Work order behavior detection method, device and equipment and storage medium

Abstract

The application discloses a work order proxy signature behavior detection method and device, equipment and a storage medium, and relates to the computer field.The method comprises the following steps: recognizing a to-be-detected work order image, obtaining a plurality of signature subgraphs contained in the to-be-detected work order image, comparing the signature subgraphs, obtaining the similarity between the signature subgraphs, judging whether similar signatures exist in the plurality of signature subgraphs based on the similarity, and performing proxy signature behavior detection on the to-be-detected work order image according to the judgment result.Because the application obtains the plurality of signature subgraphs contained in the to-be-detected work order image through identification, compares the signature subgraphs, judges whether similar signatures exist based on the similarity between the signature subgraphs, and accurately detects whether proxy signature behavior exists in the to-be-detected work order image, the proxy signature behavior detection efficiency is effectively improved, and the detection cost is greatly reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, and particularly relates to a work order sign behavior detection method and device, equipment and storage medium. BACKGROUND

[0002] The sign on the work order is prone to sign behavior, that is, the same person signs more than one sign on behalf of others. At present, the detection of sign behavior on the work order is to obtain the handwritten Chinese character data of the relevant personnel in advance, to construct a database based on the handwritten Chinese character data, and to maintain the database in real time. When detecting the work order, the sign on the work order is monitored based on the database. However, in the real work order signing environment, it is extremely difficult to obtain and maintain the handwritten Chinese character database of the personnel in the work order. This method has a large cost and low recognition efficiency, and cannot accurately identify the sign behavior in the work order.

[0003] The above content is only used to assist in understanding the technical solutions of the present application and does not represent the acknowledgement of the above content as prior art. SUMMARY

[0004] The main purpose of the present application is to provide a work order sign behavior detection method, device, equipment and storage medium, which aims to solve the technical problems of low detection efficiency and high detection cost of the prior art work order sign behavior detection.

[0005] To achieve the above purpose, the present application provides a work order sign behavior detection method, which comprises the following steps:

[0006] Recognize the to-be-detected work order image to obtain a plurality of signature subgraphs contained in the to-be-detected work order image;

[0007] Compare each signature subgraph to obtain the similarity between each signature subgraph;

[0008] Judge whether there is similar signature in the plurality of signature subgraphs based on the similarity;

[0009] According to the judgment result, the to-be-detected work order image is detected for sign behavior.

[0010] Optionally, the to-be-detected work order image is recognized to obtain a plurality of signature subgraphs contained in the to-be-detected work order image, comprising:

[0011] The to-be-detected work order image is input into a pre-constructed work order detection model for recognition to obtain the signature region bounding box and the personal signature bounding box in the to-be-detected work order image;

[0012] processing the to-be-detected work order image based on the signature region bounding box and the personal signature bounding box to obtain a signature foreground image in the to-be-detected work order image;

[0013] segmenting the signature foreground image to obtain a plurality of signature subgraphs contained in the to-be-detected work order image.

[0014] Optionally, the processing the to-be-detected work order image based on the signature region bounding box and the personal signature bounding box to obtain a signature foreground image in the to-be-detected work order image comprises:

[0015] extracting a signature region foreground image in the to-be-detected work order image based on the signature region bounding box and the personal signature bounding box;

[0016] performing connected component analysis on the signature region foreground image to obtain a vertical line mask and a horizontal line mask of the signature region foreground image;

[0017] combining the vertical line mask and the horizontal line mask to obtain a checkboard line mask result of the signature region foreground image;

[0018] processing the checkboard line mask result to obtain the signature foreground image in the to-be-detected work order image.

[0019] Optionally, the comparing each signature subgraph to obtain a similarity between each signature subgraph comprises:

[0020] processing the signature subgraph;

[0021] inputting the processed signature subgraph into a pre-constructed feature extraction model to obtain a signature feature vector of each signature subgraph;

[0022] comparing the signature feature vectors of each signature subgraph to obtain a similarity between each signature subgraph.

[0023] Optionally, the processing the signature subgraph comprises:

[0024] performing noise reduction processing on each signature subgraph;

[0025] performing connected component analysis on the noise-reduced signature subgraph to obtain coordinate information of the signature subgraph;

[0026] segmenting the signature subgraph based on the coordinate information to obtain a plurality of segmentation regions;

[0027] filtering out an effective region from the plurality of segmentation regions;

[0028] processing the signature subgraph based on the effective region.

[0029] Optionally, before the processed signature subgraphs are input into a pre-constructed feature extraction model to obtain signature feature vectors of the signature subgraphs, the method further includes:

[0030] obtaining original handwritten text data;

[0031] processing the handwritten text data to obtain a handwritten binary image set;

[0032] constructing a training set based on the handwritten binary image set;

[0033] constructing an original model;

[0034] training the original model based on the training set to obtain a feature extraction model.

