Electronic file secret label detection method and device, storage medium and electronic equipment

Abstract

The invention provides an electronic archive secret label detection method and device, a storage medium and electronic equipment, and relates to the technical field of target detection, and the method comprises the steps: obtaining an electronic archive image containing a secret-related identifier, and carrying out the preprocessing of the electronic archive image through employing a preset processing method, and obtaining a to-be-recognized secret-related image; inputting a to-be-identified secret-related image into a secret label detection network, and classifying secret-related identifiers based on a secret label detection result; the secret label detection network comprises a first sub-network used for extracting feature information; a second sub-network for optimizing a feature representation of the feature information; and the third sub-network is used for carrying out bounding box, category and confidence prediction on the optimized feature information. According to the electronic archive secret label detection method and device, the storage medium and the electronic equipment provided by the invention, scenes with different complexity and interference degrees can be adapted, and rapid detection and classification of large-scale electronic archives are realized with relatively low resource consumption.

Description

Technical Field

[0001] The present application relates to the technical field of target detection, and in particular, to a method, device, storage medium and electronic device for detecting confidential labels of electronic files. Background Art

[0002] With the advent of the information age, electronic files have become the mainstream form of file management. In order to distinguish different classified levels, it is necessary to mark different classified labels on the electronic files, and different confidentiality measures are adopted to handle the electronic files according to the categories of the classified labels of the electronic files.

[0003] If the classified labels in the electronic files cannot be effectively detected and classified, it will seriously threaten information security. For example, misidentifying a confidential level as a secret level may result in the situation that confidential information is handled according to secret information, thus leading to the leakage of confidential information. Although the detection technologies in the related art have improved the detection efficiency to a certain extent, the detection accuracy and recall rate in complex scenarios still need to be improved, and it is difficult to meet the high-standard requirements for detecting confidential labels. Summary of the Invention

[0004] The purpose of the present application is to provide a method, device, storage medium and electronic device for detecting confidential labels of electronic files, which can adapt to scenarios with different complexities and interference degrees, and realize the rapid detection and classification of large-scale electronic files with low resource consumption.

[0005] The present application provides a method for detecting confidential labels of electronic files, including: Obtaining an electronic file image containing a classified identifier, and preprocessing the electronic file image using a preset processing method to obtain a to-be-recognized classified image; inputting the to-be-recognized classified image into a confidential label detection network to obtain a confidential label detection result, and classifying the classified identifier based on the confidential label detection result to obtain a confidential label classification result; wherein, the confidential label detection network includes: a first sub-network, a second sub-network and a third sub-network; the first sub-network is used to extract feature information in the to-be-recognized classified image; the second sub-network is used to optimize the feature representation of the feature information extracted by the first sub-network and enhance the expression ability of the features; the third sub-network is used to predict the bounding box, category and confidence based on the feature information output by the second sub-network.

[0006] Optionally, the first sub-network includes: a first convolutional layer, a second convolutional layer, and a third convolutional layer; the first convolutional layer includes: a first preset number of convolutional kernels, and the size of the output feature map is a first preset size; the second convolutional layer includes: a second preset number of convolutional kernels, and the size of the output feature map is a second preset size; the third convolutional layer includes: a third preset number of convolutional kernels, and the size of the output feature map is a third preset size; the first convolutional layer, the second convolutional layer, and the third convolutional layer have the same stride; the step of inputting the to-be-recognized classified image into the classified label detection network to obtain the classified label detection result includes: inputting the to-be-recognized classified image into the first sub-network to obtain a first feature map output by the first convolutional layer, a second feature map output by the second convolutional layer, and a third feature map output by the third convolutional layer; wherein, the size of the first feature map is the first preset size, the size of the second feature map is the second preset size, and the size of the third feature map is the third preset size.

[0007] Optionally, the second sub-network includes: a fourth convolutional layer and a fifth convolutional layer; the step of inputting the to-be-recognized classified image into the classified label detection network to obtain the classified label detection result includes: the second sub-network upsamples the third feature map to obtain a fourth feature map, and splices and fuses the fourth feature map with the second feature map, and adjusts the number of channels of the spliced and fused feature map through the fourth convolutional layer to obtain a first fused feature map; the fourth feature map has the same size as the second feature map; upsamples the first fused feature map to obtain a sixth feature map, and splices and fuses the sixth feature map with the first feature map, and adjusts the number of channels of the spliced and fused feature map through the fifth convolutional layer to obtain a second fused feature map; inputs the first fused feature map and the second fused feature map into the third sub-network for predicting the bounding box, category, and confidence, and obtains the classified label detection result; wherein, the sixth feature map has the same size as the first feature map; the number of output channels of the fourth convolutional layer is twice the number of output channels of the fifth convolutional layer.

