Construction method of streaming encryption algorithm based on deep learning

Abstract

The invention discloses a construction method of a streaming encryption algorithm based on deep learning. An encryption circle is added, the edge of the encryption circle is divided into 360 arcs, the relation between the arcs of the encryption circle and characters is matched, and the relation is stored through a matching network; in order to increase the accuracy of the matching network, a detection network is added, parameters of the matching network are adjusted, and an output value of the matching network is also adjusted, so that more accurate matching data are found; therefore, the accurately encrypted second prediction position vector can be obtained only through the matching network, the detection network, the circle center and the size of the corresponding encryption circle, and then the decryption network, the detection network and the matching network are jointly trained during retraining through the decryption network.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method for constructing a streaming encryption algorithm based on deep learning. Background Art

[0002] Currently, stream cipher (CK) is a permutation-based, multi-key, symmetric encryption method. Most symmetric encryption algorithms are variations of block ciphers. This means that encryption requires alignment of the previous data. The data within a block is dependent on each other, and the previously generated ciphertext is used to generate the next ciphertext. This requires decryption of an entire block of data, necessitating data alignment during random reads and writes. Stream cipher (CK), on the other hand, performs permutation encryption only on characters. Then, using a multi-key algorithm, it generates a pseudo-random stream. The size of the pseudo-random stream is the same as the plaintext size. The algorithm can directly determine the value of the pseudo-random stream at a specific position, effectively supporting random reads and writes and allowing data to be stored directly on the storage medium as ciphertext.

[0003] Because the number and range of characters in stream encryption are fixed, the encryption method is not as complicated as block encryption. Therefore, when encrypting, it is necessary to increase the confidentiality of the encryption so that it is not easy to be decrypted by anyone other than the decryptor. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for constructing a streaming encryption algorithm based on deep learning to solve the above-mentioned problems existing in the prior art, including: Obtain data to be encrypted and a plaintext character set; the data to be encrypted represents a plurality of characters constituting data to be encrypted; the plaintext character set includes a plurality of characters capable of constituting the data to be encrypted; Based on the plaintext character set, an encrypted circle is constructed and a matching network is trained; the matching network represents a network capable of one-to-one correspondence between values ​​in the plaintext character set and arcs of the encrypted circle; Inputting the characters in the data to be encrypted into the trained matching network to obtain a set of predicted positions; obtaining multiple sets of predicted positions corresponding to multiple characters in the data to be encrypted; the set of predicted positions represents the positions corresponding to the arcs on the encryption circle predicted for the data to be encrypted; Through the detection network, based on the radius of the encryption circle, multiple predicted position sets are adjusted to obtain a second predicted position vector; the first value of the second predicted position vector represents the radius of the encryption circle, and the other values ​​represent the encrypted data to be encrypted.

[0005] Optionally, also include: Using the first value of the second predicted position vector as a decryption radius, and values ​​of the second predicted position vector other than the first value as vectors to be decrypted; The length of the arc corresponding to the decryption radius is used as the segmentation length; Splitting the to-be-decrypted vector according to the segmentation length to obtain a plurality of to-be-decrypted sets; Inputting the set to be decrypted into the decryption network to obtain a decrypted character; multiple sets to be decrypted correspondingly obtain multiple decrypted characters; Arrange the decrypted characters in the order of encryption circle division to obtain decrypted data; the decrypted data is equal to the data to be decrypted.

[0006] Optionally, constructing an encryption circle based on the plaintext character set and training a matching network includes: Acquire an encrypted circle image; wherein an encrypted circle is drawn in the encrypted circle image; Take a position on the edge of the encryption circle as the starting position; Divide the edge of the encrypted circle into 360 equal parts starting from the starting position, and obtain 360 arcs; Obtain the position of the arc in the encrypted circle image in a clockwise direction to obtain the arc position set; when there are 360 ​​arcs, 360 arc position sets are obtained. Matching the values ​​in the plaintext character set with the arc position set to obtain a matching relationship set; one value in the plaintext character set corresponds to one arc position set; A matching network is trained based on the matching relationship set.

