CN113435424B - Method and system for identifying destroying granularity of confidential medium - Google Patents
Method and system for identifying destroying granularity of confidential medium Download PDFInfo
- Publication number
- CN113435424B CN113435424B CN202110991513.3A CN202110991513A CN113435424B CN 113435424 B CN113435424 B CN 113435424B CN 202110991513 A CN202110991513 A CN 202110991513A CN 113435424 B CN113435424 B CN 113435424B
- Authority
- CN
- China
- Prior art keywords
- medium
- fragment
- confidential
- detected
- network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The invention relates to a method and a system for identifying the destroying granularity of a confidential medium. The identification method for the destroy granularity of the confidential medium obtains the medium type and the mask of the crushed medium fragments by adopting a trained medium fragment identification model, then calculates the pixel area of the corresponding medium fragments in the image according to the mask, determines whether the crushed medium fragments are unqualified fragments according to the pixel area and the medium type of the medium fragments, and then determines whether the crushing degree meets the requirement of the destroy granularity of the medium according to the number of the unqualified fragments and the total number of the crushed medium fragments, thereby realizing the judgment of the crushing effect of the confidential medium fragments.
Description
Technical Field
The invention relates to the technical field of confidential medium destruction granularity identification, in particular to a confidential medium destruction granularity identification method and system.
Background
With the rapid development of information technology and the acceleration of informatization construction in various fields, more and more information security means are widely applied to the military and civil fields. Once a loss or compromise of the confidential medium occurs, immeasurable losses in the security and interests of individuals, enterprises and even countries occur, and thus the destruction of the confidential medium to the scientific norms which end the life cycle is critical. Mechanical pulverization is the most common and effective means for destroying confidential media currently, but a reliable method for identifying the granularity of the media is lacking in the media pulverization process at the present stage, so that it is difficult to ensure that information carried by the confidential media is effectively destroyed, and thus the information security faces a great threat (data self-destruction technology [ J ] in data security system of dawn. electronic technology and software engineering, 2021, {4} (02): 249-250).
Currently, mechanical shredding relies primarily on manual visual methods to determine the size of the shredded classified media to determine if secondary shredding is required. Or the reliability of medium destruction is improved by manually setting the mechanical crushing time. The existing method not only needs to consume a large amount of time and energy, but also is easy to be influenced by subjective factors (Yan national Qing, permission to be clear, safe destruction method of information storage media and resource technology [ J ] material guide, 2013,27(03):12-17+ 31.).
Therefore, there is a need in the art for a method or system for identifying the destruction granularity of confidential media, so as to determine the mechanical crushing effect of the media by quickly and accurately identifying the size of the media fragments.
Disclosure of Invention
The invention aims to provide a method and a system for identifying the destroying granularity of confidential media, which can quickly and accurately identify the destroying granularity of the confidential media and realize the accurate judgment of the mechanical crushing effect of the media.
In order to achieve the purpose, the invention provides the following scheme:
a method of identifying the granularity at which a security media is destroyed, comprising:
acquiring an image of a confidential medium fragment to be detected;
inputting the image of the confidential medium fragments to be detected into a trained medium fragment recognition model to obtain a mask of each confidential medium fragment to be detected and a medium class to which each confidential medium fragment to be detected belongs; the medium fragment recognition model includes: a backbone network, a multi-scale feature fusion network, a regional suggestion network and a detection head;
determining an area threshold according to the medium category of each confidential medium fragment to be detected;
determining the area of a pixel according to the mask of each confidential medium fragment to be detected;
judging whether the pixel area is larger than the area threshold value or not to obtain a first judgment result;
when the first judgment result shows that the pixel area is larger than the area threshold, judging that the confidential medium fragments to be detected corresponding to the mask are unqualified fragments;
when the first judgment result is that the pixel area is smaller than or equal to the area threshold, judging that the confidential medium fragments to be detected corresponding to the mask are qualified fragments;
determining the proportion of the number of the unqualified fragments to the total number of the confidential medium fragments to be detected;
judging whether the proportion is larger than a preset proportion threshold value or not to obtain a second judgment result;
when the second judgment result is that the ratio is larger than a preset ratio threshold, judging that the image of the confidential medium fragment to be detected does not meet the requirement of medium destruction granularity;
and when the second judgment result is that the ratio is smaller than or equal to a preset ratio threshold, judging that the confidential medium fragments to be detected meet the requirement on medium destruction granularity.
Preferably, the construction process of the medium fragment identification model comprises the following steps:
taking a convolutional neural network ResNet50 as a backbone network; the convolutional neural network ResNet50 is composed of a plurality of convolutional blocks connected in sequence;
constructing a multi-scale feature fusion network based on the backbone network; the multi-scale feature fusion network comprises a multi-level pyramid network;
constructing a regional suggestion network based on the multi-scale feature fusion network;
and constructing a detection head based on the area suggestion network.