[0008] Optionally, the third sub-network includes: a sixth convolutional layer, a seventh convolutional layer, an eighth convolutional layer, and a ninth convolutional layer; the number of convolutional kernels of the sixth convolutional layer is the same as the number of channels of the fourth convolutional layer; the step of inputting the first fused feature map and the second fused feature map into the third sub-network to perform prediction of bounding boxes, classes, and confidence levels, and obtaining the classified label detection result includes: inputting the first fused feature map and the second fused feature map into the sixth convolutional layer for feature extraction, and respectively inputting the output result of the sixth convolutional layer into the seventh convolutional layer, the eighth convolutional layer, and the ninth convolutional layer to perform prediction of bounding boxes, classes, and confidence levels, and obtaining the classified label detection result; wherein, the seventh convolutional layer is used to predict the coordinates of the bounding boxes, the eighth convolutional layer is used to predict the probabilities of different classified labels, and the ninth convolutional layer is used to predict the confidence scores indicating the presence of targets within each bounding box; the number of output channels of the seventh convolutional layer is obtained based on the product of the number of anchor boxes and the number of coordinates of each anchor box; the number of output channels of the eighth convolutional layer is obtained based on the product of the number of anchor boxes and the number of classified label classes that each anchor box can predict; the number of output channels of the ninth convolutional layer is obtained based on the number of anchor boxes.

[0009] Optionally, the step of classifying the classified label based on the classified label detection result to obtain a classified label classification result includes: classifying the classified label detection result to determine the class of the identified classified label, and obtaining the classified label classification result.

[0010] This application also provides an electronic file classified label detection device, including: An image processing module, configured to obtain an electronic file image containing a classified label, and preprocess the electronic file image using a preset processing method to obtain a classified label image to be recognized; a classified label detection module, configured to input the classified label image to be recognized into a classified label detection network to obtain a classified label detection result, and classify the classified label based on the classified label detection result to obtain a classified label classification result; wherein, the classified label detection network includes: a first sub-network, a second sub-network, and a third sub-network; the first sub-network is configured to extract feature information from the classified label image to be recognized; the second sub-network is configured to optimize the feature representation of the feature information extracted by the first sub-network and enhance the expression ability of the features; the third sub-network is configured to perform prediction of bounding boxes, classes, and confidence levels based on the feature information output by the second sub-network.

[0011] Optionally, the first sub-network includes: a first convolutional layer, a second convolutional layer, and a third convolutional layer; the first convolutional layer includes: a first preset number of convolutional kernels, and the size of the output feature map is a first preset size; the second convolutional layer includes: a second preset number of convolutional kernels, and the size of the output feature map is a second preset size; the third convolutional layer includes: a third preset number of convolutional kernels, and the size of the output feature map is a third preset size; the first convolutional layer, the second convolutional layer, and the third convolutional layer have the same stride; the classified label detection module is specifically configured to input the to-be-recognized classified image into the first sub-network to obtain a first feature map output by the first convolutional layer, a second feature map output by the second convolutional layer, and a third feature map output by the third convolutional layer; wherein, the size of the first feature map is the first preset size, the size of the second feature map is the second preset size, and the size of the third feature map is the third preset size.

[0012] Optionally, the second sub-network includes: a fourth convolutional layer and a fifth convolutional layer; the classified label detection module is specifically configured to the second sub-network upsample the third feature map to obtain a fourth feature map, splice and fuse the fourth feature map with the second feature map, and adjust the number of channels of the spliced and fused feature map through the fourth convolutional layer to obtain a first fused feature map; the fourth feature map has the same size as the second feature map; the classified label detection module is specifically further configured to upsample the first fused feature map to obtain a sixth feature map, splice and fuse the sixth feature map with the first feature map, and adjust the number of channels of the spliced and fused feature map through the fifth convolutional layer to obtain a second fused feature map; the classified label detection module is specifically further configured to input the first fused feature map and the second fused feature map into the third sub-network for predicting the bounding box, category, and confidence, and obtain the classified label detection result; wherein, the sixth feature map has the same size as the first feature map; the number of output channels of the fourth convolutional layer is twice the number of output channels of the fifth convolutional layer.

[0013] Optionally, the third sub-network includes: a sixth convolutional layer, a seventh convolutional layer, an eighth convolutional layer, and a ninth convolutional layer; the number of convolutional kernels of the sixth convolutional layer is the same as the number of channels of the fourth convolutional layer; the classified label detection module is specifically configured to input the first fused feature map and the second fused feature map into the sixth convolutional layer for feature extraction, and input the output result of the sixth convolutional layer into the seventh convolutional layer, the eighth convolutional layer, and the ninth convolutional layer respectively for predicting the bounding box, category, and confidence, and obtain the classified label detection result; wherein, the seventh convolutional layer is used to predict the coordinates of the bounding box, the eighth convolutional layer is used to predict the probabilities of different classified labels, and the ninth convolutional layer is used to predict the confidence score that there is a target within each bounding box; the output channels of the seventh convolutional layer are obtained based on the product of the number of anchor boxes and the number of coordinates of each anchor box; the output channels of the eighth convolutional layer are obtained based on the product of the number of anchor boxes and the number of classified label categories that each anchor box can predict; the output channels of the ninth convolutional layer are obtained based on the number of anchor boxes.

[0014] Optionally, the classified label detection module is specifically configured to classify the classified label detection result, determine the category of the identified classified secret label, and obtain the classified label classification result.

[0015] The present application also provides a computer program product, including computer programs / instructions, which when executed by a processor implement the steps of the electronic file classified label detection method as described in any one of the above.