[0007] Optionally, the step of adjusting the plurality of predicted position sets based on the radius of the encryption circle through the detection network to obtain the second predicted position vector includes: Marking the values ​​in the predicted position set to obtain a predicted arc position image; the size of the predicted arc position image is equal to the size of the encrypted circle image; multiple predicted position sets correspond to multiple predicted arc position images; The predicted arc position image is input into the detection network to determine the state of the encrypted arc and obtain the predicted angle and predicted radius; the predicted angle represents the center angle of the arc corresponding to the predicted position set; the predicted radius represents the radius of the circle corresponding to the arc corresponding to the predicted position set; Multiple predicted position sets correspond to multiple predicted angles and multiple predicted radii; Clustering multiple prediction angles to obtain clustering angles; clustering multiple prediction radii to obtain clustering radius; reconstructing a circle according to the cluster radius to obtain a reconstructed circle image; Based on the clustering angle, the predicted arc position image and the reconstructed circle image are matched to obtain a second predicted position set; multiple predicted arc position images correspond to multiple second predicted position sets.

[0008] Optionally, the training a matching network based on the matching relationship set includes: Inputting the values ​​in the plaintext character set into a matching network, detecting the positions of the arcs of the corresponding encrypted circles, and obtaining a predicted encrypted circle position set; The predicted encrypted circle position set includes multiple predicted encrypted circle positions; the subscripts of the predicted encrypted circle positions represent the order of obtaining the predicted encrypted circle positions in a clockwise direction; The matching network is used to store the matching relationship between the values ​​in the plaintext character set and the arcs of the encryption circle; The loss is calculated between the predicted encrypted circle position set and the arc position set in the matching relationship set to train the matching network.

[0009] Optionally, matching the predicted arc position image with the reconstructed circle image based on the clustering angle to obtain a second predicted position set includes: superimposing the predicted arc position image and the reconstructed circle image into an overlapping image; The arc in the predicted arc position image and the center of the reconstructed circle form a predicted graph; Construct cluster sectors based on the center of the reconstructed circle, cluster angle and cluster radius; Taking the center of the reconstructed circle as the rotation center, rotate the cluster sectors in sequence; Calculate the area difference between the rotated cluster sector and the predicted graph to obtain the difference area; Multiple rotations correspond to obtaining multiple phase difference areas; The smallest phase difference area among the multiple phase difference areas is used as the predicted area; The position of the arc of the cluster sector corresponding to the predicted area is added to the second predicted position set.

[0010] Optionally, the detection network training method includes: Get the annotation angle and annotation radius; the annotation angle is 1; the annotation radius is the radius of the encryption circle; Calculating losses at multiple positions corresponding to the reconstructed circular image and the encrypted circular image to obtain a first loss value; Calculating a loss between the predicted angle and the marked angle to obtain a second loss value; The predicted radius and the marked radius are used to calculate the loss to obtain a third loss value.

[0011] Optionally, the input of the matching network is a character in the data to be encrypted; The output of the matching network is the input of the detection network; The output of the detection network is the input of the decryption network; The output of the decryption network is the decrypted character corresponding to the character in the encrypted data.

[0012] Optionally, the loss of the decrypted characters and the corresponding characters in the data to be encrypted is calculated to jointly train the decryption network, the detection network and the matching network.

[0013] Optionally, the length of the second predicted position vector increases as the radius of the encryption circle increases.

[0014] Compared with the prior art, the embodiments of the present invention achieve the following beneficial effects: In order to increase the complexity of encryption and make it difficult to decrypt the decryption network except for fixed training parameters during subsequent decryption, the present invention adds an encryption circle. The edge of the encryption circle is divided into 360 arcs, the relationship between the arcs of the encryption circle and the characters is matched, and the relationship is stored in the matching network. In order to increase the accuracy of the matching network, a detection network is added, the parameters of the matching network are adjusted, and the output value of the matching network is adjusted to find more accurate matching data. Therefore, the accurately encrypted second predicted position vector can be obtained only through the matching network, the detection network, the center of the circle and the size of the corresponding encryption circle. Then, through the decryption network, and during retraining, the decryption network, the detection network and the matching network are trained together, so that the trained decryption network can decrypt more accurately. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a flow chart of a method for constructing a streaming encryption algorithm based on deep learning provided by an embodiment of the present invention.

