Deep learning-based image security processing method and system, and related equipment

Abstract

The invention is suitable for the technical field of image security processing, and provides an image security processing method and system based on deep learning, and related equipment, and the method comprises the steps: obtaining image data information shot by a monitoring camera in real time through a server, and extracting a target feature region in the image data information through a convolutional neural network model; encrypting data of the target feature region through an AES encryption algorithm, and transmitting a generated first-level ciphertext to a cloud platform; the cloud platform encrypts the first-level ciphertext by using an ABE algorithm to obtain a second-level ciphertext; when an access request of the mobile terminal is received, the cloud platform judges whether the access request of the mobile terminal meets an access condition or not; if yes, selecting attribute information according to the access request, and determining a target data transmission protocol; and the cloud platform sends the second-level ciphertext to the mobile terminal according to the target data transmission protocol, so that the mobile terminal restores the image data information according to a preset layered decryption framework, and the experience feeling of image operation of a user is effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of image security processing technology, and in particular to an image security processing method, system and related equipment based on deep learning. Background Art

[0002] In today's security monitoring and remote viewing scenarios, cameras are widely deployed in various places to collect image and video data, and transmit them to cloud platforms for storage, so that users can view them at any time through their mobile phones. In this process, there are security issues with image processing. On the one hand, ordinary unencrypted network protocols are generally used in the transmission link, such as HTTP to transmit images or video streams. The data is easily intercepted by hackers on the network, and they can easily obtain sensitive information. Whether it is private life scenes in home monitoring or commercial secrets in corporate monitoring, there is a risk of privacy leakage. On the other hand, the camera side may use complex and inefficient encryption algorithms, resulting in a large amount of computing resources being occupied when encrypting images or videos, causing significant delays in data transmission. For example, the real-time monitoring screen may be stuck and cannot be displayed smoothly, which seriously affects the user experience.

[0003] In view of this, a deep learning-based image security processing method, system and related equipment are needed. Summary of the invention

[0004] The embodiments of the present application provide a deep learning-based image security processing method, system, and related equipment to solve the problem of poor user experience.

[0005] A first aspect of an embodiment of the present application provides an image security processing method based on deep learning, comprising:

[0006] The server obtains image data information captured in real time by the surveillance camera, and uses a convolutional neural network model to extract target feature areas in the image data information;

[0007] The server encrypts the data of the target feature area by using an AES encryption algorithm, and transmits the generated primary ciphertext to the cloud platform;

[0008] The cloud platform encrypts the received first-level ciphertext using the ABE algorithm to obtain a second-level ciphertext;

[0009] When receiving an access request from a mobile terminal, the cloud platform determines whether the access request from the mobile terminal satisfies an access condition;

[0010] If so, the cloud platform selects attribute information according to the access request, and determines a target data transmission protocol based on the selected attribute information;

[0011] The cloud platform sends the secondary ciphertext to the mobile terminal according to the target data transmission protocol, so that the mobile terminal restores the original image data information according to a preset hierarchical decryption framework.

[0012] Furthermore, the server obtains image data information captured in real time by the surveillance camera, and uses a convolutional neural network model to extract target feature areas in the image data information, including:

[0013] Extracting target feature vectors of video data information and picture data information in the image data information using a convolutional neural network model;

[0014] Determine a key frame based on the target feature vector of the video data information, and determine a first feature area through the key frame;

[0015] Determine a second feature area based on the target feature vector of the image data information;

[0016] A target feature region is determined according to the first feature region and the second feature region.

[0017] Furthermore, the determining of a key frame based on the target feature vector of the video data information, and determining the first feature area through the key frame, includes:

[0018] Calculating a feature metric value based on a target feature vector of the video data information, and determining a key frame set according to the feature metric value when the feature metric value is greater than or equal to a set threshold;

[0019] The feature metric value M f :

[0020]

[0021] Among them: f is the key frame, is the target feature vector, and d is the dimension of the target feature vector.

[0022] Furthermore, the cloud platform uses the ABE algorithm to encrypt the received first-level ciphertext to obtain a second-level ciphertext, including:

[0023] Set the attribute information, master public key and master private key of the ABE algorithm, and build the access structure;

[0024] The first-level ciphertext received by the cloud platform is encrypted according to the attribute information, the master public key, the master private key and the access structure to obtain the second-level ciphertext.

[0025] Furthermore, when receiving an access request from a mobile terminal, the cloud platform determines whether the access request from the mobile terminal satisfies an access condition, including:

[0026] Extracting user-related attributes and device attributes of the mobile terminal based on the access request;

[0027] The user-related attributes and the device attributes of the mobile terminal are compared one by one with the preset access conditions, and the preset access conditions are a series of access conditions set according to the security policy and business requirements of the system.

[0028] Furthermore, if so, the cloud platform selects attribute information according to the access request, and determines the target data transmission protocol based on the selected attribute information, including:

[0029] When the access request of the mobile terminal satisfies the access condition, selecting a relevant attribute subset based on the access request of the mobile terminal;

[0030] weights are set according to the subset of related attributes, and scores of various data transmission protocols in the constructed database are calculated based on the set weights.