[0016] The present application also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein when the processor executes the program, it implements the steps of the electronic file classified label detection method as described in any one of the above.

[0017] The present application also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the electronic file classified label detection method as described in any one of the above.

[0018] The electronic file confidential label detection method, device, storage medium and electronic device provided by this application. First, an electronic file image containing a confidential label is obtained, and the electronic file image is preprocessed using a preset processing method to obtain a to-be-recognized confidential image. Then, the to-be-recognized confidential image is input into a confidential label detection network to obtain a confidential label detection result, and the confidential label is classified based on the confidential label detection result to obtain a confidential label classification result. Among them, the confidential label detection network includes: a first sub-network, a second sub-network, and a third sub-network. The first sub-network is used to extract feature information from the to-be-recognized confidential image. The second sub-network is used to optimize the feature representation of the feature information extracted by the first sub-network and enhance the expression ability of the features. The third sub-network is used to predict the bounding box, category, and confidence based on the feature information output by the second sub-network. In this way, it can adapt to scenarios with different complexities and interference levels and achieve rapid detection and classification of large-scale electronic files with low resource consumption. Description of the Drawings

[0019] In order to more clearly illustrate the technical solutions in this application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of this application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0020] Figure 1 It is a flowchart of the electronic file confidential label detection method provided by this application Figure 2 It is a structural diagram of the electronic file confidential label detection network provided by this application; Figure 3 It is a structural diagram of the electronic file confidential label detection device provided by this application; Figure 4 It is a structural diagram of the electronic device provided by this application. Detailed Embodiments

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the following will clearly and completely describe the technical solutions in this application with reference to the drawings in this application. Obviously, the described embodiments are some, but not all, of the embodiments of this application. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of this application.

[0022] The terms "first", "second", etc. in the description and claims of this application are used to distinguish similar objects, rather than to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of this application can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first", "second", etc. are usually of the same type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" in the description and claims means at least one of the connected objects, and the character " / ", generally indicates an "or" relationship between the associated objects before and after.

[0023] In view of the technical problems of low accuracy and low detection efficiency of the classified identification detection of electronic files in the related art, the embodiments of this application provide an electronic file classified identification detection method, which deeply integrates deep learning, image processing technology, etc., aiming to solve the problems of low accuracy and low recall rate existing in the prior art in the detection of electronic file classified identification. The specific implementation process includes: receiving an electronic file image to be detected; preprocessing the received electronic file image, including steps such as image denoising, size adjustment, normalization, etc., to improve the image quality and meet the requirements of the detection network; accurately detecting the classified identification (such as classified seals and classified text) in the electronic file through a classified identification detection network; classifying the detection results output by the detection, and identifying the categories of the classified identification, including but not limited to top secret, secret, confidential, etc.; outputting the classified classified identification information, including the position, category, confidence level of the identification, and relevant security suggestions or handling measures. This method significantly improves the accuracy and recall rate of electronic file classified identification detection, ensures the security and integrity of classified files. It enhances the robustness and generalization ability of the system, and adapts to scenarios with different complexities and interference levels. It improves the processing efficiency and is applicable to the rapid detection and classification of large-scale electronic files.

[0024] The following will combine the drawings and specifically illustrate the electronic file classified identification detection method provided by the embodiments of this application through specific embodiments and their application scenarios.

[0025] As Figure 1 shown, an electronic file classified identification detection method provided by an embodiment of this application may include the following steps 101 and 102: Step 101, obtain an electronic file image containing a classified identification, and preprocess the electronic file image using a preset processing method to obtain a classified image to be recognized.

[0026] Exemplarily, the above electronic file image can be images in multiple formats, for example, Joint Photographic Experts Group (JPEG), Portable Network Graphics (PNG), and Tag Image File Format (TIFF), etc. Moreover, the electronic file image may contain one or more classified marks, or the electronic file image is a non-classified image.

[0027] Exemplarily, before detecting the input electronic file image, it is also necessary to preprocess the electronic file image according to a preset processing method to improve the accuracy of subsequent feature extraction. The preset processing method includes: denoising, size adjustment, grayscale conversion, binarization, normalization, edge detection, and image cropping, etc., to remove irrelevant information in the image and highlight the classified mark features.

[0028] Step 102: Input the to-be-identified classified image into the classified mark detection network to obtain a classified mark detection result, and classify the classified mark based on the classified mark detection result to obtain a classified mark classification result.

[0029] Among them, the classified mark detection network includes: a first sub-network, a second sub-network, and a third sub-network; the first sub-network is used to extract feature information in the to-be-identified classified image; the second sub-network is used to optimize the feature representation of the feature information extracted by the first sub-network and enhance the expression ability of the features; the third sub-network is used to predict the bounding box, category, and confidence based on the feature information output by the second sub-network.

