is as follows Figure 3 As shown in the figure, multiple P modules are used in the model to efficiently extract the features of the input image and reduce the feature resolution width and height to 1 / 32 of the input image. Then, global pooling and 1×1 convolution layers are designed to compress the features into flat one-dimensional features. In the P module, it is composed of a local convolution layer (PConv), a 1x1 convolution layer (Conv), a BatchNormal and ReLU activation layer, and a 1x1 convolution layer in series, and has a residual jump connection to prevent the model capability from being degraded due to an overly deep network structure, including c1×h / 4×w / 4, c2×h / 8×w / 8, c3×h / 16×w / 16, c4×h / 32×w / 32, where c, h, and w are the number of channels, height, and width. The P module can be seen in Figure 3 The local convolution layer (PConv) is as follows Figure 4 As shown, this convolution only uses C k The filter processes part of the features, and the remaining features will keep the same mapping. This design can greatly improve the operation speed of the model and also has the anti-degradation function similar to the jump connection. pis the number of partial channels, and k is the convolution kernel. The network model also inputs three eigenvalues, namely the mean, variance, and cosine distance between the input sample and the real currency sample, generated during the CPU processing stage. By designing a 1x1 convolution layer as the eigenvalue embedding layer, the eigenvalue is mapped to the network's feature space, and then all features are fused through the features generated by the cascaded image input, and finally the final true or false prediction result is generated through the fully connected layer.

[0050] Please refer to Figure 5 , which shows the process principle diagram of the multi-region two-stage RMB authenticity identification method of the present invention.

[0051] like Figure 5 As shown, first, various cameras and sensors are used to obtain various information of banknotes, such as the front view, back view, perspective view, test image, fluorescent image, thickness information, magnetic signal and electrical signal. Then, the image information is corrected and registered to obtain a unified viewpoint. Feature extraction and feature analysis are performed on different anti-counterfeiting identification areas on the CPU to obtain a preliminary regional authenticity determination result. At the same time, feature analysis is performed on some specific functional areas to obtain the version recognition result and the serial number recognition result of the banknote. In addition, the present invention designs a network model with an eigenvalue embedding layer, and the basic convolution layer of the network adopts partial convolution, which has lower latency and higher throughput. The network model can obtain the authenticity identification result of this stage on the GPU device by inputting the image ROI area and the eigenvalues ​​obtained in the previous stage. Finally, according to the prediction results combined with the two major stages, the authenticity of the final result of the banknote is determined.

[0052] Combine data processing and network model design to realize the specific implementation of the banknote authenticity identification algorithm. The specific steps are as follows: (1) Collect image data of genuine and fake banknotes through a variety of sensors.

[0053] (2) Preprocess the banknote image data, including image correction, image registration, anti-counterfeiting identification area extraction, functional area extraction, normalization, etc.

[0054] (3) Perform feature analysis on the anti-counterfeiting identification area through the mean, variance, the distance between the sample histogram and the genuine coin histogram, and the cosine distance between the sample and the existing genuine coin samples. Combined with the existing threshold and the range of positive sample feature values, a preliminary judgment is made on the authenticity of the area.

[0055] (4) The functional area is input into PaddleOCR and the classification network based on partial convolution proposed in this patent for training, and the classification network adopts the cross entropy loss function. The trained PaddleOCR network can recognize the serial number in the functional area, and the trained classification network can recognize the version of the sample banknote.

[0056] (5) Model the images of different versions and different ROI areas separately, and use the classification based on partial convolution proposed in this patent for training, and also use the cross entropy loss function. The classification network trained in this step can identify the authenticity of the sample banknote in a certain area.

[0057] (6) Multiple regions of the banknote are tested and the results are analyzed. If one region among all the regions is determined to be a fake sample, the sample banknote is identified as a fake banknote at this stage, and a second authenticity determination result of the sample banknote is obtained.

[0058] (7) Based on the authenticity identification results output by the network model, combined with the preliminary authenticity identification results, version identification results and serial number identification results obtained from the first stage feature analysis, the authenticity of the banknote is finally determined.