Preferably, the constructing a multi-scale feature fusion network based on the backbone network specifically includes:
respectively inputting the first characteristic diagram output by each convolution block in the backbone network into a convolution layer with a convolution kernel size of 1 x 1;
after down-sampling operation and up-sampling operation are carried out on the second feature graph output by the first convolutional layer, up-sampling operation and element set addition operation are carried out on the third feature graphs output by the rest convolutional layers in sequence to obtain a fourth feature graph;
and respectively inputting the fourth feature maps into a CBAM attention module to obtain a multi-level pyramid network.
Preferably, the constructing a regional suggestion network based on the multi-scale feature fusion network specifically includes:
performing convolution operation with convolution kernel of 3 × 3 on the fifth feature map output by the multi-level pyramid network, and performing traversal operation on the fifth feature map by adopting a sliding anchor frame in the convolution operation process to generate a group of candidate fragment areas;
inputting the seventh feature map for generating the subsequent fragment area into a classification branch to predict the bounding box position of the candidate fragment area;
and inputting the seventh feature map for generating the subsequent fragment region into the regression branch to predict the probability that the candidate fragment region is the medium fragment.
Preferably, the constructing a detection head based on the area recommendation network specifically includes:
mapping each candidate fragment area to a corresponding feature layer according to the area size, and outputting a series of candidate area feature maps with the same size through an ROI Align layer;
inputting the candidate region feature map into a detection branch, inputting the candidate region feature map into a regression branch after feature extraction and classification are carried out on the candidate region feature map through a first full-connection layer in the detection branch, and completing frame regression operation by utilizing a regression Loss function Smooth L1 Loss to obtain position information of the medium fragments; after the characteristic map of the candidate region is subjected to characteristic extraction and classification through a second full connection layer, inputting the characteristic map into a classification branch, and performing background and fragment class classification by using a classification Loss function Softmax Loss to determine the medium class to which the candidate fragment region belongs;
and inputting the candidate region characteristic graph into a mask branch, and in the mask branch, after the candidate region characteristic graph is subjected to characteristic extraction and classification through a third convolution layer and a fourth convolution layer, obtaining a mask corresponding to each medium fragment by using average binary cross entropy loss.
Preferably, the convolutional neural network ResNet50 is composed of 5 convolutional blocks connected in sequence.
Preferably, the training and testing process of the medium fragment recognition model includes:
acquiring a media crushing image segmentation dataset;
dividing the medium crushing image segmentation data set into a training set and a test set according to a set proportion;
after the training set is adopted to train the medium fragment recognition model, a test set is adopted to test the trained medium fragment recognition model, and in the test process, when the trained medium fragment recognition model meets a preset condition, a trained medium fragment recognition model is obtained; and when the trained medium fragment recognition model does not meet the preset condition, re-training the medium fragment recognition model.
Preferably, the acquiring the media reduction image segmentation dataset further comprises:
acquiring fragment images of various smashed confidential media;
labeling the fragment images by adopting Labelme software to obtain a labeling file corresponding to each fragment image; the format of the markup file is JSON format;
compiling a script, and converting the format of the markup file into an XML format to obtain a medium crushing image data set;
a media-shredding image-segmentation dataset in the VOC2007 dataset format is obtained based on the media-shredding image dataset and the annotation file.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
the method for identifying the destroying granularity of the confidential medium obtains the medium type and the mask of the crushed medium fragments by adopting a trained medium fragment identification model, then calculates the pixel area of the corresponding medium fragments in an image according to the mask, determines whether the crushed medium fragments are unqualified fragments according to the pixel area and the medium type of the medium fragments, and then determines whether the crushing degree meets the requirement of the medium destroying granularity according to the number of the unqualified fragments and the total number of the crushed medium fragments, thereby realizing the judgment of the crushing effect of the confidential medium fragments.
Corresponding to the identification method for the destroying granularity of the confidential medium, the invention also provides the following implementation system:
a system for identifying the granularity of destruction of a security medium, comprising:
the image acquisition module is used for acquiring an image of the confidential medium fragment to be detected;
the mask-medium type identification module is used for inputting the image of the confidential medium fragments to be detected into a trained medium fragment identification model to obtain a mask of each confidential medium fragment to be detected and a medium type to which each confidential medium fragment to be detected belongs; the medium fragment recognition model includes: a backbone network, a multi-scale feature fusion network, a regional suggestion network and a detection head;
the area threshold determining module is used for determining an area threshold according to the medium category of each confidential medium fragment to be detected;
the pixel area determining module is used for determining the pixel area according to the mask of each confidential medium fragment to be detected;
the first judgment module is used for judging whether the pixel area is larger than the area threshold value or not to obtain a first judgment result;
an unqualified fragment determining module, configured to determine that the to-be-detected confidential medium fragment corresponding to the mask is an "unqualified fragment" when the first determination result indicates that the pixel area is larger than the area threshold;
a qualified fragment determining module, configured to determine that the confidential medium fragment to be detected corresponding to the mask is a "qualified fragment" when the first determination result indicates that the pixel area is smaller than or equal to the area threshold;
the proportion determining module is used for determining the proportion of the number of the unqualified fragments in the total number of the confidential medium fragments to be detected;
the second judgment module is used for judging whether the proportion is larger than a preset proportion threshold value or not to obtain a second judgment result;
the first crushing result determining module is used for judging that the image of the confidential medium fragment to be detected does not meet the requirement of medium destruction granularity when the second judgment result is that the ratio is greater than a preset ratio threshold;
and the second crushing result determining module is used for judging that the confidential medium fragments to be detected meet the requirement on medium destruction granularity when the second judgment result shows that the proportion is smaller than or equal to a preset proportion threshold.