[0030] Exemplarily, as Figure 2 shown, the classified mark detection network provided in the embodiment of the present application may include a first sub-network, a second sub-network, and a third sub-network. Among them, the first sub-network includes: a first convolutional layer, a second convolutional layer, and a third convolutional layer; the first convolutional layer includes: a first preset number of convolutional kernels, and the size of the output feature map is a first preset size; the second convolutional layer includes: a second preset number of convolutional kernels, and the size of the output feature map is a second preset size; the third convolutional layer includes: a third preset number of convolutional kernels, and the size of the output feature map is a third preset size; the first convolutional layer, the second convolutional layer, and the third convolutional layer have the same stride.

[0031] Specifically, the structure of the above first sub-network includes: Input layer: The input is an RGB image with an input size of 224*224*3; First convolutional layer: 32 3x3 convolutional kernels, with a stride of 1, padding of same, activation function of Relu, using batch normalization (Batch Normalization), pooling is max pooling, including a 2x2 pooling kernel, with a stride of 2, and the output size is 112*112*32 (i.e., the above first preset size); Second convolutional layer: 64 3x3 convolutional kernels, with a stride of 1, padding of same, activation function of Relu, using batch normalization (Batch Normalization), pooling is max pooling, including a 2x2 pooling kernel, with a stride of 2, and the output size is 56*56*64 (i.e., the above second preset size); Third convolutional layer: 128 3x3 convolutional kernels, with a stride of 1, padding of same, activation function of Relu, using batch normalization (Batch Normalization), pooling is max pooling, including a 2x2 pooling kernel, with a stride of 2, and the output size is 28*28*128 (i.e., the above third preset size).

[0032] Specifically, based on the structure of the above first sub-network, step 102 may further include the following step 102a: Step 102a: Input the to-be-recognized classified image into the first sub-network to obtain the first feature map output by the first convolutional layer, the second feature map output by the second convolutional layer, and the third feature map output by the third convolutional layer.

[0033] Among them, the size of the first feature map is the first preset size, the size of the second feature map is the second preset size, and the size of the third feature map is the third preset size.

[0034] Exemplarily, each convolutional neural network in the above first sub-network can extract features from the input image and output a feature map. Then, the obtained feature map is input into the second sub-network to enhance the feature expression ability.

[0035] Specifically, step 102 may further include the following steps 102b1 to 102b3: Step 102b1: The second sub-network upsamples the third feature map to obtain a fourth feature map, splices and fuses the fourth feature map with the second feature map, and adjusts the number of channels of the spliced and fused feature map through a fourth convolutional layer to obtain a first fused feature map.

[0036] Among them, the fourth feature map and the second feature map have the same size.

[0037] Step 102b2: Upsample the first fused feature map to obtain a sixth feature map, splice and fuse the sixth feature map with the first feature map, and adjust the number of channels of the spliced and fused feature map through a fifth convolutional layer to obtain a second fused feature map.

[0038] Step 102b3: Input the first fused feature map and the second fused feature map into the third sub-network to predict the bounding box, category, and confidence, and obtain the encrypted label detection result.

[0039] Among them, the size of the sixth feature map is the same as that of the first feature map; the number of output channels of the fourth convolutional layer is twice that of the fifth convolutional layer.

[0040] Exemplarily, the above second sub-network includes: a fourth convolutional layer and a fifth convolutional layer; the number of output channels of the fourth convolutional layer is twice that of the fifth convolutional layer. The structure of this second sub-network is as follows: Upsample the third convolutional layer to the output size (56*56) of the second convolutional layer, and bilinear interpolation can be used. The specific upsampling method can be selected according to the actual situation. Splice and fuse the upsampled third convolutional layer and the second convolutional layer, and then adjust the number of output channels to 64 through a convolutional layer with 64 3*3 convolutional kernels. Upsample the fused feature map to the output size (112*112) of the first convolutional layer again, and bilinear interpolation can be used. The specific upsampling method can be selected according to the actual situation. Splice and fuse the upsampled convolutional layer and the first convolutional layer, and then adjust the number of output channels to 32 through a convolutional layer with 32 3*3 convolutional kernels.

[0041] Exemplarily, after the feature map output by the first sub-network is feature-enhanced in the second sub-network, it can be input into the third sub-network for object detection and classification. The third sub-network includes: a sixth convolutional layer, a seventh convolutional layer, an eighth convolutional layer, and a ninth convolutional layer; the number of convolutional kernels of the sixth convolutional layer is the same as the number of channels of the fourth convolutional layer.

[0042] Specifically, the above step 102b3 may further include the following step 102b31: Step 102b31: Input the first fused feature map and the second fused feature map into the sixth convolutional layer for feature extraction, and input the output results of the sixth convolutional layer into the seventh convolutional layer, the eighth convolutional layer, and the ninth convolutional layer respectively for predicting the bounding box, category, and confidence, and obtain the encrypted label detection result.

[0043] Among them, the seventh convolutional layer is used to predict the coordinates of the bounding boxes, the eighth convolutional layer is used to predict the probabilities of different category cipher marks, and the ninth convolutional layer is used to predict the confidence scores indicating the presence of objects within each bounding box; the number of output channels of the seventh convolutional layer is obtained based on the product of the number of anchor boxes and the number of coordinates of each anchor box; the number of output channels of the eighth convolutional layer is obtained based on the product of the number of anchor boxes and the number of cipher mark categories that each anchor box can predict; the number of output channels of the ninth convolutional layer is obtained based on the number of anchor boxes.