[0035] Optionally, the comparing of the signature feature vectors of the signature subgraphs to obtain the similarity between the signature subgraphs includes:

[0036] comparing the signature feature vectors of the signature subgraphs;

[0037] constructing a similarity matrix based on the comparison result;

[0038] obtaining the similarity between the signature subgraphs based on the similarity matrix.

[0039] In addition, to achieve the above object, the application further provides a work order proxy signature behavior detection device, which comprises:

[0040] an image recognition module, configured to recognize a to-be-detected work order image to obtain a plurality of signature subgraphs contained in the to-be-detected work order image;

[0041] a similarity obtaining module, configured to compare the signature subgraphs to obtain the similarity between the signature subgraphs;

[0042] a similarity detection module, configured to determine whether there is a similar signature in the plurality of signature subgraphs based on the similarity;

[0043] a proxy signature detection module, configured to detect a proxy signature behavior of the to-be-detected work order image according to the determination result.

[0044] In addition, to achieve the above object, the application further provides a work order proxy signature behavior detection device, which comprises a memory, a processor, and a work order proxy signature behavior detection program stored in the memory and executable on the processor, and the work order proxy signature behavior detection program is configured to implement the steps of the work order proxy signature behavior detection method.

[0045] In addition, to achieve the above object, the application further provides a storage medium, wherein the storage medium stores a work order signature behavior detection program, and the work order signature behavior detection program realizes the steps of the work order signature behavior detection method when executed by a processor.

[0046] The application acquires a plurality of signature subgraphs contained in the to-be-detected work order image by recognizing the to-be-detected work order image, compares each signature subgraph, acquires the similarity between each signature subgraph, judges whether there is a similar signature in the plurality of signature subgraphs based on the similarity, and detects the signature behavior of the to-be-detected work order image according to the judgment result. Since the application acquires a plurality of signature subgraphs contained in the to-be-detected work order image by recognition, compares each signature subgraph, judges whether there is a similar signature based on the similarity between each signature subgraph, accurately detects whether there is a signature behavior in the to-be-detected work order image, effectively improves the signature behavior detection efficiency, and greatly reduces the detection cost. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1 Fig. 1 is a structural schematic diagram of a work order signature behavior detection device of a hardware running environment related to an embodiment scheme of the application;

[0048] Figure 2 Fig. 2 is a flowchart of a first embodiment of a work order signature behavior detection method of the application;

[0049] Figure 3 Fig. 3 is a flowchart of a second embodiment of a work order signature behavior detection method of the application;

[0050] Figure 4 Fig. 4 is a flowchart of a third embodiment of a work order signature behavior detection method of the application;

[0051] Figure 5 Fig. 5 is a feature extraction model structure diagram of the third embodiment of the work order signature behavior detection method of the application;

[0052] Figure 6 Fig. 6 is a comparative learning strategy diagram of the third embodiment of the work order signature behavior detection method of the application;

[0053] Figure 7 Fig. 7 is a structural block diagram of a first embodiment of a work order signature behavior detection device of the application.

[0054] The implementation, functional features and advantages of the application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0055] It should be understood that the specific embodiments described herein are merely intended to explain the application and not to limit the application.

[0056] Refer to Figure 1 , Figure 1 The hardware environment of the embodiment of the present application involves a work order signing behavior detection device structure diagram.

[0057] As Figure 1 shown, the work order signing behavior detection device can include a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 can include a display screen, an input unit such as a keyboard, and an optional user interface 1003 can also include a standard wired interface, a wireless interface. The network interface 1004 can optionally include a standard wired interface, a wireless interface (such as a wireless fidelity (Wi-Fi) interface). The memory 1005 can be a high-speed random access memory (RAM), and can also be a stable non-volatile memory (NVM), such as a disk memory. The memory 1005 can also be an independent storage device from the aforementioned processor 1001.

[0058] Those skilled in the art can understand that Figure 1 the structure shown in the above description does not constitute a limitation on the work order signing behavior detection device, and can include more or fewer components than the diagram, or combine certain components, or different component arrangements.

[0059] As Figure 1 shown, the memory 1005 as a storage medium can include an operating system, a network communication module, a user interface module, and a work order signing behavior detection program.

[0060] In Figure 1 the work order signing behavior detection device, the network interface 1004 is mainly used for data communication with a network server; the user interface 1003 is mainly used for data interaction with a user; the processor 1001 and the memory 1005 in the work order signing behavior detection device can be arranged in the work order signing behavior detection device, and the work order signing behavior detection device calls the work order signing behavior detection program stored in the memory 1005 through the processor 1001, and executes the work order signing behavior detection method provided by the embodiment of the present application.