The technical effect achieved by the identification system for the destroying granularity of the confidential medium provided by the invention is the same as that achieved by the identification method for the destroying granularity of the confidential medium, so that the details are not repeated herein.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
FIG. 1 is a flow chart of a method for identifying the destruction granularity of a confidential medium provided by the invention;
FIG. 2 is a schematic structural diagram of a medium fragment identification model according to an embodiment of the present invention;
FIG. 3 is a schematic structural diagram of a security media destruction granularity identification system according to the present invention;
fig. 4 is a schematic structural diagram of another security media destruction granularity identification system provided by the invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention aims to provide a method and a system for identifying the destroying granularity of confidential media, which can quickly and accurately identify the destroying granularity of the confidential media and realize the accurate judgment of the mechanical crushing effect of the media.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
As shown in fig. 1, the present invention provides a method for identifying the destruction granularity of a security medium, including:
step 100: an image of the piece of the confidential medium to be detected is acquired.
Step 101: and inputting the image of the confidential medium fragment to be detected into the trained medium fragment recognition model to obtain the mask of each confidential medium fragment to be detected and the medium class to which each confidential medium fragment to be detected belongs. The medium fragment recognition model comprises: the system comprises a backbone network, a multi-scale feature fusion network, a regional suggestion network and a detection head.
Step 102: an area threshold is determined based on the media type of each of the secure media fragments to be detected.
Step 103: the pixel area is determined based on the mask for each piece of the security media to be detected.
Step 104: and judging whether the area of the pixel is larger than an area threshold value or not to obtain a first judgment result.
Step 105: and when the first judgment result is that the pixel area is larger than the area threshold, judging that the confidential medium fragment to be detected corresponding to the mask is an unqualified fragment.
Step 106: and when the first judgment result is that the pixel area is smaller than or equal to the area threshold, judging that the confidential medium fragment to be detected corresponding to the mask is a qualified fragment.
Step 107: determining a proportion of the number of "rejected pieces" to the total number of pieces of the security media to be detected.
Step 108: and judging whether the proportion is larger than a preset proportion threshold value or not to obtain a second judgment result.
Step 109: and when the second judgment result is that the proportion is larger than the preset proportion threshold value, judging that the image of the confidential medium fragment to be detected does not meet the requirement of medium destruction granularity.
Step 110: and when the second judgment result is that the proportion is smaller than or equal to the preset proportion threshold, judging that the confidential medium fragments to be detected meet the requirement on the medium destruction granularity.
The construction process of the medium fragment identification model adopted in the step 101 includes:
A. the convolutional neural network ResNet50 is used as a backbone network. The convolutional neural network ResNet50 is composed of a plurality of sequentially connected convolutional blocks. The convolutional neural network ResNet50 may be composed of 5 convolutional blocks connected in sequence. The backbone network can also be built by adopting other neural network models.
B. And constructing a multi-scale feature fusion network based on the backbone network. The multi-scale feature fusion network comprises a multi-level pyramid network. The specific implementation process of the step can be as follows:
and respectively inputting the first characteristic diagram output by each convolution block in the backbone network into a convolution layer with the convolution kernel size of 1 x 1.
And after down-sampling operation and up-sampling operation are carried out on the second feature map output by the first convolutional layer, up-sampling operation and element set addition operation are carried out on the third feature maps output by the rest convolutional layers in sequence to obtain a fourth feature map.
And respectively inputting the fourth feature maps into a CBAM attention module to obtain a multi-level pyramid network.
C. And constructing a regional suggestion network based on the multi-scale feature fusion network. The specific implementation process of the process can be as follows:
and respectively carrying out convolution operation with convolution kernel of 3 multiplied by 3 on the fifth feature map output by the multi-level pyramid network, and carrying out traversal operation on the fifth feature map by adopting a sliding anchor frame in the convolution operation process to generate a group of candidate fragment areas.
And inputting the seventh feature map for generating the subsequent fragment area into the classification branch to predict the position of the surrounding box of the candidate fragment area.
And inputting the seventh feature map for generating the subsequent fragment region into the regression branch to predict the probability that the candidate fragment region is the medium fragment.
D. And constructing a detection head based on the area suggestion network. The specific implementation process of the step can be as follows:
and mapping each candidate fragment area to a corresponding feature layer according to the area size, and outputting a series of candidate area feature maps with the same size through an ROI Align layer.
And inputting the candidate region characteristic graph into a detection branch, wherein in the detection branch, after the candidate region characteristic graph is subjected to characteristic extraction and classification through a first full connection layer, the candidate region characteristic graph is input into a regression branch, and a frame regression operation is completed by utilizing a regression Loss function Smooth L1 Loss to obtain the position information of the medium fragment. And after the feature extraction and classification are carried out on the candidate region feature map through the second full-connection layer, inputting the candidate region feature map into a classification branch, carrying out the classification of the background and the fragments by using a classification Loss function Softmax Loss, and determining the medium class to which the candidate fragment region belongs.
