Three-dimensional data encryption method

Abstract

The invention discloses a three-dimensional data encryption method. The method comprises the following steps: 1) obtaining a sequence image corresponding to a three-dimensional geographic space; 2) preprocessing the sequence image to obtain a preprocessed sequence image; 3) constructing a 3-D full convolution residual error U-net network model; 4) segmenting the preprocessed sequence image by using a 3-D full convolution residual U-net network model to obtain a plurality of segmented images; 5) extracting a low-frequency sub-band of the segmented image by using a discrete wavelet transform method to obtain an approximate image; 6) scrambling the pixel position of the approximate image by using the first chaotic sequence to obtain a first encrypted image; and 7) performing bidirectional diffusion and row-column scrambling on the first encrypted image by using the second chaotic sequence to obtain a final encrypted image. According to the method, the accuracy of three-dimensional data encryption can be effectively ensured, and the security of the encrypted three-dimensional data can be further improved.

Description

Technical Field

[0001] The invention relates to the technical field of three-dimensional data processing, and in particular to a three-dimensional data encryption method. Background Art

[0002] At present, with the development of 5G technology, image data has been popularized in the field of data transmission due to its wide and vivid advantages. Military, surveying, transportation, weather and other fields increase the accuracy of geospatial information by transmitting three-dimensional geospatial data. After the three-dimensional geospatial data (such as photogrammetry, three-dimensional architecture, BIM / CAD, instantiated elements, point cloud corresponding data) is collected or modeled by modeling software such as 3dmax and Maya, it needs to be exported as fbx or dae or obj data, and then converted into 3DTiles data format for three-dimensional data transmission.

[0003] 3D Tiles is an open specification for transmitting massive heterogeneous 3D geospatial datasets. Using tile streaming technology that is conceptually similar to terrain and imagery, 3D Tiles enables large models such as building datasets, BIM models, point clouds, and photogrammetry models to be browsed and displayed more smoothly on the web.

[0004] However, based on the requirements of the Surveying and Mapping Confidentiality Law, when the three-dimensional data generated by the surveying and mapping results reaches a certain level of confidentiality, the three-dimensional data needs to be encrypted before it can be transmitted.

[0005] In the current existing technology, the point coordinates of the object face elements in the three-dimensional model are mostly transformed to prevent other users from correctly obtaining the data of the original three-dimensional model. However, this encryption mode is not only time-consuming, but also the encrypted three-dimensional data is difficult to crack after being intercepted, and the security of the three-dimensional data is insufficient. Summary of the invention

[0006] The object of the present invention is to provide a three-dimensional data encryption method, comprising the following steps:

[0007] 1) Obtaining a sequence of images corresponding to the three-dimensional geographic space;

[0008] 2) preprocessing the sequence images to obtain preprocessed sequence images;

[0009] 3) Construct a 3-D fully convolutional residual U-net network model;

[0010] 4) Using the 3-D fully convolutional residual U-net network model to segment the preprocessed sequence images to obtain multiple segmented images;

[0011] 5) Using discrete wavelet transform method to extract low-frequency subbands of the segmented image and obtain an approximate image;

[0012] 6) Using the first chaotic sequence to scramble the pixel positions of the approximate image to obtain a first encrypted image;

[0013] 7) Using the second chaotic sequence to perform bidirectional diffusion and row-column scrambling on the first encrypted image, to obtain the final encrypted image.

[0014] Furthermore, the method for acquiring the sequence images corresponding to the three-dimensional geographic space is: continuously photographing the three-dimensional geographic space to acquire a set of sequence images.

[0015] Furthermore, the method for obtaining the sequence images corresponding to the three-dimensional geographic space is: modeling the three-dimensional geographic space to obtain a three-dimensional geographic space model; obtaining a set of sequence images of the three-dimensional geographic space model; tools for modeling the three-dimensional geographic space include 3dmax and Maya.

[0016] Furthermore, the preprocessing includes normalization processing.