[0016] Figure 2 This is a schematic diagram of the encryption circle in the flowchart of a method for constructing a streaming encryption algorithm based on deep learning provided in an embodiment of the present invention.

[0017] Figure 3 This is a schematic diagram of overlapping images in a flowchart of a method for constructing a streaming encryption algorithm based on deep learning provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The present invention will be described in detail below with reference to the accompanying drawings.

[0019] Example 1 like Figure 1 As shown, an embodiment of the present invention provides a method for constructing a stream encryption algorithm based on deep learning, the method comprising: S101: Obtain data to be encrypted and a plaintext character set; the data to be encrypted represents data to be encrypted consisting of multiple characters; the plaintext character set includes multiple characters that can constitute the data to be encrypted.

[0020] The plaintext character set includes 256 characters [0x00, 0xff].

[0021] S102: Based on the plaintext character set, construct an encryption circle and train a matching network; the matching network represents a network that can match values ​​in the plaintext character set with arcs of the encryption circle in a one-to-one manner.

[0022] S103: Input the characters in the data to be encrypted into the trained matching network to obtain a set of predicted positions; multiple characters in the data to be encrypted correspond to multiple sets of predicted positions; the predicted position sets represent the positions corresponding to the arcs on the encryption circle predicted for the data to be encrypted.

[0023] S104: By detecting the network, based on the radius of the encryption circle, multiple predicted position sets are adjusted to obtain a second predicted position vector; the first value of the second predicted position vector represents the radius of the encryption circle, and the other values ​​represent the encrypted data to be encrypted.

[0024] In this embodiment, the radius of the encryption circle is fixed. A larger setting means a longer arc length for matching, and thus a longer length of the second predicted position vector.

[0025] Using this method, arcs are used to represent characters. Even if the prediction is inaccurate in some subtle areas, using the values ​​corresponding to multiple arcs to represent a single character allows for overall judgment based on shape and position. This allows accurate prediction even when partial predictions are inaccurate, thus addressing network inaccuracies.

[0026] Optionally, also include: The first value of the second predicted position vector is used as a decryption radius, and values ​​of the second predicted position vector other than the first value are used as vectors to be decrypted.

[0027] The decryption radius is retained during the transmission process.

[0028] The length of the arc corresponding to the decryption radius is used as the segmentation length; Splitting the to-be-decrypted vector according to the segmentation length to obtain a plurality of to-be-decrypted sets; The set to be decrypted is input into the decryption network to obtain a decrypted character; multiple sets to be decrypted correspond to multiple decrypted characters.

[0029] In this embodiment, the decryption network is a convolutional neural network (CNN).

[0030] Arrange the decrypted characters in the order of encryption circle division to obtain decrypted data; the decrypted data is equal to the data to be decrypted.

[0031] Optionally, constructing an encryption circle based on the plaintext character set and training a matching network includes: An encrypted circle image is obtained; an encrypted circle is drawn in the encrypted circle image.

[0032] The encrypted circle image is a binary image; the center of the encrypted circle is at the midpoint of the encrypted circle image.

[0033] Take a position on the edge of the encryption circle as the starting position; Divide the side of the encrypted circle into 360 equal parts starting from the starting position to obtain 360 arcs.

[0034] The arc represents an arc on the edge of the encrypted circle.

[0035] Because we are using character encryption, the range that a character can represent is exactly [0x00, 0xff]. We represent the plaintext character to be encrypted with C, which must be a number between [0x00, 0xff]. Therefore, we divide the encryption circle into 360 equal parts. 360 > 256, so characters can be represented by the edges of the encryption circle.

[0036] The schematic diagram of the encryption circle is as follows: Figure 2 shown.

[0037] The values ​​in the plaintext character set are matched with arcs to obtain a matching relationship set.

[0038] Among them, one value in the plaintext character set corresponds to one arc, and one arc definitely corresponds to one value in the plaintext character set.

[0039] In this embodiment, one value in the plaintext character set is matched with an arc in clockwise order, for example, the character 0x00 corresponds to the arc rotated 1 degree clockwise from the starting position, and the character 0x01 corresponds to the arc rotated 2 degrees clockwise from the starting position.