[0059] It should be noted that the multi-regional two-stage RMB authenticity identification method provided by the present application is based on a secondary judgment algorithm, and the extracted features are subjected to feature analysis to obtain a preliminary regional authenticity judgment result, and then a secondary confirmation judgment is performed through a deep learning model, which can effectively improve the accuracy of the algorithm authenticity judgment. The feature analysis results are combined with the network model design to realize the specific implementation of the banknote authenticity identification algorithm. At the same time, a network model based on local convolution of multimodal input is also provided, which can process multimodal input, i.e., eigenvalue input and image area input, and has lower latency and higher throughput, improving the efficiency and accuracy of the algorithm. The multi-regional two-stage RMB authenticity identification method provided by the present application effectively improves the accuracy of the algorithm authenticity judgment by identifying multiple anti-counterfeiting identification areas.

[0060] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of combined actions, but those skilled in the art should be aware that the present invention is not limited by the described order of actions, because according to the present invention, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention. In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0061] In some embodiments, an embodiment of the present invention provides a storage medium, which stores one or more programs including execution instructions, and the execution instructions can be read and executed by electronic devices (including but not limited to computers, servers, or network devices, etc.) to execute any of the above-mentioned multi-zone two-stage RMB authenticity identification methods of the present invention.

[0062] In some embodiments, an embodiment of the present invention also provides a computer program product, which includes a computer program stored on a storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer executes any one of the above-mentioned multi-zone two-stage RMB authenticity identification methods.

[0063] In some embodiments, an embodiment of the present invention also provides an electronic device, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute a multi-region two-stage RMB authenticity identification method.

[0064] Figure 6 FIG. 1 is a schematic diagram of the hardware structure of an electronic device for executing a multi-region dual-stage RMB authenticity identification method provided by another embodiment of the present application, such as Figure 6 As shown, the device includes: One or more processors 610 and memory 620, Figure 6 A processor 610 is taken as an example.

[0065] The device for executing the multi-region dual-stage RMB authenticity identification method may further include: an input device 630 and an output device 640 .

[0066] The processor 610, the memory 620, the input device 630 and the output device 640 may be connected via a bus or other means. Figure 6 The example of connecting through bus is taken in the following.

[0067] The memory 620, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as program instructions / modules corresponding to the multi-region dual-stage RMB authenticity identification method in the embodiment of the present application. The processor 610 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions and modules stored in the memory 620, that is, the multi-region dual-stage RMB authenticity identification method in the above method embodiment is implemented.

[0068] The memory 620 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the electronic device, etc. In addition, the memory 620 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 620 may optionally include a memory remotely arranged relative to the processor 610, and these remote memories may be connected to the electronic device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0069] The input device 630 can receive input digital or character information and generate signals related to user settings and function control of the electronic device. The output device 640 can include a display device such as a display screen.

[0070] The one or more modules are stored in the memory 620, and when executed by the one or more processors 610, the multi-region dual-stage RMB authenticity identification method in any of the above method embodiments is executed.

[0071] The above-mentioned product can execute the method provided in the embodiment of the present application, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not fully described in this embodiment, please refer to the method provided in the embodiment of the present application.

[0072] The electronic devices of the embodiments of the present application exist in various forms, including but not limited to: (1) Mobile communication equipment: This type of equipment is characterized by having mobile communication functions and its main purpose is to provide voice and data communications. This type of terminal includes: smart phones, multimedia phones, functional phones, and low-end phones.

[0073] (2) Ultra-mobile personal computer devices: These devices belong to the category of personal computers, have computing and processing functions, and generally also have mobile Internet access features. These terminals include: PDA (personal digital assistant), MID (Mobile Internet Device), and UMPC (Ultra-mobile Personal Computer) devices.

[0074] (3) Portable entertainment devices: These devices can display and play multimedia content. They include audio and video players, handheld game consoles, e-books, smart toys, and portable car navigation devices.

[0075] (4) Other onboard electronic devices with data interaction functions, such as on-board devices installed in vehicles.

[0076] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0077] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a general hardware platform, and of course, by hardware. Based on this understanding, the above technical solution, in essence, or the part that contributes to the relevant technology, can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiment.