[0017] Furthermore, the 3-D fully convolutional residual U-net network model includes an encoder and a decoder;

[0018] The encoder comprises a plurality of downsampled residual blocks;

[0019] The encoder downsamples the preprocessed sequence images through convolution, pooling, and ReLU operations to output a high-dimensional feature map;

[0020] The decoder comprises a plurality of upsampled residual blocks;

[0021] The decoder upsamples the high-dimensional feature map through convolution, deconvolution, and ReLU operations to obtain a segmented image.

[0022] Furthermore, the 3-D fully convolutional residual U-net network model introduces an attention mechanism: taking the high-dimensional feature map of the encoder as input, outputting a feature map of the same size through a convolution transformation operation, and then splicing it with the image of the decoder to obtain fine-grained segmentation information.

[0023] Furthermore, the 3-D fully convolutional residual U-net network model is trained by labeled normalized sequence image samples.

[0024] Furthermore, the step of preprocessing the sequence images also includes:

[0025] s1) stacking the sequence images to form a three-dimensional image grayscale matrix, where each layer of the matrix represents a sequence image;

[0026] s2) normalizing the pixel values ​​in the grayscale matrix of the three-dimensional image to between 0 and 1;

[0027] s3) adjusting the pixel spacing of pixels in the three-dimensional matrix so that the pixel spacing meets the preset requirements of network training or application.

[0028] Further, the step of scrambling the pixel positions of the approximate image using the first chaotic sequence includes:

[0029] a1) generating a first chaotic sequence with a length of R+W by using logistic-sine-cosine mapping;

[0030] The mapping process is as follows:

[0031]

[0032] Among them, x i is the state variable, r and β are the control parameters, F(r, x i ) and G(1-r, x i ) are logistic and sine mappings. When β = -0.5, r∈[0,1], x∈[0,1), the system is in a chaotic state; x i+1 is the variable after mapping;

[0033] a2) dividing the first chaotic sequence into a subsequence A with a length of R and a subsequence B with a length of W;

[0034] Sort the two subsequences respectively to obtain two index vectors IA and IB;

[0035] a3) Initialize a matrix of size R*W and set each column of the matrix to IB; use each element of IA to move each column of the matrix to obtain the matrix O for scrambling;

[0036] a4) Initialize column index j=1; find the position {(1, O 1，j ), (2, O 2，j ),…,(R,O R，j )} pixels, move them upward by O 1，j Positions, get the corresponding scrambled pixel position; R，j is the element in the matrix O;

[0037] a5) Let j=j+1, return to step a4), repeat R-1 times, and obtain the first encrypted image.

[0038] Further, the step of using the second chaotic sequence to perform bidirectional diffusion and row-column scrambling on the first encrypted image includes:

[0039] b1) Construct a hyperchaotic system, namely:

[0040]

[0041] in, are the rates of change of x, y, z, and w respectively. When the system parameters a=10, b=8 / 3, c=28, and r=-1, the system is a hyperchaotic system; x, y, z, and w are the generated values ​​of the hyperchaotic system;

[0042] b2) The initial value of the hyperchaotic system (Q x0 , Q y0 , Q z0 , Q w0 ),Right now:

[0043]

[0044] Where mod(a, b) is the remainder of a / b, and fix(a) is the integer part of a. k1, k2, k3, k4 are integers;

[0045] b3) with the initial value (Q x0 , Q y0 , Q z0 , Q w0 ) is used as a reference, and four second chaotic sequences E, G, U, and V are generated using the hyperchaotic system, namely:

[0046]

[0047] In the formula, ei, gi, ui, and vi are user input parameters;

[0048] b4) using the second chaotic sequences E and G to perform bidirectional diffusion on the first encrypted image; and using the second chaotic sequences U and V to perform row and column scrambling on the first encrypted image.

[0049] The technical effect of the present invention is unquestionable. When training the 3-D fully convolutional residual U-net network model, the present invention uses the attention mechanism to splice low-dimensional information and high-dimensional information to obtain more refined image features. At the same time, it can also make full use of the continuity of sequential images to obtain more accurate segmentation results.

[0050] The present invention compresses images by discrete wavelet transform, reduces the amount of data, and improves the subsequent encryption efficiency.

[0051] The present invention performs pixel position scrambling, bidirectional diffusion and row and column replacement through a chaotic sequence, thereby improving the security of encrypted three-dimensional data.