[0061] The embodiment of the present application provides a work order signing behavior detection method, referring to Figure 2 , Figure 2A flowchart of a first embodiment of a work order proxy signature behavior detection method of the present application is shown in FIG. 1.

[0062] In this embodiment, the work order proxy signature behavior detection method includes the following steps:

[0063] Step S10: Identify the to-be-detected work order image to obtain a plurality of signature subgraphs contained in the to-be-detected work order image.

[0064] It should be noted that in this embodiment, when detecting the signature on the work order, in order to detect whether there is a proxy signature behavior on the work order, i.e., the same person signs more than one name, the embodiment identifies the plurality of signature subgraphs contained in the to-be-detected work order image, compares the signature subgraphs, and determines whether there is a similar signature based on the similarity between the signature subgraphs, thereby accurately detecting whether there is a proxy signature behavior on the work order and effectively improving the proxy signature behavior detection efficiency and accuracy.

[0065] It should be understood that the subject performing the method of this embodiment can be a work order proxy signature behavior detection device having data processing, network communication, and program running functions, such as a computer or the like, or other devices or apparatuses capable of achieving the same or similar functions, which are described hereinabove as an example of the work order proxy signature behavior detection device (hereinafter referred to as a proxy signature detection device).

[0066] It should be noted that the to-be-detected work order image can be an image of a work order that needs to be detected for proxy signature behavior. The signature subgraph can be a single signature image contained in the to-be-detected work order image, and one signature corresponds to one signature subgraph.

[0067] It can be understood that in this embodiment, the to-be-detected work order image is identified to determine the signature area in the to-be-detected work order image, the signature area is further identified to determine a plurality of signatures present in the signature area and / or near the signature area, the signatures are extracted to obtain a plurality of signature subgraphs.

[0068] In a specific implementation, the proxy signature detection device identifies the to-be-detected work order image to determine the signature area in the to-be-detected work order image, identifies the signature area to obtain a signature image containing a plurality of signatures, and segments the signatures in the signature image to obtain a plurality of signature subgraphs.

[0069] Step S20: Compare each of the signature subgraphs to obtain the similarity between each of the signature subgraphs.

[0070] It should be noted that the similarity can be the similarity of the handwritten Chinese characters between each of the signature subgraphs, for example, the similarity can be the feature similarity between each of the signature subgraphs.

[0071] It can be understood that the embodiment extracts multiple signature subgraphs, obtains feature vectors of the signature subgraphs, and compares the signature subgraphs two by two to obtain feature similarities between the signature subgraphs.

[0072] Step S30: judging whether there is a similar signature in the multiple signature subgraphs based on the similarity.

[0073] It should be noted that the similar signature can be a signature with similar handwriting in the to-be-detected work order.

[0074] It can be understood that the embodiment sets a similarity threshold in advance, judges whether the similarity between the signature subgraphs exceeds the similarity threshold, and determines that the two exceeding the threshold are signatures with similar signature styles and handwriting, that is, the two signatures are similar signatures.

[0075] Step S40: performing a signature behavior detection on the to-be-detected work order image according to the judgment result.

[0076] It should be noted that the work order with similar signatures is a suspicious work order, and the suspicious work order is reported and fed back.

[0077] The embodiment obtains multiple signature subgraphs contained in the to-be-detected work order image by recognizing the to-be-detected work order image, compares the signature subgraphs, obtains similarities between the signature subgraphs, judges whether there is a similar signature in the multiple signature subgraphs based on the similarities, and performs a signature behavior detection on the to-be-detected work order image according to the judgment result. Since the embodiment obtains multiple signature subgraphs contained in the to-be-detected work order image by recognition, compares the signature subgraphs, and judges whether there is a similar signature based on the similarities between the signature subgraphs, the signature behavior detection efficiency is effectively improved, and the detection cost is greatly reduced.

[0078] Reference Figure 3 , Figure 3 is a flowchart of a second embodiment of a work order signature behavior detection method.

[0079] Based on the above first embodiment, in the embodiment, the step S10 comprises:

[0080] Step S11: inputting a to-be-detected work order image into a pre-constructed work order detection model for recognition to obtain a signature region bounding box and a personal signature bounding box in the to-be-detected work order image.

[0081] It should be noted that the signature region bounding box can be a target detection bounding box corresponding to the signature region in the to-be-detected work order image, and the signature region bounding box covers the signature region in the to-be-detected work order image. The personal signature bounding box can be a target detection bounding box corresponding to the personal signature in the to-be-detected work order image, and the personal signature bounding box covers at least one personal signature. The work order detection model can be a pre-constructed target detection model.

[0082] It can be understood that in the embodiment, the proxy signature detection device can construct target detection training data by collecting signed work orders, train an original model based on the target detection training data to obtain a work order detection model, and input a to-be-detected work order image to the pre-constructed work order detection model for recognition.