And inputting the candidate region feature map into a mask branch, wherein in the mask branch, after feature extraction and classification are carried out on the candidate region feature map through a third convolution layer and a fourth convolution layer, a mask corresponding to each medium fragment is obtained by using average binary cross entropy loss.
The concrete structure of the medium fragment identification model obtained through the establishing process of the steps is shown in fig. 2.
The following shows the design process of the convolutional neural network by taking an input medium pulverization image 1024 × 1024 as an example:
here, ResNet50 is used as a backbone network for extracting media fragmentation characteristics, ResNet50 is composed of five convolution blocks (conv1, conv2, conv3, conv4, conv5) connected in sequence, and output feature maps of the convolution blocks are respectively represented as C1, C2, C3, C4, and C5, and feature map sizes thereof are 512 × 512 × 128, 256 × 256 × 256, 128 × 128 × 512, 64 × 64 × 1024, and 32 × 32 × 2048 in sequence.
Next, in order to improve the detection capability of the network on different-scale fragments, especially small-area fragments, a multi-scale feature fusion network is constructed. First, in order to unify the number of feature map channels, when C1 to C5 are input to one convolution layer having a convolution kernel size of 1 × 1, the feature map sizes are 512 × 512 × 256, 256 × 256 × 256, 128 × 128 × 256, 64 × 64 × 256, and 32 × 32 × 256. Then, as shown in fig. 2, the output characteristic diagram corresponding to C5 is named F5, and the F5 size is 32 × 32 × 256. Down-sampling of F5 by a factor of 0.5 was performed to obtain F6 of size 16 × 16 × 256. Then, F5 is subjected to 2-fold upsampling operation, the size of the upsampling operation is enlarged to 2-fold that of the original size, namely 64 × 64 × 256, and the upsampling operation is subjected to element-level addition with the output characteristic map corresponding to C4 with equal size, so that F4 is obtained. The above feature fusion operations, i.e., upsampling and element-level summing, are repeated sequentially for F4, F3, and F2, resulting in F3, F2, and F1 feature layers. In order to eliminate redundant features and refine key features in the feature fusion process, the F1-F6 feature maps are respectively input into a CBAM attention module, and therefore G1-G6 layers are obtained. The sizes of the G1-G6 layers are as follows in sequence: 512 × 512 × 256, 256 × 256 × 256, 128 × 128 × 256, 64 × 64 × 256, 32 × 32 × 256, and 16 × 16 × 256. The constructed multi-scale feature fusion network can enable each feature layer to obtain rich multi-scale context feature information by constructing a multi-level pyramid network, and the identification capability of an identification module on medium fragments with different sizes is enhanced.
Then, the area proposal network is constructed. The concrete structure is as follows: currently, a feature map layer of a pyramid structure is obtained based on a backbone network ResNet50 and a multi-scale feature fusion network: g1, G2, G3, G4, G5 and G6, taking G1-G6 as input feature maps of the region suggestion network, firstly, carrying out convolution operation with a convolution kernel of 3 x 3, and carrying out traversal operation on six feature maps by adopting a sliding anchor box in the process to generate a group of candidate fragment regions. And respectively inputting the medium fragments into a classification branch and a regression branch, predicting the probability of the medium fragments in a candidate region in the classification branch, and predicting the position of an enclosure frame of the candidate region in the regression branch. When the network is proposed in the training area, the candidate area with the area intersection ratio of the real target frame being more than 0.7 is judged as a fragment, and the candidate area with the area intersection ratio being less than 0.3 is judged as a background.
Finally, a fragment detection head is constructed. Firstly, mapping each candidate fragment area to a corresponding characteristic layer G according to the area sizek(k =1~ 6), a series of candidate region feature maps with the same size are output through the ROI Align layer, and the operation is to unify the size of the candidate region feature maps so as to input the subsequent fully-connected layer. Next, the candidate region feature maps are input into the detection branch and the mask branch, respectively. In the detection branch, after the feature map is subjected to feature extraction and classification through two full-connection layers, two detection branches (a regression branch and a classification branch) are respectively input: and (4) classifying the background and the fragment by using a classification Loss function Softmax Loss, and determining the medium class to which the candidate fragment region belongs. And finishing the frame regression operation by using a regression Loss function Smooth L1 Loss, and obtaining the position information of the fragments. In the mask branch, after feature extraction and classification are carried out on the feature map through two convolutional layers, the mask corresponding to each fragment is obtained by utilizing average binary cross entropy loss. The loss function value of the whole convolution neural network is obtained by adding the classification loss function, the regression loss function and the mask loss function.
And finishing the overall design of the convolutional neural network based on the processes, performing model training by adopting a medium crushing image segmentation data set, updating parameters of the whole network based on a loss function, and obtaining a final medium identification model after the training is finished.