[0052] The present invention can not only effectively ensure the accuracy of three-dimensional data encryption, but also further improve the security of the encrypted three-dimensional data. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1The figure is a flow chart of the method. DETAILED DESCRIPTION

[0054] The present invention is further described below in conjunction with the embodiments, but it should not be understood that the above subject matter of the present invention is limited to the following embodiments. Without departing from the above technical ideas of the present invention, various substitutions and changes are made according to the common technical knowledge and customary means in the art, which should all be included in the protection scope of the present invention.

[0055] Embodiment 1:

[0056] See also Figure 1 , a three-dimensional data encryption method, comprising the following steps:

[0057] 1) Obtaining a sequence of images corresponding to the three-dimensional geographic space;

[0058] 2) preprocessing the sequence images to obtain preprocessed sequence images;

[0059] 3) Construct a 3-D fully convolutional residual U-net network model;

[0060] 4) Using the 3-D fully convolutional residual U-net network model to segment the preprocessed sequence images to obtain multiple segmented images;

[0061] 5) Using discrete wavelet transform method to extract low-frequency subbands of the segmented image and obtain an approximate image;

[0062] 6) Using the first chaotic sequence to scramble the pixel positions of the approximate image to obtain a first encrypted image;

[0063] 7) Using the second chaotic sequence to perform bidirectional diffusion and row-column scrambling on the first encrypted image, to obtain the final encrypted image.

[0064] The method for acquiring the sequence images corresponding to the three-dimensional geographic space is: continuously photographing the three-dimensional geographic space to acquire a set of sequence images.

[0065] The method of acquiring the sequence images corresponding to the three-dimensional geographic space is: modeling the three-dimensional geographic space to obtain the three-dimensional geographic space model; acquiring a set of sequence images of the three-dimensional geographic space model; the tools for modeling the three-dimensional geographic space include 3dmax and Maya.

[0066] Preprocessing includes normalization.

[0067] The 3-D fully convolutional residual U-net network model includes an encoder and a decoder;

[0068] The encoder comprises a plurality of downsampled residual blocks;

[0069] The encoder downsamples the preprocessed sequence images through convolution, pooling, and ReLU operations to output a high-dimensional feature map;

[0070] The decoder comprises a plurality of upsampled residual blocks;

[0071] The decoder upsamples the high-dimensional feature map through convolution, deconvolution, and ReLU operations to obtain a segmented image.

[0072] The 3-D fully convolutional residual U-net network model introduces an attention mechanism: taking the high-dimensional feature map of the encoder as input, outputting a feature map of the same size through a convolution transformation operation, and then splicing it with the image of the decoder to obtain fine-grained segmentation information.

[0073] The 3-D fully convolutional residual U-net network model is trained by normalized sequence image samples with labels. When the region of interest is a single category, the label type uses a binary segmentation label; when it is necessary to distinguish the categories of multiple different regions of interest, the label type uses a multi-class segmentation label, and for complex objects composed of multiple discontinuous parts, the label type uses a connected component label.

[0074] The steps of preprocessing the sequence images also include:

[0075] s1) stacking the sequence images to form a three-dimensional image grayscale matrix, where each layer of the matrix represents a sequence image;

[0076] s2) normalizing the pixel values ​​in the grayscale matrix of the three-dimensional image to between 0 and 1;

[0077] s3) adjusting the pixel spacing of pixels in the three-dimensional matrix so that the pixel spacing meets the preset requirements of network training or application.

[0078] The step of scrambling the pixel positions of the approximate image by using the first chaotic sequence comprises:

[0079] a1) generating a first chaotic sequence with a length of R+W by using logistic-sine-cosine mapping;

[0080] The mapping process is as follows:

[0081]

[0082] Among them, x i is the state variable, r and β are the control parameters, F(r, x i ) and G(1-r, x i ) are logistic and sine mappings. When β = -0.5, r∈[0,1], x∈[0,1), the system is in a chaotic state; x i+1is the variable after mapping;

[0083] a2) dividing the first chaotic sequence into a subsequence A with a length of R and a subsequence B with a length of W;

[0084] Sort the two subsequences respectively to obtain two index vectors IA and IB;

[0085] a3) Initialize a matrix of size R*W and set each column of the matrix to IB; use each element of IA to move each column of the matrix to obtain the matrix O for scrambling;

[0086] a4) Initialize column index j=1; find the position {(1, O 1，j ), (2, O 2，j ),…,(R,O R，j )} pixels, move them upward by O 1，j Positions, get the corresponding scrambled pixel position; R，j is the element in the matrix O;

[0087] a5) Let j=j+1, return to step a4), repeat R-1 times, and obtain the first encrypted image.