[0040] A matching network is trained based on the matching relationship set.

[0041] Optionally, the step of adjusting the plurality of predicted position sets based on the radius of the encryption circle through the detection network to obtain the second predicted position vector includes: Marking the values ​​in the predicted position set to obtain a predicted arc position image; the size of the predicted arc position image is equal to the size of the encrypted circle image; multiple predicted position sets correspond to multiple predicted arc position images; The predicted arc position image is input into the detection network to determine the state of the encrypted arc and obtain the predicted angle and predicted radius; the predicted angle represents the center angle of the arc corresponding to the predicted position set; the predicted radius represents the radius of the circle corresponding to the arc corresponding to the predicted position set.

[0042] In this embodiment, the detection network is a fully connected neural network (FCN).

[0043] Multiple predicted position sets correspond to multiple predicted angles and multiple predicted radii; Clustering multiple prediction angles to obtain clustering angles; clustering multiple prediction radii to obtain clustering radii.

[0044] The clustering angle and clustering radius are clustering centers, and the predicted angle and predicted radius are clustered in order to obtain the same angle and radius as the encrypted circle.

[0045] In this embodiment, the k-means method is used for clustering.

[0046] The circle is reconstructed according to the cluster radius to obtain a reconstructed circle image.

[0047] The reconstructed circle image is obtained to judge the accuracy of the matching network and the detection network, and the detection network is used to train the matching network so as to obtain a network that more accurately matches arcs and characters.

[0048] The center of the encrypted circle is used as the center of the circle, and a circle is drawn with the cluster radius to obtain a reconstructed circle image. The size of the reconstructed circle image is equal to the size of the encrypted circle image.

[0049] Based on the clustering angle, the predicted arc position image and the reconstructed circle image are matched to obtain a second predicted position set; multiple predicted arc position images correspond to multiple second predicted position sets.

[0050] Since there are slight differences between the values ​​in the reconstructed circle and the predicted arc position set, the predicted arc position set is replaced by the second predicted position set on the reconstructed circle.

[0051] Optionally, the matching network training method includes: The values ​​in the plaintext character set are input into a matching network, and the positions of the arcs of the corresponding encrypted circles are detected to obtain a predicted encrypted circle position set.

[0052] Here, the value in the plain text character set is found to correspond to the arc. In this embodiment, one value in the plain text character set corresponds to one arc, and the arc can be used to determine the corresponding character.

[0053] The number of output neurons in the matching network is equal to the number of elements in the arc position set. Since the edge of the encryption circle is bisected, the arc lengths are the same, so the number of elements in the 360 ​​arc position sets is the same.

[0054] In this embodiment, the matching network is a convolutional neural network (CNN).

[0055] The predicted encrypted circle position set includes multiple predicted encrypted circle positions; the subscripts of the predicted encrypted circle positions represent the order of obtaining the predicted encrypted circle positions in a clockwise direction.

[0056] The matching network is used to store the matching relationship between the values ​​in the plaintext character set and the arcs of the encryption circle; The predicted encrypted circle position set and the arc position set are used to calculate the loss and train the matching network.

[0057] In this embodiment, the loss of the values ​​corresponding to the subscripts of the predicted encrypted circle position set and the arc position set is calculated and then averaged to obtain the loss value.

[0058] In this embodiment, a cross entropy loss function is used to calculate the loss value. The backward matching network is often trained.

[0059] Optionally, matching the predicted arc position image with the reconstructed circle image based on the clustering angle to obtain a second predicted position set includes: The predicted arc position image and the reconstructed circle image are superimposed to form an overlapping image.

[0060] In this embodiment, the overlapping image is as follows: Figure 3 shown.

[0061] The arc in the predicted arc position image and the center of the reconstructed circle form a predicted graph; Construct cluster sectors based on the center of the reconstructed circle, cluster angle and cluster radius; Taking the center of the reconstructed circle as the rotation center, rotate the cluster sectors in sequence; Calculate the area difference between the rotated cluster sector and the predicted graph to obtain the difference area; Multiple rotations correspond to obtaining multiple phase difference areas; The smallest phase difference area among the multiple phase difference areas is used as the predicted area; The position of the arc of the cluster sector corresponding to the predicted area is added to the second predicted position set.