After a medium fragment recognition model is constructed, model training is carried out by adopting a medium crushing image segmentation data set, and parameter updating is carried out on the whole network based on a loss function so as to ensure the accuracy of an output result. The training and testing process of the medium fragment recognition model comprises the following steps:
A. a media crash image segmentation dataset is acquired. The acquired medium crushing image segmentation data set is constructed by the following steps:
and acquiring fragment images of various smashed confidential media. In the process of acquiring the fragment image, the invention is based on that an industrial camera shoots and acquires the fragment image after crushing a plurality of confidential media, the storage format is JPEG, and the focal distance and the object distance of the camera in the acquisition process are required to be kept unchanged.
And labeling the fragment images by adopting Labelme software to obtain a labeling file corresponding to each fragment image. The format of the markup file is JSON format. The labeling is mainly performed to mark specific shape and outline of each fragment and a corresponding type label (such as a paper document, an optical disc, a card and the like) in each medium fragment image.
And compiling a script, and converting the format of the marked file into an XML format to obtain a medium crushed image data set.
A media-shredded image segmentation dataset in the VOC2007 dataset format is obtained based on the media-shredded image dataset and the annotation file.
B. And dividing the medium crushing image segmentation data set into a training set and a test set according to a set proportion. For example, the training set is divided by the test set at a ratio of 8: 2.
C. And after the training set is adopted to train the medium fragment recognition model, the test set is adopted to test the trained medium fragment recognition model, and in the test process, when the trained medium fragment recognition model meets the preset conditions, the trained medium fragment recognition model is obtained. And when the trained medium fragment recognition model does not meet the preset condition, re-training the medium fragment recognition model.
Corresponding to the identification method for the destroying granularity of the confidential medium, the invention also provides the following two implementation systems:
as shown in fig. 3, an identification system for the destruction granularity of a security medium includes: an image acquisition module 300, a mask-media class identification module 301, an area threshold determination module 302, a pixel area determination module 303, a first judgment module 304, an unqualified fragment determination module 305, a qualified fragment determination module 306, a proportion determination module 307, a second judgment module 308, a first shredding result determination module 309, and a second shredding result determination module 310.
The image acquisition module 300 is configured to acquire an image of a confidential medium fragment to be detected.
The mask-medium type identification module 301 is configured to input the image of the confidential medium fragments to be detected into the trained medium fragment identification model, so as to obtain a mask of each confidential medium fragment to be detected and a medium type to which each confidential medium fragment to be detected belongs. The medium fragment recognition model comprises: the system comprises a backbone network, a multi-scale feature fusion network, a regional suggestion network and a detection head.
The area threshold determination module 302 is configured to determine an area threshold based on the media type of each of the secure media fragments to be detected.
The pixel area determination module 303 is configured to determine a pixel area according to the mask of each to-be-detected classified medium fragment.
The first determining module 304 is configured to determine whether the pixel area is larger than an area threshold, so as to obtain a first determining result.
The unqualified fragment determination module 305 is configured to determine that the to-be-detected confidential medium fragment corresponding to the mask is an "unqualified fragment" when the first determination result is that the pixel area is larger than the area threshold.
The qualified debris determining module 306 is configured to determine that the confidential medium debris to be detected corresponding to the mask is "qualified debris" when the first determination result is that the pixel area is smaller than or equal to the area threshold.
The proportion determination module 307 is used to determine the proportion of the number of "rejected pieces" to the total number of pieces of the security media to be detected.
The second determining module 308 is configured to determine whether the ratio is greater than a preset ratio threshold, so as to obtain a second determining result.
The first crushing result determining module 309 is configured to determine that the image of the confidential medium fragment to be detected does not meet the requirement on the medium destruction granularity when the second determination result is that the ratio is greater than the preset ratio threshold.
The second crushing result determining module 310 is configured to determine that the confidential medium fragments to be detected meet the requirement on the medium destruction granularity when the second determination result is that the ratio is smaller than or equal to the preset ratio threshold.
As shown in fig. 4, another security media destruction granularity identification system includes: a media fragment recognition module 400, a fragment area calculation module 401, and a pulverization result determination module 402.
The medium fragment recognition module 400 is configured to obtain a mask of each confidential medium fragment to be detected and a medium type to which each confidential medium fragment to be detected belongs based on an image of the confidential medium fragments to be detected by using a trained medium fragment recognition model. The medium fragment recognition model comprises: the system comprises a backbone network, a multi-scale feature fusion network, a regional suggestion network and a detection head.
The fragment area calculation module 401 is connected to the medium fragment identification module 400, and is configured to determine an area threshold according to the medium category of each confidential medium fragment to be detected, determine a pixel area according to the mask of each confidential medium fragment to be detected, determine whether the pixel area is greater than the area threshold, obtain a first determination result, and determine whether the confidential medium fragment to be detected corresponding to the mask is an "unqualified fragment" according to the first determination result.
The crushing result judging module 402 is connected with the fragment area calculating module 401, and is configured to determine a ratio of the number of "unqualified fragments" in the confidential medium fragments to the total number of the confidential medium fragments to be detected, determine whether the ratio is greater than a preset ratio threshold value, obtain a second judgment result, and judge whether the image of the confidential medium fragments to be detected meets the requirement on medium destruction granularity according to the second judgment result.