[0031] The scores of various data transmission protocols are compared to determine the target data transmission protocol.

[0032] Furthermore, the cloud platform sends the secondary ciphertext to the mobile terminal according to the target data transmission protocol, so that the mobile terminal restores the original image data information according to a preset hierarchical decryption framework, including:

[0033] The secondary ciphertext is encapsulated according to the format corresponding to the target data transmission protocol and then sent to the mobile terminal, so that the mobile terminal performs first-layer decryption and second-layer decryption respectively according to a preset layered decryption framework to obtain the original image data information.

[0034] Furthermore, the mobile terminal is caused to perform first-layer decryption and second-layer decryption respectively according to a preset layered decryption framework to obtain original image data information, including:

[0035] The mobile terminal determines a decryption key according to the encryption algorithm type of the secondary ciphertext, and performs a first-level decryption based on the secondary ciphertext and the corresponding decryption key to obtain a primary ciphertext;

[0036] The mobile terminal determines a decryption key according to the encryption algorithm type of the first-level ciphertext, and performs a second-level decryption based on the first-level ciphertext and the corresponding decryption key to obtain original image data information.

[0037] A second aspect of an embodiment of the present application provides an image security processing system based on deep learning, comprising:

[0038] A target feature region extraction unit is used for the server to obtain image data information captured in real time by a surveillance camera, and to extract the target feature region in the image data information using a convolutional neural network model;

[0039] A first-level ciphertext generation and transmission unit, used for the server to encrypt the data of the target feature area through the AES encryption algorithm and transmit the generated first-level ciphertext to the cloud platform;

[0040] A secondary ciphertext generating unit, used for the cloud platform to encrypt the received primary ciphertext using an ABE algorithm to obtain a secondary ciphertext;

[0041] An access request judging unit, configured to, when receiving an access request from a mobile terminal, cause the cloud platform to judge whether the access request from the mobile terminal satisfies an access condition;

[0042] a target data transmission protocol determining unit, configured to, when the access request of the mobile terminal satisfies the access condition, cause the cloud platform to select attribute information according to the access request, and determine the target data transmission protocol based on the selected attribute information;

[0043] The image data information restoration unit is used for the cloud platform to send the secondary ciphertext to the mobile terminal according to the target data transmission protocol, so that the mobile terminal can restore the original image data information according to a preset hierarchical decryption framework.

[0044] A third aspect of an embodiment of the present application provides a computer device, including a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor, wherein when the processor executes the computer-readable instructions, the steps of the deep learning-based image security processing method as described in any one of the above are implemented.

[0045] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0046] The present invention obtains image data information captured by a surveillance camera in real time, and uses a convolutional neural network model to extract target feature vectors from video and picture data information, and further determines key frames, different types of feature areas and the final target feature area, which provides an important data basis for subsequent data processing; the first-level encryption is performed by the AES encryption algorithm, and the first-level ciphertext obtained retains part of the encrypted information of the image, which protects the key data in the image to a certain extent; the first-level ciphertext is encrypted by the ABE algorithm to generate a second-level ciphertext with a higher security protection level, which effectively realizes data security protection and access control; it can accurately judge whether the mobile terminal access request meets the access conditions, and if the conditions are met, reasonably determine the target data transmission protocol based on the attribute information in the request, and realize efficient and secure data transmission; the second-level ciphertext is securely transmitted to the mobile terminal by using the target data transmission protocol, and the original image data information is accurately restored through the layered decryption framework, which ensures the security of the data during transmission and processing while meeting the viewing and use requirements of the mobile terminal for the surveillance image data, thereby improving the user experience.

[0047] Other advantages, objectives, and features of the present invention will be set forth in part in the following description, and in part will be apparent to those skilled in the art based on an examination of the following or may be taught from the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A schematic diagram of an embodiment of a deep learning-based image security processing method in the present invention;

[0049] Figure 2 The figure is a flowchart of another embodiment of a method for image security processing based on deep learning in the present invention. DETAILED DESCRIPTION

[0050] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein, for example. In addition, the terms "including" and "corresponding to" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0051] Embodiment 1

[0052] See also Figure 1 , an embodiment of an image security processing method based on deep learning in the present invention comprises the following steps:

[0053] S11. The server obtains the image data information captured by the surveillance camera in real time, and uses the convolutional neural network model to extract the target feature area in the image data information;

[0054] In this embodiment, the system obtains information by connecting to the surveillance camera in a wired connection or a network protocol communication mode, and the acquired image data information includes picture data information and video data information. The picture data information includes pixel data, resolution information, and image format information; the video data information includes frame sequence data, timestamp information, and audio data. Here, the surveillance camera will select an appropriate method to provide various types of image data information based on factors such as hardware configuration, software functions, and user settings.

[0055] The image data obtained above is input into a pre-trained convolutional neural network (CNN) model. The CNN model extracts features from the image through its convolutional layer, pooling layer and other structures, automatically learns the feature representations of different targets in the image, and outputs the target feature area in the image data information, so that the system selects a suitable encryption algorithm to encrypt the data in this area.

