Banknote anomaly detection method and system based on big data

Abstract

The present application provides a banknote anomaly detection method and system based on big data, which relates to the field of intelligent detection. It extracts hue features and content semantic features of optically variable ink areas by applying image processing and analysis algorithms based on artificial intelligence and deep learning technology, so as to depict the hue and content semantic feature information of the optically variable ink area, and detects whether the banknote is an abnormal banknote based on the significant sparse interactive fusion representation between the two hierarchical features of the optically variable ink area. In this way, it is possible to capture multiple features of the banknote more comprehensively, avoid the influence of ambient lighting conditions when performing banknote anomaly detection, reduce the influence of image noise and processing errors, thereby more comprehensively identifying abnormal banknotes and improving the accuracy of detecting whether the banknote to be detected is an abnormal banknote.

Description

Technical Field

[0001] The present application relates to the field of intelligent detection, and more particularly, in an embodiment of the present application, to a banknote anomaly detection method and system based on big data. Background Art

[0002] Optically variable ink is a special printing material whose color changes with the viewing angle. This characteristic is widely used in anti-counterfeiting measures. In the field of banknote anomaly detection, the existing optically variable ink anomaly detection method mainly determines whether it is abnormal by counting the edge pixel mean of the optically variable ink part, the mean and variance of all pixels in the optically variable ink part, and then calculating the ratio of the mean of the optically variable ink part to the mean of the edge intensity pixels and the variance of the optically variable ink part. However, this method is not effective for detecting color-changing optically variable ink counterfeit banknotes because it fails to solve the problem of abnormal color change in the optically variable ink area.

[0003] In response to the above technical problems, Chinese patent CN113256873A discloses a banknote abnormality detection method, device, electronic device and machine storage medium, which can detect whether a banknote is an abnormal banknote by acquiring an image of the banknote to be detected as the banknote image, then determining the optically variable ink area in the banknote image and extracting the color tone characteristics of the optically variable ink area, and then detecting whether the banknote to be detected is an abnormal banknote based on the color tone characteristics of the optically variable ink area.

[0004] In the above-mentioned banknote abnormality detection method, the main method is to detect the color change of the optically variable ink area of ​​the banknote and compare it with the preset standard to determine whether the banknote is abnormal. However, since the color change of the optically variable ink is greatly affected by light, the images collected under different lighting conditions may be different, thus affecting the accuracy of banknote abnormality detection. In addition, this banknote abnormality detection scheme has high requirements on the image quality, and any noise in the image processing or error in the binarization processing may affect the final banknote abnormality detection result.

[0005] Therefore, an optimized banknote anomaly detection solution is desired. Summary of the invention

[0006] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a banknote anomaly detection method and system based on big data, which extracts hue features and content semantic features of the optically variable ink area by applying image processing and analysis algorithms based on artificial intelligence and deep learning technology, so as to characterize the hue and content semantic feature information of the optically variable ink area, and detects whether the banknote is an abnormal banknote based on the significant sparse interactive fusion representation between the two hierarchical features of the optically variable ink area. In this way, multiple features of the banknote can be captured more comprehensively, and the influence of ambient lighting conditions can be avoided when detecting banknote anomalies, and the influence of image noise and processing errors can be reduced, so as to more comprehensively identify abnormal banknotes and improve the detection accuracy of whether the banknote to be detected is an abnormal banknote.

[0007] According to one aspect of the present application, a banknote anomaly detection method based on big data is provided, which includes: obtaining a banknote image of a banknote to be detected; determining an optically variable ink area of ​​the banknote image; analyzing the optically variable ink area to determine whether the banknote to be detected is an abnormal banknote, including: extracting hue features and content features of the optically variable ink area to obtain hue coding features of the optically variable ink area and content semantic coding features of the optically variable ink area; performing core-related feature joint interactive coding on the hue coding features of the optically variable ink area and the content semantic coding features of the optically variable ink area to obtain hue-content semantically significant sparse interactive fusion features of the optically variable ink area, including: extracting hue features of the optically variable ink area and content semantic coding features of the optically variable ink area; performing core-related feature joint interactive coding on the hue coding features of the optically variable ink area and the content semantic coding features of the optically variable ink area to obtain hue-content semantically significant sparse interactive fusion features of the optically variable ink area The content semantic coding features are globally associated and encoded to obtain a global coding representation between an optically variable ink region hue and a content semantic feature; the optically variable ink region hue coding features and the optically variable ink region content semantic coding features are interactively associated and interacted position by position to obtain an optically variable ink region hue-content semantic feature interactive fusion feature representation; based on the optically variable ink region hue-content semantic feature interactive fusion feature representation, the global coding representation between the optically variable ink region hue-content semantic features is significantly sparsely modulated to obtain the optically variable ink region hue-content semantic significantly sparse interactive fusion feature; based on the optically variable ink region hue-content semantic significantly sparse interactive fusion feature, it is determined whether the banknote to be detected is an abnormal banknote.

[0008] In the above-mentioned banknote anomaly detection method based on big data, the optically variable ink area is subjected to hue feature extraction and content feature extraction to obtain the optically variable ink area hue coding feature and the optically variable ink area content semantic coding feature, including: inputting the optically variable ink area into a hue feature extractor based on a hole convolutional neural network model to obtain an optically variable ink area hue coding feature map as the optically variable ink area hue coding feature; inputting the optically variable ink area into a content feature extractor based on a deep separable convolutional neural network model to obtain an optically variable ink area content semantic coding feature map as the optically variable ink area content semantic coding feature.

[0009] In the above-mentioned banknote anomaly detection method based on big data, the optically variable ink area hue coding feature and the optically variable ink area content semantic coding feature are globally associated and encoded to obtain a global coding representation between the optically variable ink area hue and content semantic features, including: inputting the optically variable ink area hue coding feature map and the optically variable ink area content semantic coding feature map into a kernel feature distillation network to obtain an optically variable ink area hue kernel feature vector and an optically variable ink area content semantic kernel feature vector; and globally associated encoding the optically variable ink area hue kernel feature vector and the optically variable ink area content semantic kernel feature vector to obtain a global coding matrix between the optically variable ink area hue and content semantic kernel features as the global coding representation between the optically variable ink area hue and content semantic features.

[0010] In the above-mentioned banknote anomaly detection method based on big data, the optically variable ink area tone coding feature map and the optically variable ink area content semantic coding feature map are input into the kernel feature distillation network to obtain the optically variable ink area tone kernel feature vector and the optically variable ink area content semantic kernel feature vector, including: performing feature decomposition on the optically variable ink area tone coding feature map to obtain a sequence of optically variable ink area tone decomposition feature vectors; calculating the difference features of each optically variable ink area tone decomposition feature vector in the sequence of the optically variable ink area tone decomposition feature vector relative to other optically variable ink area tone decomposition feature vectors to obtain an optically variable ink area tone decomposition feature difference representation vector; normalizing each eigenvalue in the optically variable ink area tone decomposition feature difference representation vector to obtain a normalized optically variable ink area tone decomposition feature difference representation vector; using each eigenvalue in the normalized optically variable ink area tone decomposition feature difference representation vector as a weight, calculating the weighted sum of the sequence of the optically variable ink area tone decomposition feature vectors to obtain the optically variable ink area tone kernel feature vector.