[0062] combine.

[0063] Among them, in this embodiment, Figure 3 The character corresponding to the predicted arc position image corresponding to the predicted arc position set represented in is "a".

[0064] Optionally, the detection network training method includes: Get the annotation angle and annotation radius; the annotation angle is 1; the annotation radius is the radius of the encryption circle; Calculating losses at multiple positions corresponding to the reconstructed circular image and the encrypted circular image to obtain a first loss value; Calculating a loss between the predicted angle and the marked angle to obtain a second loss value; Calculating the loss between the predicted radius and the marked radius to obtain a third loss value; In this embodiment, a cross entropy loss function is used to obtain the loss value.

[0065] Optionally, the input of the matching network is a character in the data to be encrypted; The output of the matching network is the input of the detection network; The output of the detection network is the input of the decryption network; The output of the decryption network is the decrypted character corresponding to the character in the encrypted data.

[0066] Optionally, the loss of the decrypted characters and the corresponding characters in the data to be encrypted is calculated to jointly train the decryption network, the detection network and the matching network.

[0067] Optionally, the length of the second predicted position vector increases as the radius of the encryption circle increases.

[0068] The algorithm and display provided herein are not inherently related to any particular computer, virtual system or other device. Various general-purpose systems can also be used together with the teachings based on this. According to the above description, it is obvious that the structure required for constructing this type of system. In addition, the present invention is not directed to any specific programming language. It should be understood that various programming languages ​​can be utilized to realize the content of the present invention described herein, and the above description of specific languages ​​is for the purpose of disclosing the best mode of the present invention.

[0069] In the description provided herein, a large number of specific details are described. However, it is understood that embodiments of the present invention can be practiced without these specific details. In some instances, well-known methods and structures are not shown in detail so as not to obscure the understanding of this description.

Claims

1. A method for constructing a streaming encryption algorithm based on deep learning, characterized in that: include: Obtain data to be encrypted and a plaintext character set; the data to be encrypted represents a plurality of characters constituting data to be encrypted; the plaintext character set includes a plurality of characters capable of constituting the data to be encrypted; Based on the plaintext character set, an encrypted circle is constructed and a matching network is trained; the matching network represents a network capable of one-to-one correspondence between values ​​in the plaintext character set and arcs of the encrypted circle; Inputting the characters in the data to be encrypted into the trained matching network to obtain a set of predicted positions; obtaining multiple sets of predicted positions corresponding to multiple characters in the data to be encrypted; the set of predicted positions represents the positions corresponding to the arcs on the encryption circle predicted for the data to be encrypted; Through the detection network, based on the radius of the encryption circle, multiple predicted position sets are adjusted to obtain a second predicted position vector; the first value of the second predicted position vector represents the radius of the encryption circle, and the other values ​​represent the encrypted data to be encrypted.

2. The method for constructing a streaming encryption algorithm based on deep learning according to claim 1, characterized in that: Also includes: Using the first value of the second predicted position vector as a decryption radius, and values ​​of the second predicted position vector other than the first value as vectors to be decrypted; The length of the arc corresponding to the decryption radius is used as the segmentation length; Splitting the to-be-decrypted vector according to the segmentation length to obtain a plurality of to-be-decrypted sets; Inputting the set to be decrypted into the decryption network to obtain a decrypted character; multiple sets to be decrypted correspondingly obtain multiple decrypted characters; Arrange the decrypted characters in the order of encryption circle division to obtain decrypted data; the decrypted data is equal to the data to be decrypted.

3. The method for constructing a streaming encryption algorithm based on deep learning according to claim 1, characterized in that: The step of constructing an encryption circle based on the plaintext character set and training a matching network includes: Acquire an encrypted circle image; wherein an encrypted circle is drawn in the encrypted circle image; Take a position on the edge of the encryption circle as the starting position; Divide the edge of the encrypted circle into 360 equal parts starting from the starting position, and obtain 360 arcs; Obtain the position of the arc in the encrypted circle image in a clockwise direction to obtain the arc position set; when there are 360 ​​arcs, 360 arc position sets are obtained. Matching the values ​​in the plaintext character set with the arc position set to obtain a matching relationship set; one value in the plaintext character set corresponds to one arc position set; A matching network is trained based on the matching relationship set.