[0044] Exemplarily, the structure of the above third sub-network specifically includes: using 64 3*3 convolutional layers, with each convolutional layer followed by batch normalization and the ReLU activation function to further extract features, adjust the number of channels, and prepare for the prediction of bounding boxes, categories, and confidence scores. Using 12 1*1 convolutional layers to predict the coordinates of the bounding boxes, with 3 anchor boxes, and each anchor box predicting 4 coordinate values (center point x, y, width w, height h), and the number of output channels being 3 * 4 = 12. Using 18 1*1 convolutional layers to predict the category probabilities, with 3 anchor boxes, and each anchor box predicting 6 categories (top-secret seal, confidential seal, secret seal, top-secret text, confidential text, secret text), and the number of output channels being 3 * 6 = 18. Using 3 1*1 convolutional layers to predict the confidence scores indicating the presence of objects within each anchor box, with 3 anchor boxes, and the number of output channels being 3. Total number of output channels: number of channels for predicting bounding box coordinates + number of channels for predicting category probabilities + number of channels for predicting confidence scores. For 3 anchor boxes and 6 categories, the total number of output channels is 12 (coordinates) + 18 (categories) + 3 (confidence) = 33.

[0045] Exemplarily, in the embodiments of the present application, the prediction results are processed through post-processing steps such as non-maximum suppression (NMS) to remove duplicate detection boxes and improve detection accuracy. An adaptive anchor box mechanism is introduced to automatically adjust the scale and ratio of the anchor boxes to adapt to the shapes and sizes of classified markings in different scenarios; in the second sub-network of the classified marking detection network, a Bidirectional Feature Pyramid Network (BiFPN) structure is embedded to effectively integrate feature information of different scales and improve the detection efficiency of small target classified markings. The GAM mechanism is introduced: in the first sub-sub-network of the classified marking detection network, a Global Attention Mechanism (GAM) is introduced, and some redundant layers are reduced to improve computational efficiency, while improving image noise suppression, enhancing the stability and accuracy of the model in a changing environment, strengthening the attention to key regions of the image, and improving the recognition accuracy of classified markings. The Wise-IoU loss function: The Wise-IoU (Weighted Intersection over Union) loss function is introduced. Compared with the traditional IoU loss function, it pays more attention to the precise matching of bounding boxes, especially in scenarios with complex boundaries and dense targets, enhancing the network's sensitivity to complex boundaries and its ability to capture details, and improving detection accuracy.

[0046] The training process in the embodiments of the present application includes: Dynamic weight adjustment: During the training process, a dynamic weight adjustment strategy is adopted to balance the losses between different categories (such as top secret, secret, confidential), and improve the overall recall rate. Data augmentation: According to the characteristics of classified markings in electronic files, various data augmentation techniques such as rotation, scaling, cropping, and flipping are used to enrich the training data set and improve the generalization ability of the model. Model pruning and quantization: Pruning and quantization processing are performed on the classified marking detection network to reduce the computational amount and memory occupancy of the model, facilitating deployment to resource-constrained devices.

[0047] Exemplarily, after obtaining the classified marking detection results of the classified electronic file, it is also necessary to classify the secret level of the electronic file according to the classified marking detection results. The secret results include: top secret, secret, confidential, etc.

[0048] Specifically, step 102 above may further include the following step 102c: Step 102c: Classify the classified marking detection results, determine the category of the identified classified marking, and obtain the classified marking classification result.

[0049] Exemplarily, the classified marking classification results in the embodiments of the present application may further include: the location, category, confidence level of the classified marking, as well as relevant security suggestions or handling measures.

[0050] The electronic file confidential label detection method provided by the embodiments of this application first obtains an electronic file image containing a confidential label, and preprocesses the electronic file image using a preset processing method to obtain a confidential image to be recognized. Then, the confidential image to be recognized is input into a confidential label detection network to obtain a confidential label detection result, and the confidential label is classified based on the confidential label detection result to obtain a confidential label classification result. Among them, the confidential label detection network includes: a first sub-network, a second sub-network, and a third sub-network. The first sub-network is used to extract feature information from the confidential image to be recognized. The second sub-network is used to optimize the feature representation of the feature information extracted by the first sub-network and enhance the expression ability of the features. The third sub-network is used to predict the bounding box, category, and confidence based on the feature information output by the second sub-network. In this way, it can adapt to scenarios with different complexities and interference levels, and achieve rapid detection and classification of large-scale electronic files with low resource consumption.

[0051] It should be noted that for the electronic file confidential label detection method provided by the embodiments of this application, the execution subject can be an electronic file confidential label detection device, or a control module in the electronic file confidential label detection device for executing the electronic file confidential label detection method. In the embodiments of this application, the example of the electronic file confidential label detection device executing the electronic file confidential label detection method is used to illustrate the electronic file confidential label detection device provided by the embodiments of this application.