The construction process of the fragment area calculation module 401 is as follows: first for each media categoryiSetting area thresholdS i I.e. the maximum comminution area for that category. Based on each fragment mask output by the previous step medium fragment identification module 400jCalculating the pixel area of the mask in the imageS j . According to the media category to which each fragment output by the media fragment identification module 400 belongsiDetermining an area threshold corresponding to the media debrisS i . If the pixel area of the fragment maskS j Is greater thanS i Judging the chip to be unqualified, otherwise, judging the chip to be qualified.
The construction process of the crushing result judgment module 402 is as follows: firstly, a crushing proportion threshold value T' is set, namely, in a crushing image which meets the requirement of the granularity of the medium to be destroyed, the maximum proportion value of the number of unqualified fragments in the total number of fragments is calculated. And calculating the proportion T of the number of the unqualified fragments in the medium crushing image to the total number of the fragments based on the unqualified fragments and the qualified fragments output by the fragment area calculation module in the last step. Wherein the total number of fragments = "number of unqualified fragments +" number of qualified fragments ". And if the ratio T of the unqualified fragments is greater than the crushing ratio threshold value T', judging that the medium crushing image does not meet the requirement of the granularity of the medium destruction, starting secondary crushing, and otherwise, outputting crushing success to the medium crushing image.
In summary, the method and the system for identifying the destroying granularity of the confidential medium provided by the invention have the following advantages compared with the prior art:
1. the method identifies and analyzes the destroying granularity of the confidential medium based on the convolutional neural network, thereby automatically judging whether the crushed medium meets the destroying requirement or needs to be subjected to secondary crushing in the mechanical crushing process of the confidential medium, greatly improving the working efficiency and reliability of the medium destroying process, reducing the possibility of information leakage and filling the loophole of the mechanical crushing technology of the confidential medium.
2. According to the method, the image segmentation network based on the multi-scale feature fusion technology and the CBAM attention mechanism is constructed aiming at the characteristics that the areas of medium fragments in the image are different in size and the small-area fragments account for more, so that semantic information contained in a feature map is effectively increased, and key feature information beneficial to fragment identification is refined, so that the network can accurately detect the medium fragments with different scales, and particularly the identification capability of the network on the small-area medium fragments is improved.
3. The invention calculates the area proportion of the medium fragments in the medium crushing image based on the fragment mask output by the image segmentation network, and measures the crushing effect of the medium fragments by combining the medium types of the fragments. The method can automatically extract rich and effective fragment feature information aiming at different types of medium fragments, and has higher universality and identification efficiency compared with the traditional image processing algorithm needing manual feature design such as Hough transform.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.
Claims (8)
1. A method for identifying the granularity at which a confidential medium is destroyed, comprising:
acquiring an image of a confidential medium fragment to be detected;
inputting the image of the confidential medium fragments to be detected into a trained medium fragment recognition model to obtain a mask of each confidential medium fragment to be detected and a medium class to which each confidential medium fragment to be detected belongs; the medium fragment recognition model includes: a backbone network, a multi-scale feature fusion network, a regional suggestion network and a detection head;
determining an area threshold according to the medium category of each confidential medium fragment to be detected;
determining the area of a pixel according to the mask of each confidential medium fragment to be detected;
judging whether the pixel area is larger than the area threshold value or not to obtain a first judgment result;
when the first judgment result shows that the pixel area is larger than the area threshold, judging that the confidential medium fragments to be detected corresponding to the mask are unqualified fragments;
when the first judgment result is that the pixel area is smaller than or equal to the area threshold, judging that the confidential medium fragments to be detected corresponding to the mask are qualified fragments;
determining the proportion of the number of the unqualified fragments to the total number of the confidential medium fragments to be detected;
judging whether the proportion is larger than a preset proportion threshold value or not to obtain a second judgment result;
when the second judgment result is that the ratio is larger than a preset ratio threshold, judging that the image of the confidential medium fragment to be detected does not meet the requirement of medium destruction granularity;
when the second judgment result is that the ratio is smaller than or equal to a preset ratio threshold, judging that the confidential medium fragments to be detected meet the requirement on medium destruction granularity;
the construction process of the medium fragment identification model comprises the following steps:
taking a convolutional neural network ResNet50 as a backbone network; the convolutional neural network ResNet50 is composed of a plurality of convolutional blocks connected in sequence;
constructing a multi-scale feature fusion network based on the backbone network; the multi-scale feature fusion network comprises a multi-level pyramid network;
constructing a regional suggestion network based on the multi-scale feature fusion network;
and constructing a detection head based on the area suggestion network.
2. The method for identifying the granularity at which confidential media are destroyed according to claim 1, wherein the constructing of the multi-scale feature fusion network based on the backbone network specifically comprises:
respectively inputting the first characteristic diagram output by each convolution block in the backbone network into a convolution layer with a convolution kernel size of 1 x 1;
after down-sampling operation and up-sampling operation are carried out on the second feature graph output by the first convolutional layer, up-sampling operation and element set addition operation are carried out on the third feature graphs output by the rest convolutional layers in sequence to obtain a fourth feature graph;
and respectively inputting the fourth feature maps into a CBAM attention module to obtain a multi-level pyramid network.