[0083] In a specific implementation, the proxy signature detection device constructs target detection training data by using collected signed work orders, wherein the detection target is a signature region and a specific signature, trains an original model based on the constructed training data to obtain a work order detection model, inputs a to-be-detected work order to the obtained detection model, obtains a corresponding signature region position and a personal signature position, performs bounding box positioning based on the signature region position and the personal signature position, and obtains a signature region bounding box and a personal signature bounding box in the to-be-detected work order image.

[0084] For example, the to-be-detected work order is input to the obtained detection model to obtain a corresponding signature region position coordinate region_box=[rx,ry,rw,rh] and a personal signature position coordinate hw_box=[sx,sy,sw,sh]. The obtained coordinates are further processed by referring to the following formula 1 and formula 2, and finally the signature region bounding box region_box=[rx1,ry1,rx2,ry2] and the personal signature bounding box hw_box=[sx1,sy1,sx2,sy2] are obtained.

[0085] rx1,ry1,rx2,ry2=rx,ry,rx+rw,ry+rh Formula 1

[0086] sx1,sy1,sx2,sy2=sx,sy,sx+sw,sy+sh Formula 2

[0087] Step S12: processing the to-be-detected work order image based on the signature region bounding box and the personal signature bounding box to obtain a signature foreground image in the to-be-detected work order image.

[0088] It should be noted that the signature foreground image can be a foreground image containing only the signature text part in the to-be-detected work order image.

[0089] It should be understood that the embodiment can process the to-be-detected work order image based on the signature region bounding box and the personal signature bounding box in a foreground segmentation, grid line deletion, etc. manner to obtain a foreground, i.e., obtain a signature foreground image in the to-be-detected work order image.

[0090] Further, in order to effectively process the to-be-detected work order image, the above step S12 can include:

[0091] Step S121: extracting a signature region foreground image in the to-be-detected work order image based on the signature region bounding box and the personal signature bounding box;

[0092] Step S122: performing connected component analysis on the signature region foreground image to obtain a vertical line mask and a horizontal line mask of the signature region foreground image;

[0093] Step S123: combining the vertical line mask and the horizontal line mask to obtain a grid line mask result of the signature region foreground image;

[0094] Step S124: processing the grid line mask result to obtain a signature foreground image in the to-be-detected work order image.

[0095] It should be understood that the substitute signature detection device obtains a signature region subgraph region_img = image[ry1:ry2, rx1:rx2] based on the signature region bounding box and the personal signature bounding box and in combination with image data of the to-be-detected work order image, and obtains a signature region foreground by using an adaptive threshold segmentation algorithm (OTSU).

[0096] For the obtained foreground region, a vertical line is obtained by performing an open operation on a binary graph of the above signature region foreground by using a horizontal convolution kernel, a close operation is performed on the obtained result for breakpoint connection, connected component analysis is performed on the obtained result, height information of each connected component is calculated, the connected components are sorted based on the height from large to small, and the first two connected components are taken as a final vertical line connected component result, and the final vertical line mask is obtained by filling the final vertical line connected component result.

[0097] The obtained foreground region is subjected to an open operation on the binary image of the signature region foreground using a vertical convolution kernel to obtain horizontal lines, a close operation is performed on the obtained result for breakpoint connection, connected domain analysis is performed on the obtained result, and the width information of each connected domain is calculated. The connected domains are sorted in descending order of width, and the first four connected domains are taken as the final horizontal line connected domain result. The horizontal line mask is obtained by filling the first four connected domains. The obtained vertical line mask and horizontal line mask are combined to obtain the final grid line mask result. The obtained grid line mask is dilated, and the dilated result is inverted. The grid line region is set to 0. The obtained result is subjected to an AND operation with the binary image of the signature region foreground to remove the grid lines. A close operation is performed on the obtained result to connect the breakpoints, and a signature foreground image containing only the text part is obtained.

[0098] Step S13: segmenting the signature foreground image to obtain a plurality of signature subgraphs contained in the to-be-detected work order image.

[0099] It can be understood that the embodiment obtains the position information of the signature region and the personal signature based on the signature region bounding box and the personal signature bounding box, converts the obtained signature position coordinates to relative coordinates in the current signature region, and the conversion method refers to the following formula 3. Based on the relative coordinates, the specific signature subgraph sig_img = region_img[ty1:ty2,tx1:tx2] is obtained from the processed signature region subgraph. The intersection of the previous and the next signature subgraph is calculated in the vertical order of all obtained signature subgraphs, and the content in the intersection is set to 0 in the next signature subgraph to obtain the processed signature subgraph.