[0052] It should be noted that in the embodiments of this application, the electronic file confidential label detection methods shown in the above respective method drawings are all exemplarily illustrated by taking one drawing in the embodiments of this application as an example. Specifically in implementation, the electronic file confidential label detection methods shown in the above respective method drawings can also be implemented in combination with any other combinable drawings schemed in the above embodiments, which will not be elaborated here.

[0053] The electronic file confidential label detection device provided by this application will be described below, and the following description can be mutually referred to corresponding to the electronic file confidential label detection method described above.

[0054] Figure 3 is a structural schematic diagram of the electronic file confidential label detection device provided by the embodiments of this application, as Figure 3 shown, and specifically includes: The image processing module 301 is configured to obtain an electronic file image containing a classified label, and preprocess the electronic file image using a preset processing method to obtain a classified image to be recognized; the classified label detection module 302 is configured to input the classified image to be recognized into a classified label detection network to obtain a classified label detection result, and classify the classified label based on the classified label detection result to obtain a classified label classification result; wherein, the classified label detection network includes: a first sub-network, a second sub-network, and a third sub-network; the first sub-network is configured to extract feature information from the classified image to be recognized; the second sub-network is configured to optimize the feature representation of the feature information extracted by the first sub-network and enhance the expression ability of the features; the third sub-network is configured to predict a bounding box, a category, and a confidence level based on the feature information output by the second sub-network.

[0055] Optionally, the first sub-network includes: a first convolutional layer, a second convolutional layer, and a third convolutional layer; the first convolutional layer includes: a first preset number of convolutional kernels, and the size of the output feature map is a first preset size; the second convolutional layer includes: a second preset number of convolutional kernels, and the size of the output feature map is a second preset size; the third convolutional layer includes: a third preset number of convolutional kernels, and the size of the output feature map is a third preset size; the first convolutional layer, the second convolutional layer, and the third convolutional layer have the same stride; the classified label detection module 302 is specifically configured to input the classified image to be recognized into the first sub-network to obtain a first feature map output by the first convolutional layer, a second feature map output by the second convolutional layer, and a third feature map output by the third convolutional layer; wherein, the size of the first feature map is the first preset size, the size of the second feature map is the second preset size, and the size of the third feature map is the third preset size.

[0056] Optionally, the second sub-network includes: a fourth convolutional layer and a fifth convolutional layer; the secret label detection module 302 is specifically configured to upsample the third feature map by the second sub-network to obtain a fourth feature map, and splice and fuse the fourth feature map with the second feature map, and adjust the number of channels of the spliced and fused feature map through the fourth convolutional layer to obtain a first fused feature map; the fourth feature map has the same size as the second feature map; the secret label detection module 302 is specifically further configured to upsample the first fused feature map to obtain a sixth feature map, and splice and fuse the sixth feature map with the first feature map, and adjust the number of channels of the spliced and fused feature map through the fifth convolutional layer to obtain a second fused feature map; the secret label detection module 302 is specifically further configured to input the first fused feature map and the second fused feature map into the third sub-network to predict the bounding box, category, and confidence, and obtain the secret label detection result; wherein, the sixth feature map has the same size as the first feature map; the number of output channels of the fourth convolutional layer is twice the number of output channels of the fifth convolutional layer.

[0057] Optionally, the third sub-network includes: a sixth convolutional layer, a seventh convolutional layer, an eighth convolutional layer, and a ninth convolutional layer; the number of convolutional kernels of the sixth convolutional layer is the same as the number of channels of the fourth convolutional layer; the secret label detection module 302 is specifically configured to input the first fused feature map and the second fused feature map into the sixth convolutional layer for feature extraction, and input the output result of the sixth convolutional layer into the seventh convolutional layer, the eighth convolutional layer, and the ninth convolutional layer respectively for predicting the bounding box, category, and confidence, and obtain the secret label detection result; wherein, the seventh convolutional layer is used to predict the coordinates of the bounding box, the eighth convolutional layer is used to predict the probabilities of different category secret labels, and the ninth convolutional layer is used to predict the confidence score that there is a target in each bounding box; the number of output channels of the seventh convolutional layer is obtained based on the product of the number of anchor boxes and the number of coordinates of each anchor box; the number of output channels of the eighth convolutional layer is obtained based on the product of the number of anchor boxes and the number of secret label categories that each anchor box can predict; the number of output channels of the ninth convolutional layer is obtained based on the number of anchor boxes.

[0058] The electronic file security label detection device provided by this application first obtains an electronic file image containing a security label, and preprocesses the electronic file image using a preset processing method to obtain a to-be-recognized security image. Then, the to-be-recognized security image is input into a security label detection network to obtain a security label detection result, and the security label is classified based on the security label detection result to obtain a security label classification result. Among them, the security label detection network includes: a first sub-network, a second sub-network, and a third sub-network. The first sub-network is used to extract feature information from the to-be-recognized security image. The second sub-network is used to optimize the feature representation of the feature information extracted by the first sub-network and enhance the expression ability of the features. The third sub-network is used to predict the bounding box, category, and confidence based on the feature information output by the second sub-network. In this way, it can adapt to scenarios with different complexities and interference levels, and achieve fast detection and classification of large-scale electronic files with low resource consumption.