[0011] In the above-mentioned banknote anomaly detection method based on big data, calculating the difference characteristics of each optically variable ink area hue decomposition feature vector in the sequence of the optically variable ink area hue decomposition feature vector relative to other optically variable ink area hue decomposition feature vectors to obtain the optically variable ink area hue decomposition feature difference representation vector, including: calculating the norm of the position difference between each optically variable ink area hue decomposition feature vector in the sequence of the optically variable ink area hue decomposition feature vector and the other optically variable ink area hue decomposition feature vectors to obtain a sequence of optically variable ink area hue decomposition feature difference one-norm representation values; calculating the average value of the sequence of the optically variable ink area hue decomposition feature difference one-norm representation values ​​to obtain an optically variable ink area hue decomposition feature difference one-norm average representation value; vectorizing and arranging a plurality of the optically variable ink area hue decomposition feature difference one-norm average representation values ​​to obtain the optically variable ink area hue decomposition feature difference representation vector.

[0012] In the above-mentioned banknote anomaly detection method based on big data, each eigenvalue in the optically variable ink area hue decomposition feature difference representation vector is normalized to obtain a normalized optically variable ink area hue decomposition feature difference representation vector, including: calculating the position-based exponential function value of each eigenvalue in the optically variable ink area hue decomposition feature difference representation vector with a natural constant as the base to obtain an optically variable ink area hue decomposition feature difference class support representation vector; calculating the weighted sum of each eigenvalue in the optically variable ink area hue decomposition feature difference class support representation vector to obtain the optically variable ink area hue decomposition feature difference class support sum; calculating the position-based division between each eigenvalue in the optically variable ink area hue decomposition feature difference class support representation vector and the optically variable ink area hue decomposition feature difference class support sum to obtain the normalized optically variable ink area hue decomposition feature difference representation vector.

[0013] In the above-mentioned banknote anomaly detection method based on big data, the optically variable ink area hue coding feature and the optically variable ink area content semantic coding feature are subjected to position-by-position feature association interaction to obtain an optically variable ink area hue-content semantic feature position-by-position interactive fusion feature representation, including: inputting the global coding matrix between the optically variable ink area hue-content semantic kernel features into a sparse mask unit to obtain an optically variable ink area hue-content semantic kernel feature association sparse weight matrix; inputting the optically variable ink area hue coding feature map and the optically variable ink area content semantic coding feature map into a position-by-position association interaction network to obtain an optically variable ink area hue-content semantic position-by-position interactive fusion feature map as the optically variable ink area hue-content semantic feature position-by-position interactive fusion feature representation.

[0014] In the above-mentioned banknote anomaly detection method based on big data, based on the position-by-position interactive fusion feature representation of the optically variable ink area hue-content semantic features, the global coding representation between the optically variable ink area hue-content semantic features is subjected to feature significant sparse modulation to obtain the optically variable ink area hue-content semantic significant sparse interactive fusion feature, including: based on the associated sparse weight matrix between the optically variable ink area hue-content semantic kernel features, the feature significant sparse modulation is performed on the optically variable ink area hue-content semantic position-by-position interactive fusion feature map to obtain the optically variable ink area hue-content semantic significant sparse interactive fusion feature map as the optically variable ink area hue-content semantic significant sparse interactive fusion feature.

[0015] In the above-mentioned banknote anomaly detection method based on big data, determining whether the banknote to be detected is an abnormal banknote based on the optically variable ink area color tone-content semantically significant sparse interactive fusion feature includes: determining whether the banknote to be detected is an abnormal banknote based on the optically variable ink area color tone-content semantically significant sparse interactive fusion feature map.

[0016] According to another aspect of the present application, a banknote anomaly detection system based on big data is provided, comprising: an optically variable ink region multimodal feature extraction module, used for performing hue feature extraction and content feature extraction on the optically variable ink region to obtain hue coding features of the optically variable ink region and content semantic coding features of the optically variable ink region; an optically variable ink region multimodal core-related feature joint interactive coding module, used for performing core-related feature joint interactive coding on the hue coding features of the optically variable ink region and the content semantic coding features of the optically variable ink region to obtain hue-content semantic significant sparse interactive fusion features of the optically variable ink region; and a banknote anomaly judgment module to be detected, used for determining whether the banknote to be detected is an abnormal banknote based on the hue-content semantic significant sparse interactive fusion features of the optically variable ink region.

[0017] Compared with the prior art, the present application provides a banknote anomaly detection method and system based on big data, which extracts the hue feature and content semantic feature of the optically variable ink area by applying an image processing and analysis algorithm based on artificial intelligence and deep learning technology, so as to depict the hue and content semantic feature information of the optically variable ink area, and detects whether the banknote is an abnormal banknote based on the significant sparse interactive fusion representation between the two hierarchical features of the optically variable ink area. In this way, multiple features of the banknote can be captured more comprehensively, and the influence of ambient lighting conditions can be avoided when performing banknote anomaly detection, and the influence of image noise and processing errors can be reduced, so as to more comprehensively identify abnormal banknotes and improve the detection accuracy of whether the banknote to be detected is an abnormal banknote. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0019] Figure 1 This is a flow chart of a banknote anomaly detection method based on big data according to an embodiment of the present application.

[0020] Figure 2 The present invention is a flow chart of analyzing the optically variable ink area in the banknote abnormality detection method based on big data according to an embodiment of the present application to determine whether the banknote to be detected is an abnormal banknote.

[0021] Figure 3 The present invention is a schematic diagram of data flow for analyzing the optically variable ink area in the banknote abnormality detection method based on big data according to an embodiment of the present application to determine whether the banknote to be detected is an abnormal banknote.

[0022] Figure 4 The present invention is a flowchart of performing core-related feature joint interactive encoding of the optically variable ink area hue coding feature and the optically variable ink area content semantic coding feature in the big data-based banknote anomaly detection method according to an embodiment of the present application to obtain an optically variable ink area hue-content semantically significant sparse interactive fusion feature.

[0023] Figure 5 The present invention is a flowchart of performing global association coding on the optically variable ink area tone coding feature and the optically variable ink area content semantic coding feature in the big data-based banknote anomaly detection method according to an embodiment of the present application to obtain a global coding representation between the optically variable ink area tone-content semantic features.

[0024] Figure 6 The present invention is a flowchart of inputting the optically variable ink area hue coding feature map and the optically variable ink area content semantic coding feature map into a kernel feature distillation network to obtain an optically variable ink area hue kernel feature vector and an optically variable ink area content semantic kernel feature vector in a banknote anomaly detection method based on big data according to an embodiment of the present application.

[0025] Figure 7 It is a system block diagram of a banknote anomaly detection system based on big data according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] Various exemplary embodiments, features and aspects of the present application will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0027] In terms of banknote anomaly detection, in order to solve the problem of abnormal color change in the optically variable ink area, Chinese patent CN113256873A determines the optically variable ink area in the banknote image and extracts the color tone characteristics of the optically variable ink area, and then detects whether the banknote to be detected is an abnormal banknote based on the color tone characteristics of the optically variable ink area. However, this banknote anomaly detection scheme has high requirements on image quality, and any noise in image processing or error in binarization processing may affect the final banknote anomaly detection result.

[0028] In response to the above technical problems, this application proposes a banknote anomaly detection method based on big data. Figure 1 Flow chart of a banknote anomaly detection method based on big data according to an embodiment of the present application. Figure 1 As shown, according to the big data-based banknote abnormality detection method of the embodiment of the present application, the method includes: S110, acquiring a banknote image of the banknote to be detected; S120, determining the optically variable ink area of ​​the banknote image; S130, analyzing the optically variable ink area to determine whether the banknote to be detected is an abnormal banknote.