4. The method for constructing a streaming encryption algorithm based on deep learning according to claim 1, characterized in that: The method of adjusting the plurality of predicted position sets based on the radius of the encryption circle through the detection network to obtain a second predicted position vector includes: Marking the values ​​in the predicted position set to obtain a predicted arc position image; the size of the predicted arc position image is equal to the size of the encrypted circle image; multiple predicted position sets correspond to multiple predicted arc position images; The predicted arc position image is input into the detection network to determine the state of the encrypted arc and obtain the predicted angle and predicted radius; the predicted angle represents the center angle of the arc corresponding to the predicted position set; the predicted radius represents the radius of the circle corresponding to the arc corresponding to the predicted position set; Multiple predicted position sets correspond to multiple predicted angles and multiple predicted radii; Clustering multiple prediction angles to obtain clustering angles; clustering multiple prediction radii to obtain clustering radius; reconstructing a circle according to the cluster radius to obtain a reconstructed circle image; Based on the clustering angle, the predicted arc position image and the reconstructed circle image are matched to obtain a second predicted position set; multiple predicted arc position images correspond to multiple second predicted position sets.

5. The method for constructing a streaming encryption algorithm based on deep learning according to claim 3, characterized in that: The training of the matching network based on the matching relationship set includes: Inputting the values ​​in the plaintext character set into a matching network, detecting the positions of the arcs of the corresponding encrypted circles, and obtaining a predicted encrypted circle position set; The predicted encrypted circle position set includes multiple predicted encrypted circle positions; the subscripts of the predicted encrypted circle positions represent the order of obtaining the predicted encrypted circle positions in a clockwise direction; The matching network is used to store the matching relationship between the values ​​in the plaintext character set and the arcs of the encryption circle; The loss is calculated between the predicted encrypted circle position set and the arc position set in the matching relationship set to train the matching network.

6. The method for constructing a streaming encryption algorithm based on deep learning according to claim 4, characterized in that: The method of matching the predicted arc position image with the reconstructed circle image based on the clustering angle to obtain a second predicted position set includes: superimposing the predicted arc position image and the reconstructed circle image into an overlapping image; The arc in the predicted arc position image and the center of the reconstructed circle form a predicted graph; Construct cluster sectors based on the center of the reconstructed circle, cluster angle and cluster radius; Taking the center of the reconstructed circle as the rotation center, rotate the cluster sectors in sequence; Calculate the area difference between the rotated cluster sector and the predicted graph to obtain the difference area; Multiple rotations correspond to obtaining multiple phase difference areas; The smallest phase difference area among the multiple phase difference areas is used as the predicted area; The position of the arc of the cluster sector corresponding to the predicted area is added to the second predicted position set.

7. The method for constructing a streaming encryption algorithm based on deep learning according to claim 4, characterized in that: The training method of the detection network includes: Get the annotation angle and annotation radius; the annotation angle is 1; the annotation radius is the radius of the encryption circle; Calculating losses at multiple positions corresponding to the reconstructed circular image and the encrypted circular image to obtain a first loss value; Calculating a loss between the predicted angle and the marked angle to obtain a second loss value; The predicted radius and the marked radius are used to calculate the loss to obtain a third loss value.

8. The method for constructing a streaming encryption algorithm based on deep learning according to claim 2, characterized in that: The input of the matching network is the characters in the data to be encrypted; The output of the matching network is the input of the detection network; The output of the detection network is the input of the decryption network; The output of the decryption network is the decrypted character corresponding to the character in the encrypted data.

9. The method for constructing a streaming encryption algorithm based on deep learning according to claim 2, characterized in that: The loss of the decrypted characters is calculated from the corresponding characters in the encrypted data, and the decryption network, detection network and matching network are trained together.

10. The method for constructing a streaming encryption algorithm based on deep learning according to claim 1, characterized in that: The length of the second predicted position vector increases as the radius of the encryption circle increases.

Images