[0088] The steps of bidirectionally diffusing and scrambling rows and columns of the first encrypted image using the second chaotic sequence include:

[0089] b1) Construct a hyperchaotic system, namely:

[0090]

[0091] in, are the rates of change of x, y, z, and w respectively. When the system parameters a=10, b=8 / 3, c=28, and r=-1, the system is a hyperchaotic system; x, y, z, and w are the generated values ​​of the hyperchaotic system;

[0092] b2) The initial value of the hyperchaotic system (Q x0 , Q y0 , Q z0 , Q w0 ),Right now:

[0093]

[0094] Where mod(a, b) is the remainder of a / b, and fix(a) is the integer part of a. k1, k2, k3, k4 are integers;

[0095] b3) with the initial value (Q x0 , Q y0 , Q z0 , Q w0) is used as a reference, and four second chaotic sequences E, G, U, and V are generated using the hyperchaotic system, namely:

[0096]

[0097] In the formula, ei, gi, ui, and vi are user input parameters;

[0098] b4) using the second chaotic sequences E and G to perform bidirectional diffusion on the first encrypted image; and using the second chaotic sequences U and V to perform row and column scrambling on the first encrypted image.

[0099] Embodiment 2:

[0100] A three-dimensional data encryption method comprises the following steps:

[0101] 1) Obtaining a sequence of images corresponding to the three-dimensional geographic space;

[0102] 2) preprocessing the sequence images to obtain preprocessed sequence images;

[0103] 3) Construct a 3-D fully convolutional residual U-net network model;

[0104] 4) Using the 3-D fully convolutional residual U-net network model to segment the preprocessed sequence images to obtain multiple segmented images;

[0105] 5) Using discrete wavelet transform method to extract low-frequency subbands of the segmented image and obtain an approximate image;

[0106] 6) Using the first chaotic sequence to scramble the pixel positions of the approximate image to obtain a first encrypted image;

[0107] 7) Using the second chaotic sequence to perform bidirectional diffusion and row-column scrambling on the first encrypted image, to obtain the final encrypted image.

[0108] Embodiment 3:

[0109] A three-dimensional data encryption method, the technical content is the same as that of Example 2, and further, the method for acquiring the sequence image corresponding to the three-dimensional geographic space is: continuously photographing the three-dimensional geographic space to acquire a set of sequence images.

[0110] Embodiment 4:

[0111] A three-dimensional data encryption method, the technical content of which is the same as any one of Examples 2-3. Furthermore, a method for acquiring a sequence of images corresponding to a three-dimensional geographic space is as follows: modeling the three-dimensional geographic space to obtain a three-dimensional geographic space model; acquiring a set of sequence images of the three-dimensional geographic space model; tools for modeling the three-dimensional geographic space include 3dmax and Maya.

[0112] Embodiment 5:

[0113] A three-dimensional data encryption method, the technical content of which is the same as any one of embodiments 2-4, and further, the preprocessing includes normalization processing.

[0114] Embodiment 6:

[0115] A three-dimensional data encryption method, the technical content is the same as any one of embodiments 2-5, further, the 3-D full convolution residual U-net network model includes an encoder and a decoder;

[0116] The encoder comprises a plurality of downsampled residual blocks;

[0117] The encoder downsamples the preprocessed sequence images through convolution, pooling, and ReLU operations to output a high-dimensional feature map;

[0118] The decoder comprises a plurality of upsampled residual blocks;

[0119] The decoder upsamples the high-dimensional feature map through convolution, deconvolution, and ReLU operations to obtain a segmented image.