[0029] In the above-mentioned banknote anomaly detection method based on big data, the step S110 is to obtain a banknote image of the banknote to be detected. It should be understood that the detection system can analyze various features of the banknote through high-quality banknote images, including watermarks, security lines, tiny fonts and other security features. These features are obviously different between real banknotes and counterfeit banknotes, so the image quality directly affects the accuracy of subsequent analysis and recognition. Specifically, in the technical solution of the present application, a high-resolution flatbed scanner is used to obtain a static image of the banknote.

[0030] In the above-mentioned banknote anomaly detection method based on big data, the step S120 determines the optically variable ink area of ​​the banknote image. It should be understood that optically variable ink is a special ink whose composition contains optically variable materials, which can reflect different colors under different lighting angles. This characteristic enables optically variable ink to effectively prevent counterfeiting, because ordinary printing cannot replicate the effect of this color change. Specifically, in banknotes, optically variable ink is applied to specific patterns or symbols, such as important elements such as denomination numbers. Through optical testing, the optically variable ink area can be effectively identified. Usually, a special light source (such as ultraviolet light or white light of different wavelengths) is used to illuminate the banknote, and the characteristic changes of the optically variable ink can be observed. Under different lighting conditions, some areas of the banknote will show obvious color changes, while other areas remain unchanged. In this way, it can be determined which areas are optically variable ink coatings to further protect banknotes from counterfeiting threats.

[0031] It is worth mentioning that in a specific embodiment of the present application, the step S120, determining the optically variable ink area of ​​the banknote image, includes: S121. Determining a grayscale image of the banknote image, and determining the optically variable area based on the grayscale image; S122. Segmenting the background area and the optically variable ink area in the optically variable area, and generating a binary image; S123. Projecting the binary image to determine the optically variable ink area in the banknote image.

[0032] Specifically, in step S121, the grayscale image of the banknote image is determined, and the optically variable area is determined based on the grayscale image. It should be understood that, considering that the main characteristics of the optically variable ink in the banknote image are usually related to the brightness change of the image, the reflective characteristics of the optically variable ink under different lighting conditions will cause its grayscale value to be significantly different from the grayscale value of the surrounding background area. Therefore, the use of a grayscale image can more intuitively observe and analyze the distribution of the optically variable ink. Among them, a grayscale image is a way to convert a color image into a black and white image. By converting the color information of each pixel into a grayscale value (in the range of 0-255, 0 represents black, 255 represents white), the brightness information of the image is retained and the color information is ignored. More specifically, first, the color image file of the banknote image is loaded. Then, each pixel in the RGB color space is converted into a grayscale value using a weighted average method, wherein, considering that in the visible spectrum, the sensitivity to green light waves is the highest, followed by red, and finally blue, the grayscale value is calculated according to the formula to retain the details and contrast of the image, wherein the formula is: Gray = 0.299 × R + 0.587 × G + 0.114 × B. Then, the calculated grayscale value is used to generate a new image, which is the grayscale image of the banknote. Finally, since optically variable ink usually shows obvious brightness changes in the grayscale image, the grayscale image is divided into an optically variable area and a background area by a threshold.

[0033] Specifically, in step S122, the optically variable background area and the optically variable ink area in the optically variable area are segmented, and a binary image is generated. It should be understood that the optically variable ink area usually has a significant brightness difference with the optically variable background area in the image, and the optically variable background area may contain noise factors such as stray light and stains. Therefore, through segmentation, the two areas can be clearly separated, making the subsequent analysis more efficient and accurate, wherein the generated binary image clearly identifies the optically variable ink and the optically variable background as black and white. More specifically, first, the Otsu algorithm is used to select the optimal threshold by maximizing the inter-class variance to adapt to the brightness distribution of different images and improve the segmentation effect. Then, the optically variable ink area is marked as white and the optically variable background area is marked as black to convert the image into a binary image, thereby clearly reflecting the position and shape of the optically variable ink.

[0034] Specifically, in step S123, the binary image is projected to determine the optically variable ink area in the banknote image. It should be understood that the binary image contains only two values, black and white, which greatly reduces the amount of data. Through projection, the information can be further simplified to highlight specific areas, that is, to quickly and accurately locate and analyze the optically variable ink area, thereby ensuring the security of the banknote. More specifically, horizontal projection is used to accumulate the pixel values ​​of each row of the binary image. That is, for the binary image, each row is traversed, the number of white pixels (optically variable ink area) in each row is counted, and a one-dimensional array is generated to represent the sum of white pixels in each row. Among them, in the projection result, the projection result will show peaks and valleys, and the optically variable ink area in the banknote image can be determined based on the peak part.

[0035] In the above-mentioned banknote anomaly detection method based on big data, the step S130 analyzes the optically variable ink area to determine whether the banknote to be detected is an abnormal banknote. It should be understood that optically variable ink is an important anti-counterfeiting feature in modern banknotes, and its color changes under different lighting conditions and observation angles, so it can effectively identify the authenticity of banknotes. Considering that counterfeit banknotes may not use optically variable ink, or the ink used is of poor quality, resulting in unclear color changes. Therefore, by analyzing the optically variable ink area, it is possible to effectively determine whether the banknote is a formally issued currency. Specifically, in the technical solution of the present application, the image is feature extracted and classified by using a convolutional neural network. In the model training process, by inputting banknote images under different lighting conditions and angles, the color change characteristics of the optically variable ink are learned. After the training is completed, the performance of the model is evaluated using a test set, and the hyperparameters are adjusted to improve the classification accuracy. In the actual detection process, the image of the banknote to be detected is input into the trained model, and the model automatically analyzes the characteristics of the optically variable ink area to determine whether it meets the normal optical variable effect. If the color change is not obvious, the texture is uneven, or it is significantly different from the normal banknote characteristics, the model will mark the banknote as abnormal. This deep learning-based analysis method can improve the accuracy and efficiency of banknote detection.

[0036] Specifically, in the process of analyzing the optically variable ink area to determine whether the banknote to be detected is an abnormal banknote, the color tone feature extraction and content semantic feature extraction of the optically variable ink area are performed by applying an image processing and analysis algorithm based on artificial intelligence and deep learning technology, so as to characterize the color tone and content semantic feature information of the optically variable ink area, and detect whether the banknote is an abnormal banknote based on the significant sparse interactive fusion representation between these two hierarchical features of the optically variable ink area. In this way, by introducing the content semantics of the optically variable ink area in the prior art, multiple features of the banknote can be captured more comprehensively, which is crucial when performing banknote anomaly detection. By additionally considering the content semantic information of the optically variable ink area, the influence of ambient lighting conditions can be avoided when performing banknote anomaly detection, and the influence of image noise and processing errors can also be reduced, thereby helping to more comprehensively identify abnormal banknotes.

[0037] Figure 2 The present invention is a flow chart of analyzing the optically variable ink area in the banknote abnormality detection method based on big data according to an embodiment of the present application to determine whether the banknote to be detected is an abnormal banknote. Figure 3 The data flow diagram of analyzing the optically variable ink area to determine whether the banknote to be detected is an abnormal banknote in the banknote abnormality detection method based on big data according to an embodiment of the present application is shown. Figure 2 and Figure 3As shown, in an embodiment of the present application, the step S130, analyzing the optically variable ink area to determine whether the banknote to be detected is an abnormal banknote, includes: S131, extracting the color tone feature and the content feature of the optically variable ink area to obtain the color tone coding feature of the optically variable ink area and the content semantic coding feature of the optically variable ink area; S132, performing core-related feature joint interactive coding on the color tone coding feature of the optically variable ink area and the content semantic coding feature of the optically variable ink area to obtain the color tone-content semantic significant sparse interactive fusion feature of the optically variable ink area; S133, determining whether the banknote to be detected is an abnormal banknote based on the color tone-content semantic significant sparse interactive fusion feature of the optically variable ink area.

[0038] In an embodiment of the present application, the step S131, performing hue feature extraction and content feature extraction on the optically variable ink area to obtain an optically variable ink area hue coding feature and an optically variable ink area content semantic coding feature, includes: S1311, inputting the optically variable ink area into a hue feature extractor based on a hole convolutional neural network model to obtain an optically variable ink area hue coding feature map as the optically variable ink area hue coding feature; S1312, inputting the optically variable ink area into a content feature extractor based on a deep separable convolutional neural network model to obtain an optically variable ink area content semantic coding feature map as the optically variable ink area content semantic coding feature.

[0039] Specifically, in step S1311, the optically variable ink region is input into the tone feature extractor based on the atrous convolutional neural network model to obtain the tone coding feature map of the optically variable ink region as the tone coding feature of the optically variable ink region. It should be understood that since the atrous convolutional neural network model can expand the receptive field without sacrificing spatial resolution, the tone feature extractor based on the atrous convolutional neural network model can better capture the detailed features of the tone changes in the optically variable ink region and the different color changes that may occur, thereby facilitating the detection of color anomalies in the optically variable ink region of the banknote.

[0040] Specifically, in step S1312, the optically variable ink region is input into a content feature extractor based on a deep separable convolutional neural network model to obtain a content semantic coding feature map of the optically variable ink region as the content semantic coding feature of the optically variable ink region. That is, the optically variable ink region is input into a content feature extractor based on a deep separable convolutional neural network model for feature mining to extract the content semantic feature information in the optically variable ink region, thereby obtaining a content semantic coding feature map of the optically variable ink region. It should be understood that deep separable convolution is an efficient convolution operation method that decomposes the standard convolution operation into two independent steps: deep convolution and point convolution. Deep convolution is responsible for performing spatial convolution on each input channel, while point convolution is used for channel fusion on the output feature map. This method greatly reduces the number of parameters and computational complexity, making the model more lightweight and suitable for use in image feature extraction. Specifically, the optically variable ink region needs to be preprocessed first, including normalization, cropping and enhancement of the image, to ensure that the optically variable ink region input into the model has consistent quality and features. This step ensures that the model can learn the characteristics of optically variable ink more accurately. Next, the preprocessed optically variable ink area image is input into the content feature extractor based on the deep separable convolutional neural network model. During the feature extraction process, the model extracts the spatial features and semantic information of the image layer by layer through multiple convolutional layers. Among them, deep convolution captures local features, such as the texture, color change and glossiness of the optically variable ink, and point convolution combines these local features to form higher-level semantic features. Finally, after multi-layer convolution and activation function processing, the model outputs a content semantic encoding feature map of the optically variable ink area. This feature map not only contains the local features of the optically variable ink, but also integrates the global information of the region, providing rich semantic basis.

[0041] Furthermore, in the step S132, the color tone coding feature of the optically variable ink area and the content semantic coding feature of the optically variable ink area are jointly interactively coded with core related features to obtain the color tone-content semantic significant sparse interactive fusion feature of the optically variable ink area. It should be understood that since the color tone coding feature map of the optically variable ink area and the content semantic coding feature map of the optically variable ink area respectively contain the color tone coding feature and content semantic feature of the optically variable ink area, these two different levels of features respectively reflect different aspects of information of the optically variable ink area, and can complement each other to improve the recognition and detection capabilities of banknote anomalies. For example, when the external lighting conditions and image quality factors affect the color characteristics of the optically variable area, the content semantics of the optically variable ink area can be used as a supplement to ensure that the subsequent banknote anomaly recognition task is correctly performed. Therefore, in order to comprehensively consider the interaction between the hue features and the content semantic features of the optically variable ink area, and extract the most representative and distinguishing feature expressions therefrom, so as to provide a basis for the subsequent banknote anomaly detection task, in the technical solution of the present application, the hue coding features of the optically variable ink area and the content semantic coding features of the optically variable ink area are further subjected to core-related feature joint interactive coding to obtain the hue-content semantic significant sparse interactive fusion features of the optically variable ink area. The processing of the core-related feature joint interactive coding can extract the most representative and distinguishing feature information from the original hue coding feature map of the optically variable ink area and the content semantic coding feature map of the optically variable ink area through multi-level processing steps, and on this basis, effectively fuse these two types of feature information at different levels about the optically variable ink area to obtain a new feature representation, which includes both the color characteristics of the optically variable ink area and the content semantic information, thereby providing a more comprehensive description of the optically variable ink area.

[0042] Figure 4 The present invention is a flowchart of performing core-related feature joint interactive coding of the optically variable ink region hue coding feature and the optically variable ink region content semantic coding feature in the banknote anomaly detection method based on big data according to an embodiment of the present application to obtain the optically variable ink region hue-content semantic significant sparse interactive fusion feature. Figure 4As shown, in an embodiment of the present application, the step S132, performing core-related feature joint interactive coding on the optically variable ink area hue coding feature and the optically variable ink area content semantic coding feature to obtain an optically variable ink area hue-content semantic significant sparse interactive fusion feature, includes: S1321, performing global-domain association coding on the optically variable ink area hue coding feature and the optically variable ink area content semantic coding feature to obtain a global-domain coding representation between the optically variable ink area hue-content semantic features; S1322, performing position-by-position feature association interaction on the optically variable ink area hue coding feature and the optically variable ink area content semantic coding feature to obtain an optically variable ink area hue-content semantic feature position-by-position interactive fusion feature representation; S1323, performing feature-significant sparse modulation on the global-domain coding representation between the optically variable ink area hue-content semantic features based on the optically variable ink area hue-content semantic feature position-by-position interactive fusion feature representation to obtain the optically variable ink area hue-content semantic significant sparse interactive fusion feature.

[0043] Figure 5 The present invention is a flowchart of performing global association coding on the optically variable ink region tone coding feature and the optically variable ink region content semantic coding feature in the banknote anomaly detection method based on big data according to an embodiment of the present application to obtain a global coding representation between the optically variable ink region tone-content semantic features. Figure 5 As shown, in an embodiment of the present application, the step S1321, performing global association coding on the optically variable ink area hue coding feature and the optically variable ink area content semantic coding feature to obtain a global coding representation between the optically variable ink area hue and content semantic features, includes: S1321-1, inputting the optically variable ink area hue coding feature map and the optically variable ink area content semantic coding feature map into a kernel feature distillation network to obtain an optically variable ink area hue kernel feature vector and an optically variable ink area content semantic kernel feature vector; S1321-2, performing global association coding on the optically variable ink area hue kernel feature vector and the optically variable ink area content semantic kernel feature vector to obtain a global coding matrix between the optically variable ink area hue and content semantic kernel features as the global coding representation between the optically variable ink area hue and content semantic features.

[0044] Specifically, in step S1321-1, the optically variable ink region tone coding feature map and the optically variable ink region content semantic coding feature map are input into the kernel feature distillation network to obtain the optically variable ink region tone kernel feature vector and the optically variable ink region content semantic kernel feature vector. It should be understood that the kernel feature distillation network utilizes the powerful representation ability of the neural network, which not only reduces redundant information, but also improves the expressiveness and robustness of the optically variable ink region tone feature and the optically variable ink region content semantic feature.

[0045] Figure 6 The present invention is a flowchart of inputting the optically variable ink region tone coding feature map and the optically variable ink region content semantic coding feature map into a kernel feature distillation network to obtain an optically variable ink region tone kernel feature vector and an optically variable ink region content semantic kernel feature vector in a banknote anomaly detection method based on big data according to an embodiment of the present application. Figure 6 As shown, in an embodiment of the present application, the step S1321-1, inputting the optically variable ink region tone coding feature map and the optically variable ink region content semantic coding feature map into a kernel feature distillation network to obtain an optically variable ink region tone kernel feature vector and an optically variable ink region content semantic kernel feature vector, includes: S1321-11, performing feature decomposition on the optically variable ink region tone coding feature map to obtain a sequence of optically variable ink region tone decomposition feature vectors; S1321-12, calculating each optically variable ink region tone decomposition feature vector in the sequence of the optically variable ink region tone decomposition feature vectors. The method comprises the following steps: calculating the difference characteristics of the optically variable ink area hue decomposition feature vectors relative to other optically variable ink area hue decomposition feature vectors to obtain an optically variable ink area hue decomposition feature difference representation vector; S1321-13, normalizing each eigenvalue in the optically variable ink area hue decomposition feature difference representation vector to obtain a normalized optically variable ink area hue decomposition feature difference representation vector; S1321-14, using each eigenvalue in the normalized optically variable ink area hue decomposition feature difference representation vector as a weight, calculating the weighted sum of the sequence of the optically variable ink area hue decomposition feature vectors to obtain the optically variable ink area hue kernel feature vector.

[0046] In an embodiment of the present application, the step S1321-12, calculating the difference characteristics of each optically variable ink regional hue decomposition feature vector in the sequence of the optically variable ink regional hue decomposition feature vector relative to other optically variable ink regional hue decomposition feature vectors to obtain the optically variable ink regional hue decomposition feature difference representation vector, includes: calculating a norm of the position difference between each optically variable ink regional hue decomposition feature vector in the sequence of the optically variable ink regional hue decomposition feature vector and the other optically variable ink regional hue decomposition feature vectors to obtain a sequence of optically variable ink regional hue decomposition feature difference one-norm representation values; calculating the average value of the sequence of the optically variable ink regional hue decomposition feature difference one-norm representation values ​​to obtain an optically variable ink regional hue decomposition feature difference one-norm average representation value; and vectorizing and arranging a plurality of the optically variable ink regional hue decomposition feature difference one-norm average representation values ​​to obtain the optically variable ink regional hue decomposition feature difference representation vector.

[0047] In an embodiment of the present application, the step S1321-13, normalizing each eigenvalue in the optically variable ink area tone decomposition feature difference representation vector to obtain a normalized optically variable ink area tone decomposition feature difference representation vector, includes: calculating a position-based exponential function value with a natural constant as the base for each eigenvalue in the optically variable ink area tone decomposition feature difference representation vector to obtain an optically variable ink area tone decomposition feature difference class support representation vector; calculating a weighted sum of each eigenvalue in the optically variable ink area tone decomposition feature difference class support representation vector to obtain an optically variable ink area tone decomposition feature difference class support sum; calculating a position-based division between each eigenvalue in the optically variable ink area tone decomposition feature difference class support representation vector and the optically variable ink area tone decomposition feature difference class support sum to obtain the normalized optically variable ink area tone decomposition feature difference representation vector.

[0048] Specifically, in step S1321-2, the optically variable ink region hue kernel feature vector and the optically variable ink region content semantic kernel feature vector are globally associated and encoded to obtain a global encoding matrix between the optically variable ink region hue and content semantic kernel features as a global encoding representation between the optically variable ink region hue and content semantic features. It should be understood that the global association encoding technology is used to process the kernel feature vectors of the color features and content semantic features of the optically variable ink region to construct a global encoding matrix between kernel features that can reflect the degree of correlation and interaction between kernel features of different levels of the optically variable ink region. In particular, in a specific example of the present application, the global association encoding method is to obtain the global encoding matrix between the optically variable ink region hue and content semantic kernel features by vector multiplication. This method helps to better distinguish different types of optically variable ink regions, thereby more accurately performing the banknote anomaly detection task.

[0049] Specifically, in step S1321-2, the optically variable ink area hue kernel feature vector and the optically variable ink area content semantic kernel feature vector are globally associatedly encoded to obtain a global encoding matrix between optically variable ink area hue and content semantic kernel features as a global encoding representation between optically variable ink area hue and content semantic features, including: calculating the multiplication between the transposed vectors of the optically variable ink area hue kernel feature vector and the optically variable ink area content semantic kernel feature vector to obtain the global encoding matrix between the optically variable ink area hue and content semantic kernel features.

[0050] In an embodiment of the present application, the step S1322, performing position-by-position feature association interaction on the optically variable ink area hue coding feature and the optically variable ink area content semantic coding feature to obtain a position-by-position interactive fusion feature representation of the optically variable ink area hue-content semantic feature, includes: S1322-1, inputting the global coding matrix between the optically variable ink area hue-content semantic kernel features into a sparse mask unit to obtain a sparse weight matrix associated with the optically variable ink area hue-content semantic kernel features; S1322-2, inputting the optically variable ink area hue coding feature map and the optically variable ink area content semantic coding feature map into a position-by-position association interaction network to obtain an optically variable ink area hue-content semantic position-by-position interactive fusion feature map as the optically variable ink area hue-content semantic feature position-by-position interactive fusion feature representation.

[0051] Specifically, in step S1322-1, the global encoding matrix between the optically variable ink area hue and the content semantic kernel features is input into the sparse mask unit to obtain the sparse weight matrix between the optically variable ink area hue and the content semantic kernel features. It should be understood that the introduction of the sparsity principle filters out the less important optically variable ink area hue and the content semantic kernel associated feature information, while highlighting those parts that are most critical to the overall information transmission and subsequent detection tasks. This can effectively condense the key information flow paths in the kernel information between the original optically variable ink area hue features and the content semantic features. In this way, the system can significantly reduce the computing cost while maintaining or even improving the performance, which is particularly important for actual deployment, especially for applications in resource-constrained environments.

[0052] Specifically, in the step S1322-1, the global coding matrix between the optically variable ink region hue and the content semantic kernel features is input into a sparse mask unit to obtain a sparse weight matrix associated with the optically variable ink region hue and the content semantic kernel features, including: performing a sparse mask operation on each position feature value in the global coding matrix between the optically variable ink region hue and the content semantic kernel features to obtain a sparse weight matrix associated with the optically variable ink region hue and the content semantic kernel features; wherein, in response to the feature values ​​in the global coding matrix between the optically variable ink region hue and the content semantic kernel features, In response to the eigenvalue in the global coding matrix between the optically variable ink area hue and the content semantic kernel feature being less than or equal to the preset threshold, the eigenvalue less than or equal to the preset threshold is divided by a constant 3; in response to the eigenvalue in the global coding matrix between the optically variable ink area hue and the content semantic kernel feature being greater than the preset threshold and less than or equal to the twice of the preset threshold, the eigenvalue is taken as the original value to obtain the association sparse weight matrix between the optically variable ink area hue and the content semantic kernel feature.

[0053] Specifically, in step S1322-2, the optically variable ink region tone coding feature map and the optically variable ink region content semantic coding feature map are input into the position-by-position association interaction network to obtain an optically variable ink region tone-content semantic position-by-position interactive fusion feature map as the optically variable ink region tone-content semantic feature position-by-position interactive fusion feature representation. It should be understood that in this way, it is possible to pay attention to the feature interaction between the tone feature and the content semantic feature of the optically variable ink region at the local detail level, realize the information exchange and integration between the optically variable ink region tone and the content semantic at each corresponding position across the feature maps, and finally form an optically variable ink region tone-content semantic position-by-position interactive fusion feature map.

[0054] Specifically, in the step S1322-2, the optically variable ink area hue coding feature map and the optically variable ink area content semantic coding feature map are input into a position-by-position association interaction network to obtain an optically variable ink area hue-content semantic position-by-position interactive fusion feature map as the optically variable ink area hue-content semantic feature position-by-position interactive fusion feature representation, including: performing position-by-position point multiplication on the optically variable ink area hue coding feature map and the optically variable ink area content semantic coding feature map to obtain the optically variable ink area hue-content semantic position-by-position interactive fusion feature map.

[0055] Specifically, in step S1323, based on the position-by-position interactive fusion feature representation of the optically variable ink region hue-content semantic feature, the global coding representation between the optically variable ink region hue-content semantic feature is subjected to feature significant sparse modulation to obtain the optically variable ink region hue-content semantic significant sparse interactive fusion feature. It should be understood that the feature significant sparse modulation process adjusts the contribution of each part according to the importance level indicated by the associated sparse weight matrix, so that the adjusted result (optically variable ink region hue-content semantic significant sparse interactive fusion feature map) can better reflect the most representative feature combination in the original input and the most relevant to the subsequent banknote anomaly recognition and detection tasks. By assigning higher weights to these features, the system can better adapt to complex and changeable environmental conditions and the actual application scenario requirements of banknote anomaly detection tasks, providing a basis for subsequent banknote anomaly detection.

[0056] In an embodiment of the present application, the step S1323, based on the optically variable ink area hue-content semantic feature position-by-position interactive fusion feature representation, performs feature-significant sparse modulation on the global coding representation between the optically variable ink area hue-content semantic features to obtain the optically variable ink area hue-content semantic significant sparse interactive fusion feature, includes: based on the optically variable ink area hue-content semantic kernel feature correlation sparse weight matrix, performs feature-significant sparse modulation on the optically variable ink area hue-content semantic position-by-position interactive fusion feature map to obtain the optically variable ink area hue-content semantic significant sparse interactive fusion feature map as the optically variable ink area hue-content semantic significant sparse interactive fusion feature.

[0057] Among them, based on the sparse weight matrix associated between the optically variable ink area hue and content semantics kernel features, the optically variable ink area hue and content semantics position-by-position interactive fusion feature map is subjected to feature significant sparse modulation to obtain the optically variable ink area hue and content semantics significant sparse interactive fusion feature map as the optically variable ink area hue and content semantics significant sparse interactive fusion feature, including: multiplying each feature matrix along the channel dimension in the optically variable ink area hue and content semantics position-by-position interactive fusion feature map by the sparse weight matrix associated between the optically variable ink area hue and content semantics kernel features to obtain the optically variable ink area hue and content semantics significant sparse interactive fusion feature map.

[0058] Specifically, the optically variable ink region tone coding feature and the optically variable ink region content semantic coding feature are subjected to core-related feature joint interactive coding using the following feature joint interactive formula to obtain the optically variable ink region tone-content semantic significant sparse interactive fusion feature; wherein the feature joint interactive formula is: ;in, and are the optically variable ink area tone coding feature map and the optically variable ink area content semantic coding feature map, respectively. For kernel feature distillation, and are the hue kernel feature vector of the optically variable ink region and the content semantic kernel feature vector of the optically variable ink region, is a decomposition operation, are the first decomposition eigenvector, the second decomposition eigenvector, The decomposed eigenvectors, The decomposition eigenvectors and decomposition eigenvectors, is the one-norm of the vector, is the sequence of decomposed eigenvectors, is the number of vectors in the sequence of decomposed eigenvectors minus one, To decompose the eigenvalues ​​of each position in the feature difference representation vector, is the length of the decomposed feature difference representation vector, is the exponential operation, is the kernel eigenvector, is the global encoding matrix between the regional hue of optically variable ink and the content semantic kernel features, is the feature value of each position in the global encoding matrix between the optically variable ink regional hue and content semantic kernel features, is the preset threshold, is a sparse mask operation, is the sparse weight matrix of the correlation between the optically variable ink regional hue and the content semantic kernel features, It is the feature map of the interactive fusion of the regional hue and content semantics of optically variable ink, It is the point multiplication by position. is matrix multiplication, It is the optically variable ink regional hue-content semantically significant sparse interactive fusion feature map.

[0059] In an embodiment of the present application, the step S133, based on the optically variable ink region tone-content semantic significant sparse interactive fusion feature, determines whether the banknote to be detected is an abnormal banknote, including: based on the optically variable ink region tone-content semantic significant sparse interactive fusion feature map, determines whether the banknote to be detected is an abnormal banknote. It should be understood that in the technical solution of the present application, a deep learning classifier based on a deep neural network is selected to learn based on the feature distribution of normal banknotes and abnormal banknotes in historical data. Then, the optically variable ink region tone-content semantic significant sparse interactive fusion feature is input into the trained classifier for judgment. The model will output a probability distribution, indicating the possibility of the banknote being normal or abnormal. If the probability exceeds the set threshold, the banknote is marked as abnormal; otherwise, it is considered to be a normal banknote. This process ensures the rapid and accurate classification of banknotes, which helps prevent the circulation of counterfeit banknotes. Among them, the set threshold is determined by cross-validation and performance evaluation methods, especially using ROC curves and AUC (area under the curve) for analysis. That is to say, the sparse interaction fusion feature information between the hue feature and the content semantic feature of the optically variable ink area is used to detect whether the banknote is an abnormal banknote. In this way, by additionally considering the content semantic information of the optically variable ink area while paying attention to the color feature information of the optically variable ink area, the influence of ambient light conditions can be avoided when detecting banknote anomalies, and the influence of image noise and processing errors can also be reduced, thereby helping to more comprehensively identify abnormal banknotes.

[0060] Preferably, determining whether the banknote to be detected is an abnormal banknote based on the optically variable ink area hue-content semantically significant sparse interactive fusion feature map comprises: determining an optically variable ink area hue-content semantically significant sparse interactive fusion probability value corresponding to the optically variable ink area hue-content semantically significant sparse interactive fusion feature map, wherein the optically variable ink area hue-content semantically significant sparse interactive fusion probability value represents the probability that the banknote to be detected is an abnormal banknote; expanding the optically variable ink area hue-content semantically significant sparse interactive fusion feature map into an optically variable ink area hue-content semantically significant sparse interactive fusion feature map; eigenvector; calculating the product of the first optically variable ink area hue-content semantically significant sparse interactive fusion feature value of the optically variable ink area hue-content semantically significant sparse interactive fusion feature vector and the optically variable ink area hue-content semantically significant sparse interactive fusion probability value to obtain the first optically variable ink area hue-content semantically significant sparse interactive fusion feature probability product; calculating the second optically variable ink area hue-content semantically significant sparse interactive fusion feature value of the optically variable ink area hue-content semantically significant sparse interactive fusion feature vector and the product of the optically variable ink area hue-content semantically significant sparse interactive fusion feature probability product. The product of the difference of the interactive fusion probability values ​​is used to obtain the second optically variable ink area hue-content semantically significant sparse interactive fusion feature inverse probability product; the quotient of the first optically variable ink area hue-content semantically significant sparse interactive fusion feature value divided by one minus the difference of the optically variable ink area hue-content semantically significant sparse interactive fusion probability value is calculated to obtain the first optically variable ink area hue-content semantically significant sparse interactive fusion feature inverse probability quotient; the second optically variable ink area hue-content semantically significant sparse interactive fusion feature value is calculated with the optically variable ink area hue-content semantically significant sparse interactive fusion probability value. The quotient of the values ​​is used to obtain the probability quotient of the second optically variable ink area hue-content semantically significant sparse interactive fusion feature; after adding the first optically variable ink area hue-content semantically significant sparse interactive fusion feature probability product and the second optically variable ink area hue-content semantically significant sparse interactive fusion feature inverse probability product, the difference between the first optically variable ink area hue-content semantically significant sparse interactive fusion feature inverse probability quotient and the second optically variable ink area hue-content semantically significant sparse interactive fusion feature probability quotient is subtracted to obtain the optically variable ink area hue-content semantically significant sparse interactive fusion correction feature value;For each first optically variable ink region hue-content semantically significant sparse interactive fusion feature value and second optically variable ink region hue-content semantically significant sparse interactive fusion feature value of the optically variable ink region hue-content semantically significant sparse interactive fusion feature vector, perform matrix multiplication of the optically variable ink region hue-content semantically significant sparse interactive fusion correction feature matrix composed of the optically variable ink region hue-content semantically significant sparse interactive fusion correction feature value and the optically variable ink region hue-content semantically significant sparse interactive fusion feature vector to obtain a corrected optically variable ink region hue-content semantically significant sparse interactive fusion feature vector; and determine whether the banknote to be detected is an abnormal banknote based on the corrected optically variable ink region hue-content semantically significant sparse interactive fusion feature vector. ;