3. The method for identifying the granularity at which confidential media are destroyed according to claim 2, wherein the constructing of the area suggestion network based on the multi-scale feature fusion network specifically comprises:
performing convolution operation with convolution kernel of 3 × 3 on the fifth feature map output by the multi-level pyramid network, and performing traversal operation on the fifth feature map by adopting a sliding anchor frame in the convolution operation process to generate a group of candidate fragment areas;
inputting the seventh feature map for generating the subsequent fragment area into a classification branch to predict the bounding box position of the candidate fragment area;
and inputting the seventh feature map for generating the subsequent fragment region into the regression branch to predict the probability that the candidate fragment region is the medium fragment.
4. The method for identifying the granularity at which confidential media are destroyed according to claim 3, wherein the constructing of the detection head based on the area recommendation network specifically comprises:
mapping each candidate fragment area to a corresponding feature layer according to the area size, and outputting a series of candidate area feature maps with the same size through an ROI Align layer;
inputting the candidate region feature map into a detection branch, inputting the candidate region feature map into a regression branch after feature extraction and classification are carried out on the candidate region feature map through a first full-connection layer in the detection branch, and completing frame regression operation by utilizing a regression Loss function Smooth L1 Loss to obtain position information of the medium fragments; after the characteristic map of the candidate region is subjected to characteristic extraction and classification through a second full connection layer, inputting the characteristic map into a classification branch, and performing background and fragment class classification by using a classification Loss function Softmax Loss to determine the medium class to which the candidate fragment region belongs;
and inputting the candidate region characteristic graph into a mask branch, and in the mask branch, after the candidate region characteristic graph is subjected to characteristic extraction and classification through a third convolution layer and a fourth convolution layer, obtaining a mask corresponding to each medium fragment by using average binary cross entropy loss.
5. The method for identifying the destruction granularity of confidential media according to claim 1, wherein the convolutional neural network ResNet50 is composed of 5 convolutional blocks connected in sequence.
6. The method for identifying the granularity at which confidential media is destroyed according to claim 1, wherein the process of training and testing the media fragment identification model comprises:
acquiring a media crushing image segmentation dataset;
dividing the medium crushing image segmentation data set into a training set and a test set according to a set proportion;
after the training set is adopted to train the medium fragment recognition model, a test set is adopted to test the trained medium fragment recognition model, and in the test process, when the trained medium fragment recognition model meets a preset condition, a trained medium fragment recognition model is obtained; and when the trained medium fragment recognition model does not meet the preset condition, re-training the medium fragment recognition model.
7. The method of claim 6, wherein the obtaining the media shredding image segmentation dataset further comprises:
acquiring fragment images of various smashed confidential media;
labeling the fragment images by adopting Labelme software to obtain a labeling file corresponding to each fragment image; the format of the markup file is JSON format;
compiling a script, and converting the format of the markup file into an XML format to obtain a medium crushing image data set;
a media-shredding image-segmentation dataset in the VOC2007 dataset format is obtained based on the media-shredding image dataset and the annotation file.
8. A system for identifying the granularity at which a security medium is destroyed, comprising:
the image acquisition module is used for acquiring an image of the confidential medium fragment to be detected;
the mask-medium type identification module is used for inputting the image of the confidential medium fragments to be detected into a trained medium fragment identification model to obtain a mask of each confidential medium fragment to be detected and a medium type to which each confidential medium fragment to be detected belongs; the medium fragment recognition model includes: a backbone network, a multi-scale feature fusion network, a regional suggestion network and a detection head;
the area threshold determining module is used for determining an area threshold according to the medium category of each confidential medium fragment to be detected;
the pixel area determining module is used for determining the pixel area according to the mask of each confidential medium fragment to be detected;
the first judgment module is used for judging whether the pixel area is larger than the area threshold value or not to obtain a first judgment result;
an unqualified fragment determining module, configured to determine that the to-be-detected confidential medium fragment corresponding to the mask is an "unqualified fragment" when the first determination result indicates that the pixel area is larger than the area threshold;
a qualified fragment determining module, configured to determine that the confidential medium fragment to be detected corresponding to the mask is a "qualified fragment" when the first determination result indicates that the pixel area is smaller than or equal to the area threshold;
the proportion determining module is used for determining the proportion of the number of the unqualified fragments in the total number of the confidential medium fragments to be detected;
the second judgment module is used for judging whether the proportion is larger than a preset proportion threshold value or not to obtain a second judgment result;