[0120] Embodiment 7:

[0121] A three-dimensional data encryption method, the technical content of which is the same as any one of Examples 2-6. Furthermore, the 3-D fully convolutional residual U-net network model introduces an attention mechanism: taking the high-dimensional feature map of the encoder as input, outputting a feature map of the same size through a convolution transformation operation, and then splicing it with the image of the decoder to obtain fine-grained segmentation information.

[0122] Embodiment 8:

[0123] A three-dimensional data encryption method, the technical content of which is the same as any one of Examples 2-7, and further, the 3-D fully convolutional residual U-net network model is obtained by training with labeled normalized sequence image samples.

[0124] Embodiment 9:

[0125] A three-dimensional data encryption method, the technical content is the same as any one of embodiments 2-8, further, in order to make the input data suitable for the training and inference process of the neural network, the step of preprocessing the sequence image also includes:

[0126] (1) Stacking the sequence images to form a 3D image grayscale matrix, where each layer represents a sequence image. In this way, the entire 3D image grayscale matrix represents the data of the 3D model.

[0127] (2) Normalizing the pixel values ​​in the grayscale matrix of the three-dimensional image to between 0 and 1 can be done by calculating the minimum and maximum values ​​in the matrix and then using a linear transformation to map all values ​​to the range of 0 to 1;

[0128] (3) Adjust the pixel spacing of the pixels in the three-dimensional matrix to meet the preset requirements of network training or application. This is because the pixel spacing of the original image may not be consistent or meet the optimal parameter settings for model training. Adjusting the pixel spacing can help maintain the consistency of model predictions and, for 3D images, ensure uniform resolution in all three dimensions.

[0129] Embodiment 10:

[0130] A three-dimensional data encryption method, the technical content of which is the same as any one of Embodiments 2-9, further, the step of scrambling the pixel positions of the approximate image using a first chaotic sequence comprises:

[0131] a1) generating a first chaotic sequence with a length of R+W by using logistic-sine-cosine mapping;

[0132] The mapping process is as follows:

[0133]

[0134] Among them, x is the state variable, r and β are control parameters, and F(r, x i ) and G(1-r, x i ) are logistic and sine mappings. When β = -0.5, r∈[0, 1], x∈[0, 1), the system is in a chaotic state;

[0135] a2) dividing the first chaotic sequence into a subsequence A with a length of R and a subsequence B with a length of W;

[0136] Sort the two subsequences respectively to obtain two index vectors IA and IB;

[0137] a3) Initialize a matrix of size R*W and set each column of the matrix to IB; use each element of IA to move each column of the matrix to obtain the matrix O for scrambling;

[0138] a4) Initialize column index j=1; find the position {(1, O 1，j ), (2, O 2，j ),…,(R,O R，j )} pixels, move them upward by O 1，j Positions, get the corresponding scrambled pixel position;

[0139] a5) Let j=j+1, return to step a4), repeat R-1 times, and obtain the first encrypted image.

[0140] Embodiment 11:

[0141] A three-dimensional data encryption method, the technical content of which is the same as any one of Embodiments 2-10, further comprising the steps of using a second chaotic sequence to perform bidirectional diffusion and row-column scrambling on a first encrypted image:

[0142] b1) Construct a hyperchaotic system, namely:

[0143]

[0144] in, are the rates of change of x, y, z, and w respectively. When the system parameters a=10, b=8 / 3, c=28, and r=-1, the system is a hyperchaotic system;

[0145] b2) The initial value of the hyperchaotic system (Q x0 , Q y0 , Q z0 , Q w0 ),Right now:

[0146]

[0147] Wherein, mod(a, b) is the remainder of a / b, and fix(a) is the integer part of a. Under the condition of the initial value, the hyperchaotic system can be iterated to generate the second chaotic series.

[0148] b3) Using the hyperchaotic system to generate four second chaotic sequences E, G, U, and V, namely:

[0149]

[0150] b4) using the second chaotic sequences E and G to perform bidirectional diffusion on the first encrypted image; and using the second chaotic sequences U and V to perform row and column scrambling on the first encrypted image.

[0151] Embodiment 12:

[0152] A three-dimensional data encryption method, the steps are as follows:

[0153] Step 1:

[0154] Obtain a sequence of images corresponding to the target 3D model data. A sequence of images generally refers to a group of images continuously collected for a certain scene, and a group of sequence images can be equivalent to an image of a 3D model.