[0061] Among them, the optically variable ink area hue-content semantically significant sparse interactive fusion feature vector The correction process is expressed as: ;in, represents the i-th eigenvalue of the optically variable ink region hue-content semantically significant sparse interactive fusion feature vector, represents the probability that the banknote to be detected is an abnormal banknote, represents the jth eigenvalue of the optically variable ink region hue-content semantically significant sparse interactive fusion feature vector, Indicates the sparse interactive fusion correction feature value of the optically variable ink area hue-content semantic significance, The optically variable ink regional hue-content semantically significant sparse interactive fusion correction feature matrix is ​​composed of the optically variable ink regional hue-content semantically significant sparse interactive fusion correction feature values, represents the sparse interactive fusion feature vector of the optically variable ink region hue and content semantic significance, Represents the corrected optically variable ink region hue-content semantically salient sparse interaction fusion feature vector.

[0062] That is, when the optically variable ink area hue encoding feature map and the optically variable ink area content semantic encoding feature map respectively represent the image semantic features of the optically variable ink area based on different representation dimensions of hue features and content features, when performing feature significant joint interaction encoding based on kernel association sparse mask, the optically variable ink area hue-content semantic significant sparse interaction fusion feature map will also cause semantic significant joint interaction diversity due to the difference in sparse mask distribution of image semantic association patterns, thereby affecting the iterative consistency of classification mapping and reducing the accuracy of classification results.

[0063] Therefore, the applicant of the present application aims at the region-boundary integral relationship of the high-dimensional feature manifold of the optically variable ink area hue-content semantically significant sparse interactive fusion feature map, and defines the probability distribution representation of the feature set of the optically variable ink area hue-content semantically significant sparse interactive fusion feature map as a manifold constraint boundary to approximate the high-dimensional manifold representation of the feature set of the optically variable ink area hue-content semantically significant sparse interactive fusion feature map composed of eigenvalue pairs, so as to avoid the ambiguity in the mapping of the diversified feature representation of the optically variable ink area hue-content semantically significant sparse interactive fusion feature map to the local manifold representation in the high-dimensional feature class convergence space, improve the execution iteration consistency of each local feature distribution in the mapping task, so as to improve the feature classification convergence effect, and improve the detection accuracy of whether the banknote to be detected is an abnormal banknote.

[0064] In summary, the banknote anomaly detection method based on big data according to the embodiment of the present application is explained, which extracts the hue feature and content semantic feature of the optically variable ink area by applying the image processing and analysis algorithm based on artificial intelligence and deep learning technology, so as to characterize the hue and content semantic feature information of the optically variable ink area, and detects whether the banknote is an abnormal banknote based on the significant sparse interactive fusion representation between the two hierarchical features of the optically variable ink area. In this way, it is possible to capture multiple features of the banknote more comprehensively, avoid the influence of ambient lighting conditions when performing banknote anomaly detection, reduce the influence of image noise and processing errors, thereby more comprehensively identifying abnormal banknotes and improving the detection accuracy of whether the banknote to be detected is an abnormal banknote.

[0065] Figure 7 : is a system block diagram of a banknote anomaly detection system based on big data according to an embodiment of the present application. Figure 7 As shown, according to the big data-based banknote anomaly detection system 100 of the embodiment of the present application, it includes: an optically variable ink region multimodal feature extraction module 110, which is used to extract the hue feature and the content feature of the optically variable ink region to obtain the hue coding feature of the optically variable ink region and the content semantic coding feature of the optically variable ink region; an optically variable ink region multimodal core-related feature joint interactive coding module 120, which is used to perform core-related feature joint interactive coding on the hue coding feature of the optically variable ink region and the content semantic coding feature of the optically variable ink region to obtain the hue-content semantic significant sparse interactive fusion feature of the optically variable ink region; a banknote to be detected abnormality judgment module 130, which is used to determine whether the banknote to be detected is an abnormal banknote based on the hue-content semantic significant sparse interactive fusion feature of the optically variable ink region.

[0066] Here, those skilled in the art can understand that the specific operations of each step in the above-mentioned banknote anomaly detection system based on big data have been referred to above. Figures 1 to 6 The invention has been introduced in detail in the description of the banknote anomaly detection method based on big data, and therefore, its repeated description will be omitted.

[0067] As described above, the banknote anomaly detection system 100 based on big data according to the embodiment of the present application can be implemented in various terminal devices. In one example, the banknote anomaly detection system 100 based on big data can be integrated into the terminal device as a software module and / or a hardware module. For example, the banknote anomaly detection system 100 based on big data can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the banknote anomaly detection system 100 based on big data can also be one of the many hardware modules of the terminal device.

[0068] In summary, a big data-based banknote anomaly detection system based on an embodiment of the present application is explained, which performs hue feature extraction and content semantic feature extraction on the optically variable ink area by applying image processing and analysis algorithms based on artificial intelligence and deep learning technology, so as to characterize the hue and content semantic feature information of the optically variable ink area, and detects whether the banknote is an abnormal banknote based on the significant sparse interactive fusion representation between the two hierarchical features of the optically variable ink area, so as to more comprehensively capture the various features of the banknote, avoid the influence of ambient lighting conditions, reduce the influence of image noise and processing errors, and thus improve the detection accuracy of whether the banknote to be detected is an abnormal banknote.

Claims

1. A banknote anomaly detection method based on big data, characterized in that: include: Acquire a banknote image of a banknote to be detected; Determining an optically variable ink area of ​​the banknote image; Analyzing the optically variable ink area to determine whether the banknote to be detected is an abnormal banknote includes: Performing color tone feature extraction and content feature extraction on the optically variable ink area to obtain color tone coding features of the optically variable ink area and content semantic coding features of the optically variable ink area; The optically variable ink region tone coding feature and the optically variable ink region content semantic coding feature are jointly interactively encoded with core associated features to obtain optically variable ink region tone-content semantic significant sparse interactive fusion features, including: performing global association coding on the optically variable ink region tone coding feature and the optically variable ink region content semantic coding feature to obtain a global coding representation between optically variable ink region tone-content semantic features; performing position-by-position feature association interaction on the optically variable ink region tone coding feature and the optically variable ink region content semantic coding feature to obtain a position-by-position interactive fusion feature representation of optically variable ink region tone-content semantic features; performing feature significant sparse modulation on the global coding representation between the optically variable ink region tone-content semantic features based on the position-by-position interactive fusion feature representation of the optically variable ink region tone-content semantic features to obtain the optically variable ink region tone-content semantic significant sparse interactive fusion feature; Based on the optically variable ink area tone-content semantically significant sparse interactive fusion feature, it is determined whether the banknote to be detected is an abnormal banknote.

2. The banknote anomaly detection method based on big data according to claim 1 is characterized in that: Performing color tone feature extraction and content feature extraction on the optically variable ink area to obtain color tone coding features of the optically variable ink area and content semantic coding features of the optically variable ink area includes: Inputting the optically variable ink region into a tone feature extractor based on a dilated convolutional neural network model to obtain an optically variable ink region tone coding feature map as the optically variable ink region tone coding feature; The optically variable ink area is input into a content feature extractor based on a deep separable convolutional neural network model to obtain an optically variable ink area content semantic coding feature map as the optically variable ink area content semantic coding feature.

3. The banknote anomaly detection method based on big data according to claim 2 is characterized in that: Performing global association coding on the optically variable ink region tone coding feature and the optically variable ink region content semantic coding feature to obtain a global coding representation between the optically variable ink region tone and content semantic features, including: Inputting the optically variable ink region tone coding feature map and the optically variable ink region content semantic coding feature map into a kernel feature distillation network to obtain an optically variable ink region tone kernel feature vector and an optically variable ink region content semantic kernel feature vector; The optically variable ink region hue kernel feature vector and the optically variable ink region content semantic kernel feature vector are globally associated and encoded to obtain a global encoding matrix between optically variable ink region hue and content semantic kernel features as a global encoding representation between optically variable ink region hue and content semantic features.

4. The banknote anomaly detection method based on big data according to claim 3 is characterized in that: Inputting the optically variable ink region tone coding feature map and the optically variable ink region content semantic coding feature map into a kernel feature distillation network to obtain an optically variable ink region tone kernel feature vector and an optically variable ink region content semantic kernel feature vector, comprising: Performing feature decomposition on the optically variable ink regional tone coding feature map to obtain a sequence of optically variable ink regional tone decomposition feature vectors; Calculating the difference feature of each optically variable ink regional tone decomposition feature vector in the sequence of the optically variable ink regional tone decomposition feature vectors relative to other optically variable ink regional tone decomposition feature vectors to obtain an optically variable ink regional tone decomposition feature difference representation vector; Normalizing each eigenvalue in the optically variable ink regional tone decomposition feature difference representation vector to obtain a normalized optically variable ink regional tone decomposition feature difference representation vector; Taking each eigenvalue in the normalized optically variable ink regional tone decomposition feature difference representation vector as a weight, a weighted sum of a sequence of the optically variable ink regional tone decomposition feature vectors is calculated to obtain the optically variable ink regional tone kernel feature vector.

5. The banknote anomaly detection method based on big data according to claim 4 is characterized in that: Calculating the difference feature of each optically variable ink regional tone decomposition feature vector in the sequence of the optically variable ink regional tone decomposition feature vectors relative to other optically variable ink regional tone decomposition feature vectors to obtain an optically variable ink regional tone decomposition feature difference representation vector, including: Calculating the norm of the positional difference between each optically variable ink regional tone decomposition feature vector in the sequence of the optically variable ink regional tone decomposition feature vectors and the other optically variable ink regional tone decomposition feature vectors to obtain a sequence of values ​​representing the norm of the optically variable ink regional tone decomposition feature difference; Calculating the average value of the sequence of the optically variable ink regional tone decomposition feature difference-norm representation values ​​to obtain the optically variable ink regional tone decomposition feature difference-norm average representation value; A plurality of the optically variable ink region tone decomposition feature difference one-norm average representation values ​​are vectorized and arranged to obtain the optically variable ink region tone decomposition feature difference representation vector.

6. The banknote anomaly detection method based on big data according to claim 5 is characterized in that: Normalizing each eigenvalue in the optically variable ink regional tone decomposition feature difference representation vector to obtain a normalized optically variable ink regional tone decomposition feature difference representation vector, including: Calculating the position-based exponential function value of each eigenvalue in the optically variable ink regional tone decomposition feature difference representation vector with a natural constant as the base to obtain an optically variable ink regional tone decomposition feature difference class support representation vector; Calculating the weighted sum of each eigenvalue in the optically variable ink regional tone decomposition feature difference class support representation vector to obtain the total optically variable ink regional tone decomposition feature difference class support; The normalized optically variable ink area hue decomposition feature difference representation vector is obtained by calculating the positional division between each eigenvalue in the optically variable ink area hue decomposition feature difference class support representation vector and the optically variable ink area hue decomposition feature difference class support sum.

7. The banknote anomaly detection method based on big data according to claim 6, characterized in that: Performing position-by-position feature association interaction on the optically variable ink region tone coding feature and the optically variable ink region content semantic coding feature to obtain a position-by-position interactive fusion feature representation of the optically variable ink region tone-content semantic feature, including: Inputting the global encoding matrix between the optically variable ink regional hue and the content semantic kernel features into a sparse mask unit to obtain a correlation sparse weight matrix between the optically variable ink regional hue and the content semantic kernel features; The optically variable ink area hue coding feature map and the optically variable ink area content semantic coding feature map are input into a position-by-position association interaction network to obtain an optically variable ink area hue-content semantic position-by-position interactive fusion feature map as the optically variable ink area hue-content semantic feature position-by-position interactive fusion feature representation.

8. The banknote anomaly detection method based on big data according to claim 7, characterized in that: Based on the position-by-position interactive fusion feature representation of the optically variable ink regional hue-content semantic features, the global coding representation between the optically variable ink regional hue-content semantic features is subjected to feature significant sparse modulation to obtain the optically variable ink regional hue-content semantic significant sparse interactive fusion feature, including: based on the associated sparse weight matrix between the optically variable ink regional hue-content semantic kernel features, the optically variable ink regional hue-content semantic position-by-position interactive fusion feature map is subjected to feature significant sparse modulation to obtain the optically variable ink regional hue-content semantic significant sparse interactive fusion feature map as the optically variable ink regional hue-content semantic significant sparse interactive fusion feature.

9. The banknote anomaly detection method based on big data according to claim 8, characterized in that: Determining whether the banknote to be detected is an abnormal banknote based on the optically variable ink area color tone-content semantically significant sparse interactive fusion feature includes: determining whether the banknote to be detected is an abnormal banknote based on the optically variable ink area color tone-content semantically significant sparse interactive fusion feature map.

10. A banknote anomaly detection system based on big data according to the banknote anomaly detection method based on big data according to claim 1, characterized in that: include: An optically variable ink region multimodal feature extraction module, used for performing hue feature extraction and content feature extraction on the optically variable ink region to obtain a hue coding feature of the optically variable ink region and a content semantic coding feature of the optically variable ink region; An optically variable ink region multimodal core-related feature joint interactive coding module is used to perform core-related feature joint interactive coding on the optically variable ink region hue coding feature and the optically variable ink region content semantic coding feature to obtain an optically variable ink region hue-content semantic significant sparse interactive fusion feature; The banknote abnormality judgment module to be detected is used to determine whether the banknote to be detected is an abnormal banknote based on the optically variable ink area color tone-content semantic significant sparse interactive fusion feature.

Images