[0059] Figure 4 An example of the physical structure diagram of an electronic device is shown as Figure 4 shown. The electronic device may include: a processor 410, a communication interface 420, a memory 430, and a communication bus 440. Among them, the processor 410, the communication interface 420, and the memory 430 communicate with each other through the communication bus 440. The processor 410 can call the logical instructions in the memory 430 to execute the electronic file security label detection method, which includes: first, obtaining an electronic file image containing a security label, and preprocessing the electronic file image using a preset processing method to obtain a to-be-recognized security image. Then, the to-be-recognized security image is input into a security label detection network to obtain a security label detection result, and the security label is classified based on the security label detection result to obtain a security label classification result. Among them, the security label detection network includes: a first sub-network, a second sub-network, and a third sub-network. The first sub-network is used to extract feature information from the to-be-recognized security image. The second sub-network is used to optimize the feature representation of the feature information extracted by the first sub-network and enhance the expression ability of the features. The third sub-network is used to predict the bounding box, category, and confidence based on the feature information output by the second sub-network. In this way, it can adapt to scenarios with different complexities and interference levels, and achieve fast detection and classification of large-scale electronic files with low resource consumption.

[0060] In addition, when the logical instructions in the above-mentioned memory 430 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of this application. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs.

[0061] On the other hand, this application also provides a computer program product. The computer program product includes a computer program stored on a computer-readable storage medium. The computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the electronic file secret label detection method provided by the above-mentioned various methods. The method includes: First, obtain an electronic file image containing a classified label, and use a preset processing method to preprocess the electronic file image to obtain a to-be-recognized classified image; After that, input the to-be-recognized classified image into a secret label detection network to obtain a secret label detection result, and classify the classified label based on the secret label detection result to obtain a secret label classification result; Among them, the secret label detection network includes: a first sub-network, a second sub-network, and a third sub-network; The first sub-network is used to extract feature information from the to-be-recognized classified image; The second sub-network is used to optimize the feature representation of the feature information extracted by the first sub-network and enhance the expression ability of the features; The third sub-network is used to predict the bounding box, category, and confidence based on the feature information output by the second sub-network. In this way, it can adapt to scenarios with different complexities and interference levels and achieve the rapid detection and classification of large-scale electronic files with low resource consumption.

[0062] In another aspect, the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is configured to execute the above-provided electronic file secret label detection method, which includes: First, obtain an electronic file image containing a classified label, and preprocess the electronic file image using a preset processing method to obtain a to-be-recognized classified image; then, input the to-be-recognized classified image into a secret label detection network to obtain a secret label detection result, and classify the classified label based on the secret label detection result to obtain a secret label classification result; wherein, the secret label detection network includes: a first sub-network, a second sub-network, and a third sub-network; the first sub-network is used to extract feature information from the to-be-recognized classified image; the second sub-network is used to optimize the feature representation of the feature information extracted by the first sub-network and enhance the expression ability of the features; the third sub-network is used to predict the bounding box, category, and confidence based on the feature information output by the second sub-network. In this way, it is possible to adapt to scenarios with different complexities and interference levels, and achieve rapid detection and classification of large-scale electronic files with low resource consumption.

[0063] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0064] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the essence of the above technical solutions, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disc, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0065] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for detecting electronic archive secret labels, characterized in that: include: Acquire an electronic file image containing a confidential mark, and pre-process the electronic file image using a preset processing method to obtain a confidential image to be identified; Inputting the confidential image to be identified into a confidential mark detection network to obtain a confidential mark detection result, and classifying the confidential mark based on the confidential mark detection result to obtain a confidential mark classification result; Among them, the secret mark detection network includes: a first subnetwork, a second subnetwork and a third subnetwork; the first subnetwork is used to extract feature information in the confidential image to be identified; the second subnetwork is used to optimize the feature representation of the feature information extracted by the first subnetwork and enhance the feature expression ability; the third subnetwork is used to predict the bounding box, category and confidence based on the feature information output by the second subnetwork.

2. The method according to claim 1, characterized in that The first sub-network includes: a first convolutional layer, a second convolutional layer and a third convolutional layer; the first convolutional layer includes: a first preset number of convolutional kernels, and the size of the output feature map is the first preset size; the second convolutional layer includes: a second preset number of convolutional kernels, and the size of the output feature map is the second preset size; the third convolutional layer includes: a third preset number of convolutional kernels, and the size of the output feature map is the third preset size; the first convolutional layer, the second convolutional layer and the third convolutional layer have the same step size; The step of inputting the to-be-identified confidential image into a confidential mark detection network to obtain a confidential mark detection result includes: Inputting the confidential image to be identified into the first sub-network to obtain a first feature map output by the first convolution layer, a second feature map output by the second convolution layer, and a third feature map output by the third convolution layer; Among them, the size of the first feature map is the first preset size, the size of the second feature map is the second preset size, and the size of the third feature map is the third preset size.