[0155] Step 2:

[0156] The sequence images are normalized, and then the normalized sequence images are segmented using a pre-trained 3-D fully convolutional residual U-net network model to obtain multiple segmented images. Using a fully convolutional model to achieve three-dimensional image segmentation can make the segmentation results more accurate and achieve a smoother segmentation effect.

[0157] Step 3:

[0158] The low-frequency subbands of multiple segmented images are extracted by discrete wavelet transform method to compress the images, reduce the amount of data to be encrypted, and thus improve the encryption efficiency.

[0159] Step 4:

[0160] A first chaotic sequence is generated by LSC mapping, and then the pixel positions of the image are scrambled by a scrambling matrix generated by the first chaotic sequence.

[0161] Step 5:

[0162] The second chaotic sequence is generated by the hyperchaotic system, and then the scrambled image is bidirectionally diffused and row-column scrambled by the second chaotic sequence, and finally the encrypted sequence image corresponding to the target three-dimensional model is obtained.

[0163] Embodiment 13:

[0164] A three-dimensional data encryption method, the steps are as follows:

[0165] 1. 3D image segmentation:

[0166] Before 3D image segmentation, pre-prepared sequence image samples are obtained, and regions of interest in the sequence image samples are manually annotated as labels.

[0167] The labeled sequence image samples are normalized, and then the normalized sequence image samples are transferred to the pre-built 3-D fully convolutional residual U-net network model for supervised training until the convergence condition is reached, and then the trained 3-D fully convolutional residual U-net network model is obtained.

[0168] The 3-D fully convolutional residual U-net network model includes an encoder, a decoder, and an attention mechanism, which may include operations such as convolution, pooling, and deconvolution. The encoder includes 4 downsampling residual blocks. The size of the feature map remains unchanged after each convolution operation, and the size of the feature map becomes 1 / 2 of the original after each pooling operation. ReLU is used as a nonlinear activation function to increase the nonlinear expression ability of the network; the decoder includes 4 upsampling residual blocks. The size of the feature map becomes twice the original after each deconvolution operation. The encoder takes the preprocessed three-dimensional image as input, downsamples the image through a series of convolution, pooling, ReLU and other operations, and extracts a high-dimensional feature map as output; the decoder takes the high-dimensional feature map output by the encoder as input, upsamples the high-dimensional features through a series of convolution, deconvolution, ReLU and other operations, and obtains a segmentation result map of the same size as the original image as output; the attention mechanism inputs the shallow feature map of the encoder, and after convolution transformation, outputs a feature map of the same size, which is then spliced ​​with the corresponding feature map of the decoder to obtain more fine-grained segmentation information.

[0169] The sequence images corresponding to the target three-dimensional model data are normalized, and then segmented using the above-trained 3-D fully convolutional residual U-net network model to obtain multiple segmented images.

[0170] The three-dimensional image grayscale matrix can be obtained through the sequence of images, and then the three-dimensional image grayscale matrix is ​​normalized to 0-1, and finally the pixel spacing of the normalized image is adjusted to a preset value.

[0171] 2. Discrete wavelet transform:

[0172] The main purpose is to discard the high-frequency edge detail information and extract the low-frequency component to obtain an approximate image of the original image. Discrete wavelet transform can decompose the image in two dimensions to obtain four sub-band images, namely: low-frequency component, horizontal high-frequency component, vertical high-frequency component, and diagonal high-frequency component. The low-frequency component is a thumbnail of the image content, which contains most of the content information in the original image, and the remaining three components contain high-frequency edge information in different directions.

[0173] 3. First Chaos Encryption:

[0174] A first chaotic sequence is generated by logistic-sine-cosine mapping, and then the pixel positions of the image are scrambled by a scrambling matrix generated by the first chaotic sequence.

[0175] The representations of the logistic-sine-cosine mapping include:

[0176]

[0177] Among them, x is the state variable, r and β are control parameters, and F(r, x i ) and G(1-r, x i ) is a logistic and sine map. When β = -0.5, r∈[0, 1], x∈[0, 1), the system is in a chaotic state. The key determines the initial state of the LSC map. The length is set to 256 bits, and the key structure can be divided into 6 parts.

[0178] The first chaotic sequence is generated by using LSC mapping with an initial state, and then the first chaotic sequence is used to scramble the global pixel position, and the positions of pixel rows and columns are changed at the same time, so that the strong correlation between adjacent pixels can be effectively reduced. When the pixel position is scrambled, the following steps are included: the initial state (x0, p) is input into the LSC mapping to generate a random sequence of length R+W, and the sequence is divided into a subsequence A of length R and a subsequence B of length W; the two subsequences are sorted respectively to obtain two index vectors IA and IB; a matrix of size R*W is initialized, and each column of the matrix is ​​set to IB; each element of IA is used to move each column of the matrix to obtain a matrix O for scrambling; the column index j=1 is initialized; the position of {(1, O 1，j ), (2, O 2，j ),…,(R,O R，j )}; treat these pixels as a string connected together and move them upward by O 1，j positions, and obtain the corresponding scrambled pixel positions; for j, stack from 2 to W, repeat the above steps R-1 times.

[0179] 4. Second Chaos Encryption:

[0180] Based on the Lorenz system, a nonlinear controller w is introduced to form a hyperchaotic system, which can generate 4 different second chaotic sequences. The hyperchaotic system has high system dimension and complex structure. It is a mature chaotic system that helps to improve the security performance of encryption schemes.

[0181] The representation of a hyperchaotic system is:

[0182]

[0183] in, are the change rates of x, y, z, and w respectively. When the system parameters a=10, b=8 / 3, c=28, and r=-1, the system is a hyperchaotic system.

[0184] The generation process of the initial value of the hyperchaotic system is related to the integer randomly input by the user and the pixel value of the plaintext image. The initial value of the hyperchaotic system (Q x0 , Qy0 , Q z0 , Q w0 ) is represented as:

[0185]

[0186] Among them, mod(a, b) is the remainder of a / b, and fix(a) is the integer part of a.

[0187] The representation of the four second chaotic sequences E, G, U, and V is:

[0188] E=mod[fix(ei×10 5 ), 256]

[0189] G = mod[fix(gi×10 5 ), 256]

[0190] U=mod[fix((ui+100)×10 10 ), 256]+1

[0191] V = mod[fix((vi+100)×10 10 ), 256]+1

[0192] The processed chaotic sequences E and G are used in the subsequent diffusion algorithm of image pixel values, and the sequences U and V are used for row and column scrambling.

[0193] In order to enhance the security of the algorithm, bidirectional diffusion processing is performed on the scrambled image, with the aim of changing every pixel of the image, making the information in the image difficult to be read by external attackers.

[0194] Row and column scrambling is used to further disrupt the positions of image pixels, enhance the randomness and security of the encryption algorithm, and obtain the final ciphertext image.

[0195] 5. Decryption:

[0196] The same key is used for encryption and decryption, so the decryption process is the inverse operation of the encryption process.

Claims

1. A three-dimensional data encryption method, characterized in that: The following steps are involved: 1) Obtain a sequence of images corresponding to the three-dimensional geographic space. 2) preprocessing the sequence images to obtain preprocessed sequence images; 3) Construct a 3-D fully convolutional residual U-net network model; 4) Using the 3-D fully convolutional residual U-net network model to segment the preprocessed sequence images to obtain multiple segmented images; 5) Using discrete wavelet transform method to extract low-frequency subbands of the segmented image and obtain an approximate image; 6) Using the first chaotic sequence to scramble the pixel positions of the approximate image to obtain a first encrypted image; 7) Using the second chaotic sequence to perform bidirectional diffusion and row-column scrambling on the first encrypted image, to obtain the final encrypted image.

2. A three-dimensional data encryption method according to claim 1, characterized in that: The method for acquiring the sequence images corresponding to the three-dimensional geographic space is: continuously photographing the three-dimensional geographic space to acquire a set of sequence images.

3. A three-dimensional data encryption method according to claim 1, characterized in that: The method of acquiring the sequence images corresponding to the three-dimensional geographic space is: modeling the three-dimensional geographic space to obtain the three-dimensional geographic space model; acquiring a set of sequence images of the three-dimensional geographic space model; the tools for modeling the three-dimensional geographic space include 3dmax and Maya.