[0100] tx1,ty1,tx2,ty2=sx1-rx1,sy1-ry1,sx2-rx1,sy2-ry1 Formula 3

[0101] The embodiment inputs a to-be-detected work order image into a pre-constructed work order detection model for recognition, obtains a signature region bounding box and a personal signature bounding box in the to-be-detected work order image, processes the to-be-detected work order image based on the signature region bounding box and the personal signature bounding box, obtains a signature foreground image in the to-be-detected work order image, segments the signature foreground image, and obtains a plurality of signature subgraphs contained in the to-be-detected work order image. Since the signature region bounding box and the personal signature bounding box in the to-be-detected work order image are obtained through the pre-constructed work order detection model, the signature in the work order image is subjected to target detection, the signature foreground image containing the signature is effectively extracted by processing the to-be-detected work order image based on the signature region bounding box and the personal signature bounding box, the signature foreground image is segmented to obtain a plurality of signature subgraphs corresponding to the signatures in the work order, the signature recognition accuracy is improved, and the detection efficiency of the signature substitution behavior is improved.

[0102] Reference Figure 4 , Figure 4 The flowchart of the third embodiment of the work order signature substitution behavior detection method is shown.

[0103] Based on the first embodiment, in the embodiment, the step S20 comprises:

[0104] Step S21: processing the signature subgraph.

[0105] It can be understood that the signature subgraph can be preprocessed, for example, the signature subgraph can be subjected to preprocessing operations such as noise reduction, expansion, and segmentation, so as to improve the image quality and improve the subsequent detection efficiency.

[0106] Further, in order to effectively improve the quality of the signature subgraph, the above step S21 can comprise:

[0107] Step S211: performing noise reduction processing on each signature subgraph;

[0108] Step S212: performing connected domain analysis on the noise-reduced signature subgraph to obtain coordinate information of the signature subgraph;

[0109] Step S213: segmenting the signature subgraph based on the coordinate information to obtain a plurality of segmentation regions;

[0110] Step S214: screening an effective region from the plurality of segmentation regions;

[0111] Step S215: processing the signature subgraph based on the effective region.

[0112] It can be understood that the opening operation is performed on the signature subgraph, the influence of discrete noise points is eliminated, and the further dilation operation is performed on the result after noise reduction; the result after the dilation operation is subjected to connected domain calculation, and the circumscribed rectangle of the connected domain is further obtained to obtain coordinate information, based on the obtained coordinate information, the signature image data is cut in the horizontal direction, each cutting region is judged, the invalid region is deleted, and the remaining region is combined in the original order to obtain new signature image data; based on the obtained coordinate information, the maximum value and the minimum value of the y coordinate are retained in the vertical direction based on the new signature image data, and the corresponding subgraph is intercepted to update the signature image data; the updated signature image data is further filled with 0 based on the longest side as the reference, so that the effective region is located at the overall center position, the data after processing is scaled and normalized to meet the model input requirement, and for the new image data combined, the width and height are w and h respectively, and the filling rule is referred to formula 4.

[0113] L = max (w, h) Formula 4

[0114] pad_left = (L-w) / / 2; pad_right = L-w-pad_left

[0115] pad_top = (L-h) / / 2; pad_bottom = L-h-pad_top

[0116] Step S22: input the processed signature subgraph into the pre-constructed feature extraction model to obtain the signature feature vector of each signature subgraph.

[0117] It should be noted that in some embodiments, referring to Figure 5 , Figure 5 is a feature extraction model structure diagram, and the image basic feature extraction adopts the convolution structure of the classification network ResNet50; conv(512 / 3) identifies the convolution calculation on the feature matrix calculated by ResNet50, wherein the convolution kernel is 3x3, and the number of convolution kernels is 512; Flatten represents that the obtained feature matrix is flattened into a one-dimensional feature vector; Dropout represents that part of the neural nodes are inhibited; Dense(256) represents full connection calculation, and outputs a 256-dimensional feature vector; Relu represents that the obtained 256-dimensional feature vector is subjected to relu activation, that is, f(x) = max(0,x); Dense(128) represents full connection calculation, and outputs a 128-dimensional feature vector; and sigmoid represents that the output 128-dimensional feature is further subjected to sigmoid activation, that is, f(x) = 1 / (1+e-x).

[0118] Further, in order to accurately extract the features of the signature subgraph, the above step S22 can include the following steps before the step S22:

[0119] Step S221: obtaining original handwritten text data;

[0120] Step S222: processing the handwritten text data to obtain a handwritten binary image set;

[0121] Step S223: constructing a training set based on the handwritten binary image set;

[0122] Step S224: constructing an original model;

[0123] Step S225: training the original model based on the training set to obtain a feature extraction model.

[0124] It can be understood that the embodiment constructs a feature extraction model and constructs a training data set by using the collected handwritten text data of different persons, wherein the training set can not come from the handwritten data text in the real production environment, and a contrast learning strategy is used to train the feature extraction model, so that the handwritten data feature distance of the same person is closer, and the handwritten data feature distance of different persons is farther.

[0125] In a specific implementation, the signature detection device collects handwritten Chinese text data, one text data is written by one person, and the text data is split into scattered single character graphs or multi-character graphs, which are used to simulate signature handwritten data. The specific steps are as follows:

[0126] A01. Foreground segmentation is performed on the text data;

[0127] A02. The segmented results are dilated;

[0128] A03. Connected component analysis is performed on the dilated binary graph;

[0129] A04. The bounding rectangle of each connected component is obtained, and the smaller width target in the two adjacent rectangles and the obtained too small rectangle are connected to synthesize a rectangular region through the prior information of the character width;

[0130] A05. The subgraph is cut out from the original text image data according to the obtained region coordinate information, and finally a plurality of handwritten characters or connected binary images (not more than 4 characters) of different persons are obtained;

[0131] A06. The obtained data is used to construct a training data set. For image data pairs written by the same person, the label is 1, indicating that the writing style is consistent, and for image data pairs written by different persons, the label is 0, indicating that the writing style is inconsistent.

[0132] Step S23: comparing the signature feature vectors of each of the signature subgraphs to obtain the similarity between each of the signature subgraphs.

[0133] It should be understood that the substitute signature detection device inputs each pre-processed signature subgraph into the feature extraction model, calculates the features of each signature subgraph, and further calculates the cosine similarity between the obtained feature vectors two by two to obtain a similarity matrix, and obtains the similarity between each pair of signature subgraphs based on the similarity matrix.

[0134] Further, in order to improve the comparison accuracy, the above step S23 can include:

[0135] Step S231: comparing the signature feature vectors of each of the signature subgraphs;

[0136] Step S232: constructing a similarity matrix based on the comparison result;

[0137] Step S233: obtaining the similarity between each of the signature subgraphs based on the similarity matrix.

[0138] It can be understood that in the embodiment, the processed signature subgraphs are input into the pre-constructed feature extraction model two by two to form input data pairs, a similarity matrix is obtained, the cosine similarity between each of the signature subgraphs is obtained based on the similarity matrix, and it is determined whether the cosine similarity exceeds a threshold value based on the pre-set threshold value. If the cosine similarity exceeds the threshold value, it is considered that the two signature styles are similar and there is a possibility of substitute signature, and the suspicious work order is fed back.

[0139] In a specific implementation, the substitute signature detection device can perform similarity comparison by constructing a contrast learning strategy, referring to Figure 6 , Figure 6 is a schematic diagram of the contrast learning strategy. Based on the above contrast strategy, a contrast learning loss is constructed, referring to the following formula 5. Based on the constructed feature extraction model and the contrast learning strategy, the feature extraction model weight is learned using the collected training data. During the training process, the model input uses a size of (96, 96), a stochastic gradient descent optimization method is used to optimize the above contrast learning loss, and during the training process, the learning rate is modified by adding a cosine annealing strategy. After the training is completed, the feature extraction model with the smallest loss calculation on the validation set is selected as the final calculation model. The obtained single signature image data is further pre-processed, and the feature extraction model is used to calculate the features of each signature image data, and further calculate the cosine similarity between the obtained feature vectors two by two to obtain a similarity matrix.

[0140]

[0141] The embodiment obtains the signature feature vectors of the signature subgraphs by extracting the signature feature vectors of the signature subgraphs, compares the signature feature vectors of the signature subgraphs, and obtains the similarity between the signature subgraphs, thereby effectively detecting similar signatures existing in the work order and improving the detection accuracy of the signature behavior.

[0142] In addition, the embodiment of the present application further provides a storage medium, and the storage medium stores a work order signature behavior detection program. When the work order signature behavior detection program is executed by a processor, the steps of the work order signature behavior detection method described above are implemented.

[0143] Since the storage medium adopts all the technical solutions of the above-mentioned embodiments, at least all the beneficial effects brought by the technical solutions of the above-mentioned embodiments are achieved, which will not be repeated here.

[0144] Reference Figure 7 , Figure 7 is a structural block diagram of the first embodiment of the work order signature behavior detection device of the present application.

[0145] As shown in Figure 7 , the work order signature behavior detection device provided by the embodiment of the present application comprises:

[0146] An image recognition module 10 is configured to recognize a to-be-detected work order image and obtain a plurality of signature subgraphs contained in the to-be-detected work order image.

[0147] A similarity obtaining module 20 is configured to compare the signature subgraphs and obtain the similarity between the signature subgraphs.

[0148] A similarity detection module 30 is configured to determine whether similar signatures exist in the plurality of signature subgraphs based on the similarity.