3. The method according to claim 2, characterized in that The second sub-network includes: a fourth convolutional layer and a fifth convolutional layer; The step of inputting the to-be-identified confidential image into a confidential mark detection network to obtain a confidential mark detection result includes: The second sub-network upsamples the third feature map to obtain a fourth feature map, concatenates and fuses the fourth feature map with the second feature map, and adjusts the number of channels of the concatenated and fused feature map through a fourth convolutional layer to obtain a first fused feature map; the fourth feature map has the same size as the second feature map; The first fused feature map is upsampled to obtain a sixth feature map, the sixth feature map is concatenated and fused with the first feature map, and the number of channels of the concatenated and fused feature map is adjusted through a fifth convolutional layer to obtain a second fused feature map; Inputting the first fused feature map and the second fused feature map into the third sub-network to predict the bounding box, category and confidence, and obtaining the secret mark detection result; The sixth feature map has the same size as the first feature map; and the number of output channels of the fourth convolutional layer is twice the number of output channels of the fifth convolutional layer.

4. The method according to claim 3, characterized in that The third sub-network includes: a sixth convolutional layer, a seventh convolutional layer, an eighth convolutional layer and a ninth convolutional layer; the number of convolutional kernels of the sixth convolutional layer is the same as the number of channels of the fourth convolutional layer; The step of inputting the first fused feature map and the second fused feature map into the third sub-network to predict the bounding box, category and confidence, and obtaining the secret mark detection result includes: Inputting the first fused feature map and the second fused feature map into the sixth convolutional layer for feature extraction, and inputting the output result of the sixth convolutional layer into the seventh convolutional layer, the eighth convolutional layer, and the ninth convolutional layer respectively for prediction of bounding box, category, and confidence, and obtaining the secret mark detection result; Among them, the seventh convolution layer is used to predict the coordinates of the bounding box, the eighth convolution layer is used to predict the probability of different categories of secret labels, and the ninth convolution layer is used to predict the confidence score of the existence of the target in each bounding box; the output channel of the seventh convolution layer is obtained based on the product of the number of anchor boxes and the number of coordinates of each anchor box; the output channel of the eighth convolution layer is obtained based on the product of the number of anchor boxes and the number of secret label categories that can be predicted by each anchor frame; the output channel of the ninth convolution layer is obtained based on the number of anchor boxes.

5. The method according to any one of claims 1 to 4, characterized in that The classifying the confidential identification based on the secret mark detection result to obtain the secret mark classification result includes: The secret mark detection result is classified to determine the category of the secret mark to obtain the secret mark classification result.

6. An electronic file secret mark detection device, characterized in that: The device comprises: An image processing module is used to obtain an electronic file image containing a confidential mark, and pre-process the electronic file image using a preset processing method to obtain a confidential image to be identified; A secret mark detection module, used for inputting the secret-related image to be identified into a secret mark detection network to obtain a secret mark detection result, and classifying the secret-related mark based on the secret mark detection result to obtain a secret mark classification result; Among them, the secret mark detection network includes: a first subnetwork, a second subnetwork and a third subnetwork; the first subnetwork is used to extract feature information in the confidential image to be identified; the second subnetwork is used to optimize the feature representation of the feature information extracted by the first subnetwork and enhance the feature expression ability; the third subnetwork is used to predict the bounding box, category and confidence based on the feature information output by the second subnetwork.

7. The device according to claim 6, characterized in that The first sub-network includes: a first convolutional layer, a second convolutional layer and a third convolutional layer; the first convolutional layer includes: a first preset number of convolutional kernels, and the size of the output feature map is the first preset size; the second convolutional layer includes: a second preset number of convolutional kernels, and the size of the output feature map is the second preset size; the third convolutional layer includes: a third preset number of convolutional kernels, and the size of the output feature map is the third preset size; the first convolutional layer, the second convolutional layer and the third convolutional layer have the same step size; The secret mark detection module is specifically used to input the secret image to be identified into the first sub-network to obtain the first feature map output by the first convolution layer, the second feature map output by the second convolution layer, and the third feature map output by the third convolution layer; Among them, the size of the first feature map is the first preset size, the size of the second feature map is the second preset size, and the size of the third feature map is the third preset size.

8. The device according to claim 7, characterized in that The second sub-network includes: a fourth convolutional layer and a fifth convolutional layer; The secret mark detection module is specifically used for the second sub-network to upsample the third feature map to obtain a fourth feature map, and to splice and fuse the fourth feature map with the second feature map, and to adjust the number of channels of the spliced ​​and fused feature map through a fourth convolutional layer to obtain a first fused feature map; the fourth feature map has the same size as the second feature map; The secret mark detection module is further configured to obtain a sixth feature map after upsampling the first fused feature map, concatenate and fuse the sixth feature map with the first feature map, and adjust the number of channels of the concatenated and fused feature map through a fifth convolutional layer to obtain a second fused feature map; The secret mark detection module is further configured to input the first fused feature map and the second fused feature map into the third sub-network to predict the bounding box, category and confidence, and obtain the secret mark detection result; The sixth feature map has the same size as the first feature map; and the number of output channels of the fourth convolutional layer is twice the number of output channels of the fifth convolutional layer.

9. An electronic device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the electronic archive secret mark detection method as claimed in any one of claims 1 to 5 are implemented.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor, the steps of the electronic archive secret mark detection method as claimed in any one of claims 1 to 5 are implemented.

Images