[0078] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A multi-region dual-stage method for distinguishing the authenticity of RMB, characterized in that: include: Acquiring image information of the ticket sample and correcting and registering the image information to obtain image data of the ticket sample; Extracting a feature value of at least one anti-counterfeiting identification area in the image data and determining whether the feature value is normal; If the feature value is abnormal, the feature value of the anti-counterfeiting identification area and the image of the functional area of ​​the ticket sample are input into the classification network model training to obtain the classification recognition result of the ticket sample and the serial number recognition result and version recognition result of the ticket sample; The version recognition result of the ticket sample, the serial number recognition result of the ticket sample and the classification recognition result of the ticket sample are analyzed to determine the authenticity of the ticket sample.

2. The multi-region dual-stage RMB authenticity identification method according to claim 1 is characterized in that: The correcting and registering the image information comprises: Performing image geometric correction on the image of the ticket sample by using a calibration dot matrix and a perspective calibration method, wherein the image geometric correction can make the angle information, size information, position information and illumination information of the image of the ticket sample consistent; When there is uneven grayscale distribution in the image of the ticket sample, performing bright field correction on the image of the ticket sample; According to the registration algorithm of feature point matching, feature points in the image of the ticket sample are extracted and matched with feature points in the preset ticket sample image.

3. The multi-region dual-stage RMB authenticity identification method according to claim 2 is characterized in that: The extracting feature points from the ticket sample image and matching the feature points with feature points in a preset ticket sample image comprises: Extracting feature points from the image of the ticket sample using a feature point detection algorithm; Matching the feature points in the image of the ticket sample with the feature points in the preset ticket sample image to obtain at least one set of feature point pairs; Using the RANSAC algorithm, a transformation matrix between the feature point pairs is calculated, wherein the transformation matrix includes an affine transformation matrix and a perspective transformation matrix; The ticket sample image is transformed according to the transformation matrix to align the ticket sample image with the preset ticket sample image.

4. The multi-region dual-stage RMB authenticity identification method according to claim 3 is characterized in that: The feature point detection algorithms include SIFT, SURF and ORB.

5. The multi-region dual-stage RMB authenticity identification method according to claim 1 is characterized in that: The extracting the feature value of at least one anti-counterfeiting identification area in the image data and determining whether the feature value is normal includes: The anti-counterfeiting identification area of ​​the ticket sample is feature analyzed through the mean, variance, the distance between the ticket sample histogram and the preset ticket sample histogram, and the cosine distance between the ticket sample and the preset ticket sample, and the authenticity of the anti-counterfeiting identification area of ​​the ticket sample is determined.

6. The multi-region dual-stage RMB authenticity identification method according to claim 1 is characterized in that: The step of analyzing the version recognition result of the ticket sample, the serial number recognition result of the ticket sample, and the classification recognition result of the ticket sample, and determining whether the ticket sample is genuine includes: The anti-counterfeiting identification area is tested and the result is analyzed. If there is a counterfeit banknote sample in the anti-counterfeiting identification area, the banknote sample is identified as a counterfeit currency at this stage, and a second authenticity determination result of the banknote sample is obtained.

7. The multi-region dual-stage RMB authenticity identification method according to claim 6 is characterized in that: Before the anti-counterfeiting identification area is detected and the result is analyzed, it includes: Using the classification network to train known counterfeit ticket samples, extracting the fc layer of the classification network as the feature vector after feature dimension reduction, clustering each type of feature vector, and calculating the average cluster radius and cluster center of known types; The distance between the feature vector of the feature to be detected and each category is calculated. When the distance between the feature vector of the feature to be detected and each category exceeds the preset multiple of the clustering radius, the user is prompted whether to add a new category. If necessary, add a new category and retrain the classification network; If not necessary, add the features to be detected to the known categories and retrain the classification network.

8. The multi-region dual-stage RMB authenticity identification method according to claim 1 is characterized in that: The step of obtaining the image information of the ticket sample comprises: The image information of the ticket sample is obtained based on sensors and camera devices, wherein the image information of the ticket sample includes a front view, a back view, a perspective view, a test image, a fluorescent image, thickness information, a magnetic signal and an electrical signal of the ticket sample.

9. An electronic device, characterized in that: include: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform the steps of the multi-region dual-stage RMB authenticity identification method described in any one of claims 1 to 8.

10. A storage medium, characterized in that: A computer program is stored thereon, and when the program is executed by a processor, the steps of the multi-region dual-stage RMB authenticity identification method described in any one of claims 1 to 8 are implemented.

Images