the first crushing result determining module is used for judging that the image of the confidential medium fragment to be detected does not meet the requirement of medium destruction granularity when the second judgment result is that the ratio is greater than a preset ratio threshold;
the second crushing result determining module is used for determining that the confidential medium fragments to be detected meet the requirement on medium destruction granularity when the second judgment result shows that the proportion is smaller than or equal to a preset proportion threshold;
the construction process of the medium fragment identification model comprises the following steps:
taking a convolutional neural network ResNet50 as a backbone network; the convolutional neural network ResNet50 is composed of a plurality of convolutional blocks connected in sequence;
constructing a multi-scale feature fusion network based on the backbone network; the multi-scale feature fusion network comprises a multi-level pyramid network;
constructing a regional suggestion network based on the multi-scale feature fusion network;
and constructing a detection head based on the area suggestion network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110991513.3A CN113435424B (en) | 2021-08-27 | 2021-08-27 | Method and system for identifying destroying granularity of confidential medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110991513.3A CN113435424B (en) | 2021-08-27 | 2021-08-27 | Method and system for identifying destroying granularity of confidential medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113435424A CN113435424A (en) | 2021-09-24 |
| CN113435424B true CN113435424B (en) | 2021-11-26 |
Family
ID=77798117
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110991513.3A Active CN113435424B (en) | 2021-08-27 | 2021-08-27 | Method and system for identifying destroying granularity of confidential medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113435424B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114528950B (en) * | 2022-04-24 | 2022-07-12 | 北京中超伟业信息安全技术股份有限公司 | Destroying method and system for identifying type of confidential medium based on three-dimensional point cloud |
| CN116109636B (en) * | 2023-04-13 | 2023-06-23 | 北京中超伟业信息安全技术股份有限公司 | Magnetic medium destruction processing method and system based on visual monitoring |
| CN116758547B (en) * | 2023-06-27 | 2024-03-12 | 北京中超伟业信息安全技术股份有限公司 | Paper medium carbonization method, system and storage medium |
| CN116561809B (en) * | 2023-07-10 | 2023-10-24 | 北京中超伟业信息安全技术股份有限公司 | Destroying method for identifying security medium based on point cloud |
| CN116702069B (en) * | 2023-08-08 | 2023-11-24 | 北京中超伟业信息安全技术股份有限公司 | Medium destroying method based on state sensing |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101494095B1 (en) * | 2013-09-11 | 2015-02-16 | 한국컴퓨터재생센터주식회사 | System and method for monitoring erasing data of hard disk drive |
| CN112287303B (en) * | 2020-12-24 | 2021-04-02 | 北京中超伟业信息安全技术股份有限公司 | Destroying equipment suitable for traceable confidential medium |
| CN112285479B (en) * | 2020-12-24 | 2021-03-30 | 北京中超伟业信息安全技术股份有限公司 | System and method for detecting integrity of confidential medium |
| CN112257099B (en) * | 2020-12-24 | 2021-04-27 | 北京中超伟业信息安全技术股份有限公司 | Confidential equipment identification device suitable for emergency destruction vehicle |
-
2021
- 2021-08-27 CN CN202110991513.3A patent/CN113435424B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN113435424A (en) | 2021-09-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113435424B (en) | Method and system for identifying destroying granularity of confidential medium | |
| CN108876780B (en) | Bridge crack image crack detection method under complex background | |
| CN107808358B (en) | Automatic detection method for image watermark | |
| CN110598800A (en) | Garbage classification and identification method based on artificial intelligence | |
| Li et al. | Sewer pipe defect detection via deep learning with local and global feature fusion | |
| Zhang et al. | A dense u-net with cross-layer intersection for detection and localization of image forgery | |
| Kadam et al. | Detection and localization of multiple image splicing using MobileNet V1 | |
| Zhou et al. | Deep learning-based crack segmentation for civil infrastructure: Data types, architectures, and benchmarked performance | |
| CN105069447A (en) | Facial expression identification method | |
| CN114528950B (en) | Destroying method and system for identifying type of confidential medium based on three-dimensional point cloud | |
| CN113642390A (en) | Street view image semantic segmentation method based on local attention network | |
| CN114841972A (en) | Power transmission line defect identification method based on saliency map and semantic embedded feature pyramid | |
| CN112257665A (en) | Image content recognition method, image recognition model training method, and medium | |
| Sarmiento | Pavement distress detection and segmentation using YOLOv4 and DeepLabv3 on pavements in the Philippines | |
| Li et al. | A pavement crack detection method based on multiscale Attention and HFS | |
| CN114842524B (en) | Face false distinguishing method based on irregular significant pixel cluster | |
| CN104156963A (en) | Automatic authenticating method of electrical fire hazard melted mark physical evidence | |
| CN112818818B (en) | Novel ultra-high-definition remote sensing image change detection method based on AFFPN | |
| CN116189063B (en) | Key frame optimization method and device for intelligent video monitoring | |
| CN116168348B (en) | Security monitoring method, system and storage medium based on image processing | |
| Jia et al. | Sample generation of semi‐automatic pavement crack labelling and robustness in detection of pavement diseases | |
| CN114332536A (en) | Forged image detection method, system and storage medium based on posterior probability | |
| CN114519819A (en) | Remote sensing image target detection method based on global context awareness | |
| CN114202765A (en) | Image text recognition method and storage medium | |
| Shufang et al. | Detection and Classification of Small Traffic Signs Based on Cascade Network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |