Signature verification method and device based on artificial intelligence, computer equipment and medium

Abstract

The invention belongs to the field of artificial intelligence and financial science and technology, and relates to a signature verification method and device based on artificial intelligence, computer equipment and a storage medium, and the method comprises the steps: judging whether a signature verification request triggered by a user is received or not; if yes, carrying out validity verification on the signature verification request; if the signature verification request passes the validity verification, extracting a receipt image from the signature verification request; performing preprocessing operation on the document image to obtain a target document image; calling a preset signature recognition model; performing signature recognition processing on the target document image based on a signature recognition model to obtain a corresponding signature recognition result; and generating a receipt verification result corresponding to the receipt image based on the signature recognition result, and returning the receipt verification result to the user. In addition, the invention also relates to a block chain technology, and the receipt verification result can be stored in a block chain. According to the application, the processing efficiency and the processing accuracy of signature recognition processing can be effectively improved based on the use of the signature recognition model.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence development technology and the field of financial technology, and in particular to artificial intelligence-based signature verification methods, devices, computer equipment and storage media. Background Art

[0002] In the daily operation of the financial industry, the review and processing of documents is a key link to ensure the authenticity and legality of transactions. The document verification system, especially the verification function for the signature on the document, plays a vital role as the first line of defense against fraud and protecting the security of funds. With the development of technology, the automated document verification system has gradually become the industry standard, aiming to improve the efficiency and accuracy of the review and reduce human errors.

[0003] At present, the technical means widely used in the field of document signature verification mainly include methods based on image recognition and image detection. These technologies were originally widely used in the fields of face recognition, document scanning, etc., and have demonstrated powerful image parsing and processing capabilities. However, when directly migrating such general technologies to the specific field of signature verification, a series of challenges and limitations gradually emerge:

[0004] As a unique biometric identifier, signatures vary from person to person. They are not only affected by the writer's personal habits, emotional state, health status and other factors, but may also show significant differences at different times and on different paper materials. This high degree of information dispersion and diversity makes traditional image recognition technology face huge challenges when dealing with signature verification. Direct application of these technologies often makes it difficult to accurately capture and distinguish the subtle differences between different signatures, resulting in low recognition accuracy and difficulty in meeting the high security requirements of the financial field. Summary of the invention

[0005] The purpose of the embodiments of the present application is to propose a signature verification method, device, computer equipment and storage medium based on artificial intelligence to solve the technical problem that traditional image recognition technology has low recognition accuracy in processing signature verification.

[0006] In order to solve the above technical problems, the embodiment of the present application provides a signature verification method based on artificial intelligence, which adopts the following technical solution:

[0007] Determine whether a signature verification request triggered by a user is received; wherein the signature verification request carries a document image;

[0008] If so, verify the validity of the signature verification request;

[0009] If the signature verification request passes the validity verification, extracting the document image from the signature verification request;

[0010] Performing a preprocessing operation on the document image to obtain a corresponding target document image;

[0011] Calling a preset signature recognition model; wherein the signature recognition model is a model obtained by fine-tuning a pre-trained initial signature recognition model using a preset fine-tuning document data set;

[0012] Performing signature recognition processing on the target document image based on the signature recognition model to obtain a corresponding signature recognition result;

[0013] A document verification result corresponding to the document image is generated based on the signature recognition result, and the document verification result is returned to the user.

[0014] Furthermore, the step of verifying the validity of the signature verification request specifically includes:

[0015] Performing authentication information verification on the signature verification request;

[0016] If the signature verification request passes the authentication information verification, performing request format verification on the signature verification request;

[0017] If the signature verification request passes the request format verification, performing security verification on the signature verification request;

[0018] If the signature verification request passes the security verification, then it is determined that the signature verification request passes the validity verification; otherwise, it is determined that the signature verification request fails the validity verification.

[0019] Furthermore, the signature recognition result at least includes a signature position and a signature confidence level; and the step of generating a document verification result corresponding to the document image based on the signature recognition result specifically includes:

[0020] Obtaining a preset expected area and obtaining a preset confidence threshold;

[0021] Extracting the signature position from the signature recognition result, and determining whether the signature position is within the expected area;

[0022] If the signature position is within the expected area, extracting the signature confidence from the signature recognition result, and determining whether the signature confidence is greater than the signature confidence;

[0023] If the signature confidence is greater than or equal to the signature confidence, it is determined that the document image has passed the signature verification, and a first document verification result indicating that the document has passed the signature verification is generated;

[0024] If the signature confidence is less than the signature confidence, it is determined that the document image has not passed the signature verification, and a second document verification result indicating that the document has not passed the signature verification is generated.

[0025] Furthermore, the step of performing a preprocessing operation on the document image to obtain a corresponding target document image specifically includes:

[0026] Scaling the document image based on a preset size to obtain a corresponding first image;

[0027] Performing color correction processing on the first image to obtain a corresponding second image;

[0028] Performing denoising processing on the second image to obtain a corresponding third image;

[0029] Normalizing the third image to obtain a corresponding fourth image;

[0030] The fourth image is used as the target document image.

[0031] Furthermore, before the step of calling the preset signature recognition model, the method further includes:

[0032] Obtaining a pre-built initial network model; wherein the initial network model is composed of a target network structure, a bottleneck layer and a prediction layer;

[0033] Acquire a pre-collected first document image data set, and perform labeling processing on the first document image data set to obtain corresponding sample data;

[0034] Obtain a preset target optimization algorithm, and determine an intersection-over-union loss function, a cross entropy loss function, and a target L1 loss function corresponding to the initial network model;

[0035] Generate a total loss function based on the intersection-over-union loss function, the cross entropy loss function and the target L1 loss function;

[0036] Based on the target optimization algorithm and the total loss function, the initial network model is trained using the sample data to obtain a trained specified network model;

[0037] The designated network model is used as the initial signature recognition model.

[0038] Furthermore, the step of generating a total loss function based on the intersection-over-union loss function, the cross entropy loss function and the target L1 loss function specifically includes:

[0039] Obtain a first weight, a second weight, and a third weight corresponding to the intersection-over-union loss function, the cross entropy loss function, and the target L1 loss function, respectively;

[0040] Call the preset loss generation formula;

[0041] Based on the loss generation formula, the intersection-over-union loss function, the cross entropy loss function, the target L1 loss function, the first weight, the second weight, and the third weight are calculated to obtain corresponding calculation data;

[0042] The calculated data is used as the total loss function.

[0043] Furthermore, before the step of calling the preset signature recognition model, the method further includes:

[0044] Acquire a pre-collected unsigned second document image dataset;

[0045] Performing data enhancement processing on the second document image dataset to obtain a corresponding fine-tuned document dataset;

[0046] Loading the main structure of the initial signature recognition model;

[0047] Constructing a corresponding first model based on the preset pooling layer, classification layer and the main structure;

[0048] Determining target training parameters and target training strategies corresponding to the first model;

[0049] Based on the target training parameters and the target training strategy, fine-tune the first model using the fine-tuning document data set to obtain a second model that meets the fine-tuning expectations;

[0050] The second model is used as the signature recognition model.

[0051] In order to solve the above technical problems, the embodiment of the present application also provides a signature verification device based on artificial intelligence, which adopts the following technical solution:

[0052] A judgment module, used to judge whether a signature verification request triggered by a user is received; wherein the signature verification request carries a document image;

[0053] A verification module, configured to, if yes, verify the validity of the signature verification request;

[0054] an extraction module, configured to extract the document image from the signature verification request if the signature verification request passes the validity verification;

[0055] A preprocessing module, used for performing a preprocessing operation on the document image to obtain a corresponding target document image;

[0056] A calling module, used to call a preset signature recognition model; wherein the signature recognition model is a model obtained by fine-tuning a pre-trained initial signature recognition model using a preset fine-tuning document data set;

[0057] A recognition module, used to perform signature recognition processing on the target document image based on the signature recognition model to obtain a corresponding signature recognition result;

[0058] A processing module is used to generate a document verification result corresponding to the document image based on the signature recognition result, and return the document verification result to the user.

[0059] In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical solution:

[0060] The computer device includes a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps of any one of the artificial intelligence-based signature verification methods proposed in the embodiments of the present application are implemented.

[0061] In order to solve the above technical problems, the embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solution:

[0062] The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, the steps of any one of the artificial intelligence-based signature verification methods proposed in the embodiments of the present application are implemented.

[0063] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0064] The present application first determines whether a signature verification request triggered by a user is received; wherein the signature verification request carries a document image; if so, the validity of the signature verification request is verified; if the signature verification request passes the validity verification, the document image is extracted from the signature verification request; the document image is then preprocessed to obtain a corresponding target document image; thereafter a preset signature recognition model is called; wherein the signature recognition model is a model obtained by fine-tuning a pre-trained initial signature recognition model using a preset fine-tuning document data set; subsequently, signature recognition processing is performed on the target document image based on the signature recognition model to obtain a corresponding signature recognition result; finally, a document verification result corresponding to the document image is generated based on the signature recognition result, and the document verification result is returned to the user. When the present application detects that the signature verification request triggered by the user has passed the validity verification, the document image carried in the signature verification request will be preprocessed to obtain the target document image, and then the target document image will be processed for signature recognition based on the signature recognition model obtained after fine-tuning the pre-trained initial signature recognition model using the preset fine-tuning document data set to generate a corresponding signature recognition result, and then the document verification result corresponding to the document image can be accurately generated based on the signature recognition result. Since the signature recognition model generated after fine-tuning has a simpler structure, it runs faster, which is conducive to improving the processing efficiency of the signature recognition process, and can effectively improve the processing accuracy of the signature recognition process, thereby ensuring the accuracy of the obtained signature recognition result. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0066] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;

[0067] Figure 2 A flowchart of an embodiment of a signature verification method based on artificial intelligence according to the present application;

[0068] Figure 3 is a structural schematic diagram of an embodiment of an artificial intelligence-based signature verification device according to the present application;

[0069] Figure 4 It is a structural diagram of an embodiment of a computer device according to the present application. DETAILED DESCRIPTION

[0070] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0071] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0072] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0073] like Figure 1 As shown, the system architecture 100 may include a terminal device 101, a network 102 and a server 103. The terminal device 101 may be a laptop 1011, a tablet computer 1012 or a mobile phone 1013. The network 102 is used to provide a medium for a communication link between the terminal device 101 and the server 103. The network 102 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0074] The user can use the terminal device 101 to interact with the server 103 through the network 102 to receive or send messages, etc. Various communication client applications can be installed on the terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0075] The terminal device 101 can be various electronic devices with a display screen and supporting web browsing. In addition to the laptop computer 1011, the tablet computer 1012 or the mobile phone 1013, the terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 (Moving Picture Experts Group Audio Layer IV) player, a laptop computer, a desktop computer, etc.

[0076] The server 103 may be a server that provides various services, such as a background server that provides support for a web page displayed on the terminal device 101 .

[0077] It should be noted that the artificial intelligence-based signature verification method provided in the embodiment of the present application is generally executed by a server / terminal device, and accordingly, the artificial intelligence-based signature verification device is generally set in the server / terminal device.

[0078] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0079] Continue to refer Figure 2 , showing a flow chart of an embodiment of an artificial intelligence-based signature verification method according to the present application. According to different requirements, the order of the steps in the flow chart can be changed, and some steps can be omitted. The artificial intelligence-based signature verification method provided in the embodiment of the present application can be applied to any scenario where signature verification is required, and the artificial intelligence-based signature verification method can be applied to products in these scenarios, for example, document signature verification in the field of finance and insurance. The artificial intelligence-based signature verification method comprises the following steps:

[0080] Step S201, determining whether a signature verification request triggered by a user is received; wherein the signature verification request carries a document image.

[0081] In this embodiment, the electronic device (e.g. Figure 1The server / terminal device shown in the figure) can obtain the signature verification request through a wired connection or a wireless connection. It should be noted that the above-mentioned wireless connection method may include but is not limited to 3G / 4G / 5G connection, Wi-Fi connection, Bluetooth connection, Wi MAX connection, Zigbee connection, UWB (ultra wi deband) connection, and other wireless connection methods currently known or to be developed in the future. The executor of the present application is specifically a signature verification system, or simply referred to as the system. The above-mentioned signature verification request is a request triggered by the user in the system for document signature verification of the input document image. In the business scenario of finance and insurance, documents may include insurance documents, claim documents, etc.

[0082] Step S202: If yes, verify the validity of the signature verification request.

[0083] In this embodiment, the specific implementation process of the validity verification of the signature verification request will be further described in detail in subsequent specific embodiments of the present application, and will not be elaborated on here.

[0084] Step S203: If the signature verification request passes the validity verification, extracting the document image from the signature verification request.

[0085] In this embodiment, the document image in the request body of the signature verification request can be parsed according to the content type of the request.

[0086] Step S204, performing a preprocessing operation on the document image to obtain a corresponding target document image.

[0087] In this embodiment, the above-mentioned preprocessing operation is performed on the document image to obtain the specific implementation process of the corresponding target document image. This application will provide further details on this in subsequent specific embodiments and will not be elaborated on here.

[0088] Step S205: calling a preset signature recognition model.

[0089] In this embodiment, the signature recognition model is a model obtained by fine-tuning the pre-trained initial signature recognition model using a preset fine-tuning document data set. The specific construction process of the signature recognition model will be further described in detail in the subsequent specific embodiments of this application, and will not be elaborated on here.

[0090] Step S206: Perform signature recognition processing on the target document image based on the signature recognition model to obtain a corresponding signature recognition result.

[0091] In this embodiment, the target document image is input into the signature recognition model, so as to perform signature recognition processing on the target document image through the signature recognition model, and then output the corresponding signature recognition result. The signature recognition result includes the signature position and the signature confidence. The signature position is in the form of pixel coordinates. The signature confidence is a probability value indicating the recognition accuracy.

[0092] Step S207: generating a document verification result corresponding to the document image based on the signature recognition result, and returning the document verification result to the user.

[0093] In this embodiment, the specific implementation process of generating the document verification result corresponding to the document image based on the signature recognition result will be further described in detail in subsequent specific embodiments of the present application, and will not be elaborated on here.

[0094] When the present application detects that the signature verification request triggered by the user has passed the validity verification, the document image carried in the signature verification request will be preprocessed to obtain the target document image, and then the target document image will be processed for signature recognition based on the signature recognition model obtained after fine-tuning the pre-trained initial signature recognition model using the preset fine-tuning document data set to generate a corresponding signature recognition result, and then the document verification result corresponding to the document image can be accurately generated based on the signature recognition result. Since the signature recognition model generated after fine-tuning has a simpler structure, it runs faster, which is conducive to improving the processing efficiency of the signature recognition process, and can effectively improve the processing accuracy of the signature recognition process, thereby ensuring the accuracy of the obtained signature recognition result.

[0095] In some optional implementations, step S202 includes the following steps:

[0096] Perform authentication information verification on the signature verification request.

[0097] In this embodiment, the authentication information verification includes: checking whether the signature verification request contains valid authentication information, such as an API key, an OAuth token, or a JWT (JSON Web Token). These authentication information are used to identify and verify the identity of the requester. If the signature verification request contains valid authentication information, it is determined that the signature verification request passes the authentication information verification; otherwise, it is determined that the signature verification request fails the authentication information verification, and further determines that the signature verification request fails the validity verification.

[0098] If the signature verification request passes the authentication information verification, a request format verification is performed on the signature verification request.

[0099] In this embodiment, the request format verification includes verifying whether the format of the signature verification request complies with a predetermined specification, including HTTP method (such as GET, POST), request header (such as Content-Type), request body structure, etc. If the format of the signature verification request complies with the predetermined specification, it is determined that the signature verification request passes the request format verification, otherwise, the signature verification request fails the request format verification, and further determines that the signature verification request fails the validity verification.

[0100] If the signature verification request passes the request format verification, security verification is performed on the signature verification request.

[0101] In this embodiment, the security verification includes checking whether the signature verification request contains common security threats, such as SQL injection, cross-site scripting attack documents and other image data. If the signature verification request does not contain security threats, it is determined that the signature verification request has not passed the security verification, otherwise it is determined that the signature verification request has not passed the security verification, and further determines that the signature verification request has not passed the validity verification.

[0102] If the signature verification request passes the security verification, then it is determined that the signature verification request passes the validity verification; otherwise, it is determined that the signature verification request fails the validity verification.

[0103] In this embodiment, if the signature verification request passes the validity verification, it is determined that the signature verification request is a legitimate request from a legitimate client and complies with a prescribed format.

[0104] This application verifies the authentication information of the signature verification request; if the signature verification request passes the authentication information verification, the request format verification is performed on the signature verification request; and if the signature verification request passes the request format verification, the signature verification request is security verified; if the signature verification request passes the security verification, the signature verification request is determined to have passed the validity verification, otherwise the signature verification request is determined to have failed the validity verification. This application verifies the authentication information, request format verification, and security of the signature verification request, thereby combining multiple verification methods to complete the validity verification of the signature verification request, effectively improving the accuracy of the validity verification processing and ensuring the accuracy of the validity verification result of the obtained signature verification request. And only when it is detected that the signature verification request has passed the validity verification, will the subsequent response processing be performed on the signature verification request, thereby improving the processing compliance of the signature verification request, thereby effectively protecting the system from illegal access.

[0105] In some optional implementations of this embodiment, the signature recognition result includes at least the signature position and the signature confidence; step S207 includes the following steps:

[0106] Get a preset expected area and a preset confidence threshold.

[0107] In this embodiment, there is no specific limitation on the selection of the expected area and the value of the confidence threshold, which can be set according to the actual evaluation requirements of the actual document signature verification.

[0108] The signature position is extracted from the signature recognition result, and it is determined whether the signature position is within the expected area.

[0109] In this embodiment, the signature position can be obtained by extracting information from the signature recognition result, and the signature position is in the form of pixel coordinates. Whether the signature position is within the expected area can be determined by detecting whether the pixel coordinates corresponding to the signature position are within the coordinate interval corresponding to the expected area. If the signature position is within the expected area, it is directly determined that the document image has not passed the signature verification, and a document verification result indicating that the document has not passed the signature verification is generated.

[0110] If the signature position is within the expected area, the signature confidence is extracted from the signature recognition result, and it is determined whether the signature confidence is greater than the signature confidence.

[0111] In this embodiment, information may be extracted from the signature recognition result to obtain the signature confidence, which is a probability value representing the recognition accuracy.

[0112] If the signature confidence is greater than or equal to the signature confidence, it is determined that the document image has passed the signature verification, and a first document verification result indicating that the document has passed the signature verification is generated.

[0113] In this embodiment, if the signature position in the signature recognition result is within the expected area and the signature confidence is greater than or equal to the signature confidence, it is determined that the document image passes the signature verification, and a first document verification result indicating that the document passes the signature verification is generated.

[0114] If the signature confidence is less than the signature confidence, it is determined that the document image has not passed the signature verification, and a second document verification result indicating that the document has not passed the signature verification is generated.

[0115] The present application obtains a preset expected area and a preset confidence threshold; then extracts the signature position from the signature recognition result, and determines whether the signature position is within the expected area; if the signature position is within the expected area, extracts the signature confidence from the signature recognition result, and determines whether the signature confidence is greater than the signature confidence; if the signature confidence is greater than or equal to the signature confidence, determines that the document image has passed the signature verification, and generates a first document verification result indicating that the document has passed the signature verification; and if the signature confidence is less than the signature confidence, determines that the document image has not passed the signature verification, and generates a second document verification result indicating that the document has not passed the signature verification. The present application verifies the signature recognition result based on the preset expected area and confidence threshold, and can quickly and accurately generate a document verification result corresponding to the document image, thereby ensuring the data accuracy of the obtained document verification result.

[0116] In some optional implementations, step S204 includes the following steps:

[0117] The document image is scaled based on a preset size to obtain a corresponding first image.

[0118] In this embodiment, the preset size is a uniform size predetermined according to actual needs, such as 640x640 pixels. Specifically, the document image can be cropped as needed to remove unnecessary edges or backgrounds, and the image can be scaled to a uniform size using a resolution function.

[0119] Performing color correction processing on the first image to obtain a corresponding second image.

[0120] In this embodiment, the color correction process includes: adjusting the brightness, contrast and color balance of the image by applying image processing technology to improve the image quality, thereby obtaining a corresponding second image. The image processing technology may adopt histogram equalization, gamma correction, etc.

[0121] The second image is subjected to denoising processing to obtain a corresponding third image.

[0122] In this embodiment, the denoising process includes applying Gaussian filtering or other denoising algorithms (such as median filtering, bilateral filtering) to remove noise in the image, thereby improving the image clarity.

[0123] The third image is normalized to obtain a corresponding fourth image.

[0124] In this embodiment, the normalization process includes: normalizing the image data to the range used in the signature recognition model training to ensure the consistency of the input data.

[0125] The fourth image is used as the target document image.

[0126] The present application can quickly and accurately complete the pre-processing operation of the document image by performing scaling, color correction, denoising and normalization on the document image, effectively ensuring the data accuracy of the obtained target document image, which is conducive to the subsequent signature recognition processing of the target document image by using a signature recognition model to improve the data accuracy of the generated signature recognition result.

[0127] In some optional implementations, before step S205, the electronic device may further perform the following steps:

[0128] Obtain a pre-built initial network model; wherein the initial network model consists of a target network structure, a bottleneck layer and a prediction layer.

[0129] In this embodiment, the initial network model can use YOLOv5 as the basic model. The target network structure is composed of a convolution layer, an activation layer, and a pooling layer. The sub-unit structure adopts a traditional bottleneck structure and a shortCut connection structure (residual connection), and the backbone network is formed by stacking layers. After the features are extracted in the backbone (target network structure), the features enter the bottleneck layer. The bottleneck layer consists of a concat layer, an upsampling layer, and a convolution layer, and the features of different resolutions are input to the prediction layer. By fusing feature maps of different resolutions, the feature expression ability is improved. The above prediction layer is composed of a single convolution layer without BN. Assuming that the input image size is 640x640, after feature extraction, large (80x80), medium (40x40) and small (20x20) resolution features are output and the depth is c, which is used to detect signatures of different sizes. Among them, c=5*anchor_num. The five dimensions represented by 5 are the center coordinates xc, yc of the predicted signature, the width and height w, h of the signature, and the confidence of the target. In the target detection algorithm, the length and width of the target box are regressed based on the size of the anchor. anchor_num represents the number of pre-set anchors. There are three output layers. In the prediction dimensions corresponding to each anchor, the length and width are resolutions, and the points correspond to the pixel positions of the original image. The depth represents the information of the corresponding position of the original image. Assuming that in the prediction output of a certain anchor (anchor_w, anchor_h) in a certain layer, xc_offset represents the offset from the true value xc_gt, yc_offset represents the offset from the true value yc_gt, w represents the target box width, and h represents the target box height. The corresponding relationship of the training target obtained includes: xc_gt=xc+σ(xc_offset); yc_gt=yc+σ(yc_offset); w_gt=anchor_w*exp w ; h_gt = anchor_h*exp h xc,yc are the index values ​​predicted to be 1 in the confidence dimension.

[0130] A first document image data set collected in advance is obtained, and the first document image data set is annotated to obtain corresponding sample data.

[0131] In this embodiment, the first document image dataset is a certain number of document images containing real signatures. The first document image dataset can be annotated using an annotation tool to obtain corresponding sample data. During annotation, the signature boundary box is accurately outlined and a category label (usually "signature") is assigned.

[0132] Obtain a preset target optimization algorithm, and determine an intersection-over-union loss function, a cross entropy loss function, and a target L1 loss function corresponding to the initial network model.

[0133] In this embodiment, the above-mentioned target optimization algorithm is specifically a gradient descent algorithm. The above-mentioned intersection-over-union loss function (Iou_Loss loss) refers to: calculating the intersection-over-union (IoU) loss between the predicted box and the true box. This helps to optimize the position and size of the prediction box. The above-mentioned cross entropy loss function (ceLoss function) refers to: used for classification tasks to distinguish whether the predicted box is a signature. This can be achieved by assigning a binary classification label (signature / non-signature) to each prediction box. The above-mentioned target L1 loss function (Smooth L1 loss function, SmoothLoss) refers to: used for border regression to further optimize the position and size of the prediction box. Smooth L1 loss is a smoothed version of L1 loss, which is smooth near zero, which helps to avoid the gradient explosion problem. Numerical substitution can be performed according to the above-mentioned training target correspondence to establish the intersection-over-union loss function, cross entropy loss function and target L1 loss function corresponding to the initial network model.

[0134] A total loss function is generated based on the intersection-over-union loss function, the cross entropy loss function, and the target L1 loss function.

[0135] In this embodiment, the specific implementation process of generating the total loss function based on the intersection-over-union loss function, the cross entropy loss function and the target L1 loss function will be further described in detail in subsequent specific embodiments of the present application and will not be elaborated here.

[0136] Based on the target optimization algorithm and the total loss function, the initial network model is trained using the sample data to obtain a trained specified network model.

[0137] In this embodiment, training is performed by using labeled sample data. A target optimization algorithm (such as Adam) is selected to update model parameters. During the training process, the changes in the total loss function are monitored regularly, and hyperparameters such as the learning rate and batch size are adjusted as needed. The model performance is evaluated using a validation set, and necessary adjustments (such as early stopping, learning rate decay, etc.) are made based on the evaluation results to ensure that the model converges during the training process, and after the training is completed, a trained specified network model is obtained, and the trained model parameters are saved and processed.

[0138] The designated network model is used as the initial signature recognition model.

[0139] The present application obtains sample data by annotating a pre-collected first document image data set, and then uses the sample data to train a pre-constructed initial network model according to a target optimization algorithm and a total loss function generated based on an intersection-over-union loss function, a cross entropy loss function, and a target L1 loss function, thereby achieving rapid and accurate construction of a corresponding initial signature recognition model, improving the construction efficiency of the initial signature recognition model, and ensuring the model effect of the obtained initial signature recognition model.

[0140] In some optional implementations of this embodiment, generating a total loss function based on the intersection-over-union loss function, the cross entropy loss function, and the target L1 loss function includes the following steps:

[0141] Obtain a first weight, a second weight, and a third weight corresponding to the intersection-over-union loss function, the cross entropy loss function, and the target L1 loss function, respectively.

[0142] In this embodiment, there is no specific limitation on the numerical selection of the first weight, the second weight, and the third weight, which can be determined according to the specific requirements of the task and the relative importance between the loss functions. In addition, a weight generation algorithm can be used to generate the first weight, the second weight, and the third weight corresponding to the intersection-over-union loss function, the cross entropy loss function, and the target L1 loss function, respectively, so as to improve the accuracy of the generated weights. Specifically, the values ​​of the first weight, the second weight, and the third weight can be 0.3, 0.6, and 0.1.

[0143] Call the preset loss generation formula.

[0144] In this embodiment, the above loss generation formula specifically includes: Loss_total=W1*Iou_Loss+W2*ceLoss+W3*SmoothLoss, wherein W1 is the first weight, Iou_Loss is the intersection-over-union loss function, W2 is the second weight, ceLoss is the cross entropy loss function, W3 is the third weight, and smoothLoss is the target L1 loss function.

[0145] Based on the loss generation formula, the intersection-over-union loss function, the cross entropy loss function, the target L1 loss function, the first weight, the second weight and the third weight are calculated and processed to obtain corresponding calculation data.

[0146] In this embodiment, the intersection-over-union loss function, the cross entropy loss function, the target L1 loss function, the first weight, the second weight and the third weight are substituted into the corresponding positions in the above loss generation formula for calculation and processing to obtain the corresponding calculation data.

[0147] The calculated data is used as the total loss function.

[0148] The present application obtains the first weight, second weight and third weight corresponding to the intersection-over-union loss function, the cross entropy loss function and the target L1 loss function respectively, and then uses a preset loss generation formula to calculate and process the intersection-over-union loss function, the cross entropy loss function and the target L1 loss function to quickly and accurately generate the corresponding total loss function, thereby improving the generation efficiency of the total loss function and ensuring the accuracy of the generated total loss function.

[0149] In some optional implementations of this embodiment, before step S205, the electronic device may further perform the following steps:

[0150] A pre-collected unsigned second document image dataset is obtained.

[0151] In this embodiment, the second document image data set is a certain number of document images without signatures collected in advance. These data will be used to train the model to distinguish signed documents from unsigned documents, and can be pre-processed accordingly.

[0152] Data enhancement processing is performed on the second document image dataset to obtain a corresponding fine-tuned document dataset.

[0153] In this embodiment, the data enhancement processing includes: performing data enhancement processing on the second document image data set using a data enhancement strategy to generate a large number of enhanced images, and integrating the enhanced images with the second document image data set to obtain the fine-tuned document data set. The data enhancement strategy may include color transformation (such as brightness and contrast adjustment), rotation, scaling, cropping, affine transformation, etc. In addition, considering the particularity of the signature recognition task, avoid using data enhancement methods that will destroy the signature features (such as extreme blurring or severe distortion). Exemplarily, all data in the second document image data set are enhanced according to color transformation, and random noise is added in the signature area according to Gaussian distribution, so as to obtain a comprehensive enhanced image.

[0154] The main structure of the initial signature recognition model is loaded.

[0155] In this embodiment, the main structure is loaded into the above-mentioned pre-trained initial signature recognition model, and the main structure includes a pre-trained convolutional layer and some customized layers, but does not include a prediction layer for signature detection.

[0156] A corresponding first model is constructed based on the preset pooling layer, classification layer and the main structure.

[0157] In this embodiment, a global average pooling layer is added after the main structure of the initial signature recognition model to convert the feature map into a feature vector of fixed length. A fully connected layer (Dense layer, also called classification layer) is added after the global average pooling layer, and a softmax activation function is used to adjust the output to two categories (signature / non-signature).

[0158] Determine target training parameters and target training strategies corresponding to the first model.

[0159] In this embodiment, the above-mentioned target training parameters include: using a new loss function (such as CELoss, i.e., cross entropy loss) for training. Set a lower learning rate (such as 5e-5) because this is fine-tuning based on the pre-trained model and does not require a large learning step size. Select an appropriate batch size (such as 64), which depends on computing resources and memory limitations. The optimizer uses the adam optimizer. The above-mentioned target training strategy includes: using the adam optimizer for training. In each training iteration, a batch of images is randomly selected from the enhanced fine-tuning document data set, forward propagated through the model, the loss value is calculated, and back propagated to update the model parameters until the loss function converges, the corresponding model parameters are saved, and a trained model that meets the fine-tuning expectations is obtained. The above-mentioned fine-tuning expectations may refer to the loss value or accuracy of the model reaching the expected value.

[0160] Based on the target training parameters and the target training strategy, the first model is fine-tuned using the fine-tuning document dataset to obtain a second model that meets the fine-tuning expectations.

[0161] In this embodiment, according to the policy content of the above-mentioned target training policy, the first model can be fine-tuned using the fine-tuning document data set and the target training parameters to obtain a second model that meets the fine-tuning expectations.

[0162] The second model is used as the signature recognition model.

[0163] In this embodiment, the generated signature recognition model can be deployed to a cloud service, and after importing the saved model parameters, it can be applied in the signature verification system.

[0164] The present application obtains a fine-tuned document data set by performing data enhancement processing on a pre-collected unsigned second document image data set, and constructs a first model based on a preset pooling layer, a classification layer, and the main structure of the initial signature recognition model, and then uses the fine-tuned document data set to fine-tune the first model based on the target training parameters and target training strategy corresponding to the first model, thereby obtaining a second model that meets the fine-tuning expectations and serves as the final signature recognition model. Since the generated signature recognition model has a simpler structure than the initial signature recognition model, it runs faster, which is conducive to improving the processing efficiency of the signature recognition model. In addition, the accuracy of the fine-tuned signature recognition model is further improved compared to the initial signature recognition model.

[0165] In some optional implementations, the user information obtained is subject to the user's consent and complies with relevant laws and policies.

[0166] In addition, the non-Company software tools or components that appear in the embodiments of the present application are merely examples and do not represent actual use.

[0167] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0168] It should be emphasized that in order to further ensure the privacy and security of the above-mentioned document verification results, the above-mentioned document verification results can also be stored in a node of a blockchain.

[0169] The blockchain referred to in this application is a new application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm, etc. Blockchain is essentially a decentralized database, a string of data blocks generated by cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, platform product service layer, and application service layer.

[0170] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0171] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0172] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through computer-readable instructions, and the computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0173] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0174] Further references Figure 3 , as a response to the above Figure 2 The present application provides an embodiment of a signature verification device based on artificial intelligence, and the device embodiment is similar to Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0175] like Figure 3 As shown, the signature verification device 300 based on artificial intelligence described in this embodiment includes: a judgment module 301, a verification module 302, an extraction module 303, a pre-processing module 304, a calling module 305, an identification module 306 and a processing module 307. Among them:

[0176] The judgment module 301 is used to judge whether a signature verification request triggered by a user is received; wherein the signature verification request carries a document image;

[0177] Verification module 302, configured to verify the validity of the signature verification request if yes;

[0178] An extraction module 303, configured to extract the document image from the signature verification request if the signature verification request passes the validity verification;

[0179] A preprocessing module 304 is used to perform a preprocessing operation on the document image to obtain a corresponding target document image;

[0180] The calling module 305 is used to call a preset signature recognition model; wherein the signature recognition model is a model obtained by fine-tuning a pre-trained initial signature recognition model using a preset fine-tuning document data set;

[0181] The recognition module 306 is used to perform signature recognition processing on the target document image based on the signature recognition model to obtain a corresponding signature recognition result;

[0182] The processing module 307 is used to generate a document verification result corresponding to the document image based on the signature recognition result, and return the document verification result to the user.

[0183] In some optional implementations of this embodiment, the verification module 302 includes:

[0184] A first verification submodule, used to verify the authentication information of the signature verification request;

[0185] A second verification submodule, configured to perform request format verification on the signature verification request if the signature verification request passes the authentication information verification;

[0186] A third verification submodule, configured to perform security verification on the signature verification request if the signature verification request passes the request format verification;

[0187] The determination submodule is used to determine that the signature verification request passes the validity verification if the signature verification request passes the security verification, and otherwise determine that the signature verification request fails the validity verification.

[0188] In some optional implementations of this embodiment, the signature recognition result includes at least the signature position and the signature confidence; the processing module 307 includes:

[0189] A first acquisition submodule is used to acquire a preset expected area and a preset confidence threshold;

[0190] A first judgment submodule is used to extract the signature position from the signature recognition result and judge whether the signature position is within the expected area;

[0191] A second judgment submodule is used for extracting the signature confidence from the signature recognition result if the signature position is within the expected area, and judging whether the signature confidence is greater than the signature confidence;

[0192] A first generating submodule, configured to determine that the document image has passed the signature verification if the signature confidence is greater than or equal to the signature confidence, and generate a first document verification result indicating that the document has passed the signature verification;

[0193] The second generating submodule is used for determining that the document image fails to pass the signature verification if the signature confidence is less than the signature confidence, and generating a second document verification result indicating that the document fails to pass the signature verification.

[0194] In some optional implementations of this embodiment, the preprocessing module 304 includes:

[0195] A first processing submodule, configured to scale the document image based on a preset size to obtain a corresponding first image;

[0196] A second processing submodule, configured to perform color correction processing on the first image to obtain a corresponding second image;

[0197] A third processing submodule, configured to perform denoising on the second image to obtain a corresponding third image;

[0198] a fourth processing submodule, configured to perform normalization processing on the third image to obtain a corresponding fourth image;

[0199] The first determining submodule is configured to use the fourth image as the target document image.

[0200] In some optional implementations of this embodiment, the signature verification device based on artificial intelligence further includes:

[0201] A first acquisition module is used to acquire a pre-built initial network model; wherein the initial network model is composed of a target network structure, a bottleneck layer and a prediction layer;

[0202] A labeling module, used to obtain a pre-collected first document image data set, and label the first document image data set to obtain corresponding sample data;

[0203] A second acquisition module is used to acquire a preset target optimization algorithm and determine an intersection-over-union loss function, a cross entropy loss function, and a target L1 loss function corresponding to the initial network model;

[0204] A generation module, used to generate a total loss function based on the intersection-over-union loss function, the cross entropy loss function and the target L1 loss function;

[0205] A training module, used to train the initial network model using the sample data based on the target optimization algorithm and the total loss function to obtain a trained specified network model;

[0206] The first determining module is used to use the specified network model as the initial signature recognition model.

[0207] In some optional implementations of this embodiment, the generating module includes:

[0208] A second acquisition submodule is used to acquire a first weight, a second weight, and a third weight corresponding to the intersection-over-union loss function, the cross entropy loss function, and the target L1 loss function, respectively;

[0209] Call submodule, used to call the preset loss generation formula;

[0210] A calculation submodule, configured to calculate the intersection-over-union loss function, the cross entropy loss function, the target L1 loss function, the first weight, the second weight, and the third weight based on the loss generation formula to obtain corresponding calculation data;

[0211] The second determination submodule is used to use the calculated data as the total loss function.

[0212] In some optional implementations of this embodiment, the signature verification device based on artificial intelligence further includes:

[0213] A third acquisition module, used to acquire a pre-collected unsigned second document image data set;

[0214] An enhancement module, used to perform data enhancement processing on the second document image dataset to obtain a corresponding fine-tuned document dataset;

[0215] A loading module, used to load the main structure of the initial signature recognition model;

[0216] A construction module, used to construct a corresponding first model based on a preset pooling layer, a classification layer and the main structure;

[0217] A second determination module, used to determine target training parameters and target training strategies corresponding to the first model;

[0218] A fine-tuning module, configured to perform fine-tuning training on the first model using the fine-tuning document data set based on the target training parameters and the target training strategy, so as to obtain a second model that meets the fine-tuning expectations;

[0219] The third determining module is used to use the second model as the signature recognition model.

[0220] To solve the above technical problems, the present application also provides a computer device. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0221] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected through a system bus. It should be noted that the figure only shows a computer device 4 with components 41-43, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital processors (DSPs), embedded devices, etc.

[0222] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.

[0223] The memory 41 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 41 can be an internal storage unit of the computer device 4, such as a hard disk or memory of the computer device 4. In other embodiments, the memory 41 can also be an external storage device of the computer device 4, such as a plug-in hard disk equipped on the computer device 4, a smart memory card (Smart Med ia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. Of course, the memory 41 can also include both the internal storage unit of the computer device 4 and its external storage device. In this embodiment, the memory 41 is generally used to store the operating system and various application software installed on the computer device 4, such as computer-readable instructions of a signature verification method based on artificial intelligence, etc. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or are to be output.

[0224] The processor 42 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips in some embodiments. The processor 42 is generally used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to run computer-readable instructions stored in the memory 41 or process data, such as computer-readable instructions for running the artificial intelligence-based signature verification method.

[0225] The network interface 43 may include a wireless network interface or a wired network interface. The network interface 43 is generally used to establish a communication connection between the computer device 4 and other electronic devices.

[0226] The present application also provides another implementation, namely, providing a computer-readable storage medium, wherein the computer-readable storage medium stores computer-readable instructions, and the computer-readable instructions can be executed by at least one processor to enable the at least one processor to perform the steps of the above-mentioned artificial intelligence-based signature verification method.

[0227] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0228] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. A signature verification method based on artificial intelligence, characterized in that: The steps include: Determine whether a signature verification request triggered by a user is received; wherein the signature verification request carries a document image; If so, verify the validity of the signature verification request; If the signature verification request passes the validity verification, extracting the document image from the signature verification request; Performing a preprocessing operation on the document image to obtain a corresponding target document image; Calling a preset signature recognition model; wherein the signature recognition model is a model obtained by fine-tuning a pre-trained initial signature recognition model using a preset fine-tuning document data set; Performing signature recognition processing on the target document image based on the signature recognition model to obtain a corresponding signature recognition result; A document verification result corresponding to the document image is generated based on the signature recognition result, and the document verification result is returned to the user.

2. The signature verification method based on artificial intelligence according to claim 1, characterized in that: The step of verifying the validity of the signature verification request specifically includes: Performing authentication information verification on the signature verification request; If the signature verification request passes the authentication information verification, performing request format verification on the signature verification request; If the signature verification request passes the request format verification, performing security verification on the signature verification request; If the signature verification request passes the security verification, then it is determined that the signature verification request passes the validity verification; otherwise, it is determined that the signature verification request fails the validity verification.

3. The signature verification method based on artificial intelligence according to claim 1, characterized in that: The signature recognition result at least includes a signature position and a signature confidence level; the step of generating a document verification result corresponding to the document image based on the signature recognition result specifically includes: Obtaining a preset expected area and obtaining a preset confidence threshold; Extracting the signature position from the signature recognition result, and determining whether the signature position is within the expected area; If the signature position is within the expected area, extracting the signature confidence from the signature recognition result, and determining whether the signature confidence is greater than the signature confidence; If the signature confidence is greater than or equal to the signature confidence, it is determined that the document image has passed the signature verification, and a first document verification result indicating that the document has passed the signature verification is generated; If the signature confidence is less than the signature confidence, it is determined that the document image has not passed the signature verification, and a second document verification result indicating that the document has not passed the signature verification is generated.

4. The signature verification method based on artificial intelligence according to claim 1, characterized in that: The step of performing a preprocessing operation on the document image to obtain a corresponding target document image specifically includes: Scaling the document image based on a preset size to obtain a corresponding first image; Performing color correction processing on the first image to obtain a corresponding second image; Performing denoising processing on the second image to obtain a corresponding third image; Normalizing the third image to obtain a corresponding fourth image; The fourth image is used as the target document image.

5. The signature verification method based on artificial intelligence according to claim 1, characterized in that: Before the step of calling the preset signature recognition model, the method further includes: Obtaining a pre-built initial network model; wherein the initial network model is composed of a target network structure, a bottleneck layer and a prediction layer; Acquire a pre-collected first document image data set, and perform labeling processing on the first document image data set to obtain corresponding sample data; Obtain a preset target optimization algorithm, and determine an intersection-over-union loss function, a cross entropy loss function, and a target L1 loss function corresponding to the initial network model; Generate a total loss function based on the intersection-over-union loss function, the cross entropy loss function and the target L1 loss function; Based on the target optimization algorithm and the total loss function, the initial network model is trained using the sample data to obtain a trained specified network model; The designated network model is used as the initial signature recognition model.

6. The signature verification method based on artificial intelligence according to claim 5, characterized in that: The step of generating a total loss function based on the intersection-over-union loss function, the cross entropy loss function and the target L1 loss function specifically includes: Obtain a first weight, a second weight, and a third weight corresponding to the intersection-over-union loss function, the cross entropy loss function, and the target L1 loss function, respectively; Call the preset loss generation formula; Based on the loss generation formula, the intersection-over-union loss function, the cross entropy loss function, the target L1 loss function, the first weight, the second weight, and the third weight are calculated to obtain corresponding calculation data; The calculated data is used as the total loss function.

7. The signature verification method based on artificial intelligence according to claim 1, characterized in that: Before the step of calling the preset signature recognition model, the method further includes: Acquire a pre-collected unsigned second document image dataset; Performing data enhancement processing on the second document image dataset to obtain a corresponding fine-tuned document dataset; Loading the main structure of the initial signature recognition model; Constructing a corresponding first model based on the preset pooling layer, classification layer and the main structure; Determining target training parameters and target training strategies corresponding to the first model; Based on the target training parameters and the target training strategy, fine-tune the first model using the fine-tuning document data set to obtain a second model that meets the fine-tuning expectations; The second model is used as the signature recognition model.

8. A signature verification device based on artificial intelligence, characterized in that: include: A judgment module, used to judge whether a signature verification request triggered by a user is received; wherein the signature verification request carries a document image; A verification module, configured to, if yes, verify the validity of the signature verification request; an extraction module, configured to extract the document image from the signature verification request if the signature verification request passes the validity verification; A preprocessing module, used for performing a preprocessing operation on the document image to obtain a corresponding target document image; A calling module, used to call a preset signature recognition model; wherein the signature recognition model is a model obtained by fine-tuning a pre-trained initial signature recognition model using a preset fine-tuning document data set; A recognition module, used to perform signature recognition processing on the target document image based on the signature recognition model to obtain a corresponding signature recognition result; A processing module is used to generate a document verification result corresponding to the document image based on the signature recognition result, and return the document verification result to the user.

9. A computer device, characterized in that: It comprises a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the steps of the signature verification method based on artificial intelligence as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the steps of the artificial intelligence-based signature verification method as described in any one of claims 1 to 7 are implemented.

Images