[0149] A signature detection module 40 is configured to perform signature behavior detection on the to-be-detected work order image according to the determination result.

[0150] Further, the image recognition module 10 is further configured to input the to-be-detected work order image into a pre-constructed work order detection model for recognition, obtain a signature region bounding box and a personal signature bounding box in the to-be-detected work order image, process the to-be-detected work order image based on the signature region bounding box and the personal signature bounding box, obtain a signature foreground image in the to-be-detected work order image, segment the signature foreground image, and obtain the plurality of signature subgraphs contained in the to-be-detected work order image.

[0151] Further, the image recognition module 10 is further configured to extract a signature region foreground image in the to-be-detected work order image based on the signature region bounding box and the personal signature bounding box; perform connected component analysis on the signature region foreground image to obtain a vertical line mask and a horizontal line mask of the signature region foreground image; combine the vertical line mask and the horizontal line mask to obtain a grid line mask result of the signature region foreground image; and process the grid line mask result to obtain a signature foreground image in the to-be-detected work order image.

[0152] Further, the similarity obtaining module 20 is further configured to process the signature subgraphs; input the processed signature subgraphs into a pre-constructed feature extraction model to obtain signature feature vectors of the signature subgraphs; and compare the signature feature vectors of the signature subgraphs to obtain similarities between the signature subgraphs.

[0153] Further, the similarity obtaining module 20 is further configured to perform noise reduction processing on the signature subgraphs; perform connected component analysis on the noise-reduced signature subgraphs to obtain coordinate information of the signature subgraphs; split the signature subgraphs based on the coordinate information to obtain a plurality of split regions; select effective regions from the plurality of split regions; and process the signature subgraphs based on the effective regions.

[0154] Further, the similarity obtaining module 20 is further configured to obtain original handwritten text data; process the handwritten text data to obtain a handwritten binary image set; construct a training set based on the handwritten binary image set; construct an original model; train the original model based on the training set to obtain a feature extraction model.

[0155] Further, the similarity obtaining module 20 is further configured to compare the signature feature vectors of the signature subgraphs; construct a similarity matrix based on a comparison result; and obtain similarities between the signature subgraphs based on the similarity matrix.

[0156] The embodiment obtains a plurality of signature subgraphs contained in the to-be-detected work order image by recognizing the to-be-detected work order image, compares the signature subgraphs to obtain similarities between the signature subgraphs, determines whether there is a similar signature in the plurality of signature subgraphs based on the similarities, and detects a signature surrogate behavior in the to-be-detected work order image according to a determination result. Since the embodiment obtains a plurality of signature subgraphs contained in the to-be-detected work order image by recognition, compares the signature subgraphs, and determines whether there is a similar signature based on the similarities between the signature subgraphs, the signature surrogate behavior detection efficiency is effectively improved, and the detection cost is greatly reduced.

[0157] It should be understood that the above is only illustrative, and does not constitute any limitation on the technical solutions of the present application. In specific applications, those skilled in the art can set up according to the needs, and the present application does not limit this.

[0158] It should be noted that the above-described workflow is only illustrative and does not limit the scope of protection of the present application. In actual application, those skilled in the art can select part or all of them to achieve the purpose of the embodiment scheme according to the actual needs, which is not limited here.

[0159] In addition, technical details not described in detail in this embodiment can be referred to the work order signing behavior detection method provided by any embodiment of the present application, which will not be repeated here.

[0160] In addition, it should be noted that in this paper, the terms "include", "contain" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or system including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or includes elements inherent to such process, method, article or system. Without more limitations, the element defined by the statement "includes a" does not exclude the presence of another identical element in the process, method, article or system including the element.

[0161] The above embodiment numbers of the present application are only for description, not representing the advantages and disadvantages of the embodiments.

[0162] Through the description of the above embodiments, those skilled in the art can clearly understand that the above embodiment methods can be realized by software and necessary general hardware platform, of course, also can be realized by hardware, but in many cases, the former is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of software product, which is stored in a storage medium (such as read only memory (Read Only Memory, ROM) / RAM, magnetic disk, optical disk), including a plurality of instructions for making a terminal device (which can be a mobile phone, computer, server, or network device, etc.) execute the method described in each embodiment of the present application.