4. A three-dimensional data encryption method according to claim 1, characterized in that: Preprocessing includes normalization.

5. A three-dimensional data encryption method according to claim 4, characterized in that: The steps of preprocessing the sequence images also include: 1) Stack the sequence images to form a three-dimensional image grayscale matrix, where each layer of the matrix represents a sequence image; 2) Normalize the pixel values ​​in the grayscale matrix of the three-dimensional image to between 0 and 1; 3) Adjust the pixel spacing of pixels in the three-dimensional matrix so that the pixel spacing meets the preset requirements of network training or application.

6. A three-dimensional data encryption method according to claim 1, characterized in that: The 3-D fully convolutional residual U-net network model includes an encoder and a decoder; The encoder comprises a plurality of downsampled residual blocks; The encoder downsamples the preprocessed sequence images through convolution, pooling, and ReLU operations to output a high-dimensional feature map; The decoder comprises a plurality of upsampled residual blocks; The decoder upsamples the high-dimensional feature map through convolution, deconvolution, and ReLU operations to obtain a segmented image.

7. A three-dimensional data encryption method according to claim 6, characterized in that: The 3-D fully convolutional residual U-net network model introduces an attention mechanism: taking the high-dimensional feature map of the encoder as input, outputting a feature map of the same size through a convolution transformation operation, and then splicing it with the image of the decoder to obtain fine-grained segmentation information.

8. A three-dimensional data encryption method according to claim 1, characterized in that: The 3-D fully convolutional residual U-net network model is trained by normalized sequence image samples with labels. The three-dimensional image grayscale matrix is ​​obtained through the sequence image, and then the three-dimensional image grayscale matrix is ​​normalized to 0-1, and finally the pixel spacing of the normalized image is adjusted to a preset value.

9. A three-dimensional data encryption method according to claim 1, characterized in that: The step of scrambling the pixel positions of the approximate image by using the first chaotic sequence comprises: 1) Generate a first chaotic sequence with a length of R+W using logistic-sine-cosine mapping; The mapping process is as follows: Among them, x i is the state variable, r and β are the control parameters, F(r, x i ) and G(1-r, x i ) are logistic and sine mappings. When β = -0.5, r∈[0,1], x∈[0,1), the system is in a chaotic state; x i+1 is the variable after mapping; 2) Divide the first chaotic sequence into a subsequence A with a length of R and a subsequence B with a length of W; Sort the two subsequences respectively to obtain two index vectors IA and IB; 3) Initialize a matrix of size R*W and set each column of the matrix to IB; use each element of IA to move each column of the matrix to obtain the matrix O for scrambling; 4) Initialize column index j = 1; find the position {(1, O 1，j ), (2, O 2，j ),…,(R,O R，j )} pixels, move them upward by O 1，j Positions, get the corresponding scrambled pixel position; R，j is the element in the matrix O; 5) Let j=j+1, return to step 4), repeat R-1 times, and obtain the first encrypted image.

10. A three-dimensional data encryption method according to claim 1, characterized in that: The steps of bidirectionally diffusing and scrambling rows and columns of the first encrypted image using the second chaotic sequence include: 1) Construct a hyperchaotic system, namely: in, are the rates of change of x, y, z, and w respectively. When the system parameters a=10, b=8 / 3, c=28, and r=-1, the system is a hyperchaotic system. x, y, z, and w are the generated values ​​of the hyperchaotic system. 2) Hyperchaotic system generates initial value (Q x0 , Q y0 , Q z0 , Q w0 ),Right now: Where, mod(a, b) is the remainder of a / b, fix(a) is the integer part of a; k1, k2, k3, k4 are integers; 3) With the initial value (Q x0 , Q y0 , Q z0 , Q w0 ) is used as a reference, and four second chaotic sequences E, G, U, and V are generated using the hyperchaotic system, namely: Where ei, gi, ui, and vi are user input parameters; 4) Using the second chaotic sequences E and G to perform bidirectional diffusion on the first encrypted image; using the second chaotic sequences U and V to perform row and column scrambling on the first encrypted image.

Images