[0163] The above is only the preferred embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made by using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for detecting work order forgery, characterized in that, The method for detecting work order forgery includes: The work order image to be detected is identified to obtain multiple signature sub-images contained in the work order image to be detected; The signature sub-images are compared to obtain the similarity between them. Based on the similarity, determine whether there are similar signatures in the multiple signature sub-graphs; Based on the judgment result, the work order image to be detected is subjected to forged signature detection; The step of identifying the work order image to be detected and obtaining multiple signature sub-images contained in the work order image to be detected includes: The work order image to be detected is input into a pre-built work order detection model for recognition, and the signature region bounding box and the personal signature bounding box in the work order image to be detected are obtained. The signature region bounding box is the target detection bounding box corresponding to the signature region in the work order image to be detected, covering the signature region in the work order image to be detected. The personal signature bounding box is the target detection bounding box corresponding to the personal signature in the work order image to be detected, covering at least one personal signature. The signature region bounding box and the personal signature bounding box are obtained by bounding box localization based on the signature region position and the personal signature position obtained by the work order detection model. The work order image to be detected is processed based on the signature region bounding box and the personal signature bounding box to obtain the signature foreground image in the work order image to be detected. The signature foreground image is segmented to obtain multiple signature sub-images contained in the work order image to be detected.

2. The method for detecting work order forgery as described in claim 1, characterized in that, The step of processing the work order image to be detected based on the signature region bounding box and the personal signature bounding box to obtain the signature foreground image in the work order image to be detected includes: Based on the signature region bounding box and the personal signature bounding box, extract the foreground image of the signature region in the work order image to be detected. Connectivity analysis of the foreground image of the signature region is performed to obtain the vertical and horizontal line masks of the foreground image of the signature region. The vertical line mask and the horizontal line mask are combined to obtain the grid line mask result of the foreground image of the signature area; The grid mask result is processed to obtain the signature foreground image in the work order image to be detected.

3. The method for detecting work order forgery as described in claim 1, characterized in that, The step of comparing each of the signature sub-images to obtain the similarity between each of the signature sub-images includes: The signature subgraph is processed; The processed signature subgraphs are input into a pre-built feature extraction model to obtain the signature feature vectors of each signature subgraph. The signature feature vectors of each signature subgraph are compared to obtain the similarity between each signature subgraph.

4. The method for detecting work order forgery as described in claim 3, characterized in that, The processing of the signature subgraph includes: Denoising processing is performed on each of the aforementioned signature sub-images; Connectivity analysis is performed on the denoised signature subgraph to obtain the coordinate information of the signature subgraph; The signature sub-image is segmented based on the coordinate information to obtain multiple segmentation regions; Valid regions are selected from the multiple segmented regions; The signature subgraph is processed based on the effective region.

5. The method for detecting work order forgery as described in claim 3, characterized in that, Before inputting the processed signature subgraphs into a pre-built feature extraction model to obtain the signature feature vectors of each signature subgraph, the method further includes: Obtain the raw handwritten text data; The handwritten text data is processed to obtain a set of handwritten binary images; A training set is constructed based on the handwritten binary image set; Construct the original model; The original model is trained based on the training set to obtain a feature extraction model.

6. The method for detecting work order forgery as described in claim 3, characterized in that, The step of comparing the signature feature vectors of each signature sub-graph to obtain the similarity between each signature sub-graph includes: The signature feature vectors of each of the aforementioned signature subgraphs are compared; A similarity matrix is ​​constructed based on the comparison results; The similarity between each signature subgraph is obtained based on the similarity matrix.

7. A device for detecting work order forgery, characterized in that, The work order forgery detection device includes: The image recognition module is used to recognize the work order image to be detected and obtain multiple signature sub-images contained in the work order image to be detected; The similarity acquisition module is used to compare each of the signature sub-images and obtain the similarity between each of the signature sub-images; The similarity detection module is used to determine whether there are similar signatures in the plurality of signature subgraphs based on the similarity. The proxy signature detection module is used to detect proxy signature behavior on the work order image to be detected based on the judgment result; The image recognition module is further configured to: input the work order image to be detected into a pre-built work order detection model for recognition, and obtain the signature region bounding box and the personal signature bounding box in the work order image to be detected. The signature region bounding box is a target detection bounding box corresponding to the signature region in the work order image to be detected, covering the signature region in the work order image to be detected. The personal signature bounding box is a target detection bounding box corresponding to the personal signature in the work order image to be detected, covering at least one personal signature. The signature region bounding box and the personal signature bounding box are obtained by bounding box localization based on the signature region position and the personal signature position obtained by the work order detection model. The module further configures the work order image to be detected based on the signature region bounding box and the personal signature bounding box to obtain a signature foreground image in the work order image to be detected. Finally, the module segments the signature foreground image to obtain multiple signature sub-images contained in the work order image to be detected.

8. A device for detecting work order forgery, characterized in that, The work order signing forgery detection device includes: a memory, a processor, and a work order signing forgery detection program stored in the memory and executable on the processor. The work order signing forgery detection program is configured to implement the work order signing forgery detection method as described in any one of claims 1 to 6.

9. A storage medium, characterized in that, The storage medium stores a work order forgery detection program, which, when executed by the processor, implements the work order forgery detection method as described in any one of claims 1 to 6.

Images