[0056] S12. The server encrypts the data of the target feature area using the AES encryption algorithm and transmits the generated first-level ciphertext to the cloud platform;

[0057] For each feature region in the key frame determined above, the data in the region is encrypted using the AES encryption algorithm. The encryption process converts the original image data of the feature region into ciphertext to obtain the first-level ciphertext. Here, the first-level ciphertext retains part of the encrypted information of the image, protecting the key data in the image to a certain extent.

[0058] S13. The cloud platform uses the ABE algorithm to encrypt the received first-level ciphertext to obtain a second-level ciphertext; step S13 also includes the following steps:

[0059] 1. Set the attribute information, master public key and master private key of the ABE algorithm, and build the access structure;

[0060] First, it is necessary to determine the set U of all attributes that may be involved in the entire encryption system. These attributes include but are not limited to data sources, such as specific surveillance camera IDs, surveillance areas to which departments belong, etc., user permissions, such as administrator permissions, read-only permissions, etc., data types, such as video data, image data, etc., and time-related attributes, such as the time period for data generation.

[0061] Run the initialization algorithm Setup(λ, U) of the ABE algorithm, where λ is the security factor, which is used to determine the security characteristics such as the length of the key. The algorithm will generate the master public key PK and the master private key MK. The master public key PK will be publicly used for encryption operations, while the master private key MK must be kept strictly confidential for subsequent operations such as generating user private keys.

[0062] The system analyzes the data attributes related to the first-level ciphertext received from the cloud platform. For example, the system obtains the camera ID, data type, and possible timestamp of the data source from the metadata or accompanying identification information of the first-level ciphertext to form a data attribute set A. data ={Camera ID=C1, data type-video data, time period-T1}. At the same time, according to the access policy set by the system and the current business scenario, determine the user attributes that may affect data access, such as user authority level, department, etc., to form a user attribute set A user .

[0063] According to the system's security policy and business requirements, an access structure A is constructed. The access structure specifies what attribute combinations can access encrypted data. For example, a possible access structure can be: only when the user's authority level is administrator authority and the department to which he belongs is the security department, and the data source is the surveillance camera in a specific area, can the relevant data be accessed. user and data attribute set A data Associate with the constructed access structure A to ensure that subsequent encryption operations can implement appropriate access control based on this access structure.

[0064] 2. Encrypt the first-level ciphertext received by the cloud platform according to the attribute information, master public key, master private key and access structure to obtain the second-level ciphertext.

[0065] Using the above PK, A data and A calls the ABE encryption algorithm Encrypt(PK, C1, A data , A), where C1 is the received first-level ciphertext. By performing encryption operations on the first-level ciphertext, the second-level ciphertext C2 is generated. Based on C1, C2 integrates various attribute information and access structure requirements to achieve a higher level of security protection. The cloud platform stores the generated second-level ciphertext C2 so that it can be processed accordingly according to the access request of the mobile terminal.

[0066] S14. When receiving the access request from the mobile terminal, the cloud platform determines whether the access request from the mobile terminal meets the access conditions; step S14 also includes the following steps:

[0067] 1. Extract user-related attributes and mobile device attributes based on the access request;

[0068] 2. Compare the user-related attributes and mobile device attributes with the preset access conditions one by one. The preset access conditions are a series of access conditions set according to the system's security policies and business requirements.

[0069] A series of access conditions are set according to the security policies and business requirements of the system. These access conditions may be based on various factors, such as data sensitivity, user authority level and role, device security level, etc. For example, for highly sensitive surveillance video data, the access conditions may stipulate that only administrators of the security department and using devices with high security levels through a specific network can access it. These preset access condition information is extracted from the system configuration or database for subsequent comparison and judgment. The user-related attributes and the mobile device attributes contained in the mobile access request are compared and analyzed with the extracted access conditions one by one. Taking the user authority level as an example, if the access condition requires the visitor to have administrator authority, and the user authority level in the mobile access request is ordinary user authority, then the access condition is not met in this item. Similarly, similar comparison and judgment are performed for attributes such as device security level and network type. Only when all relevant attributes in the mobile access request meet the preset access conditions, the access request is judged to meet the access conditions; otherwise, it is judged to not meet the access conditions.

[0070] S15. When the access request from the mobile terminal meets the access conditions, the cloud platform selects attribute information according to the access request and determines the target data transmission protocol based on the selected attribute information; step S15 also includes the following:

[0071] 1. When the access request from the mobile terminal meets the access condition, a relevant attribute subset is selected based on the access request from the mobile terminal;

[0072] When the access request from the mobile terminal meets the access conditions, attribute information that is helpful in determining the data transmission protocol is selected from the access request. This attribute information is not all the information in the access request, but the part that is closely related to the transmission protocol selection after screening. For example, attributes such as the user's permission level, network type, device security level, and requested resource type may be selected. This is because these attributes can greatly affect the security and efficiency requirements of data transmission.

[0073] 2. Set weights according to the relevant attribute subsets, and calculate the scores of each data transmission protocol in the constructed database based on the set weights.

[0074] The system pre-builds a database containing various data transmission protocols and their characteristics. Each transmission protocol in the database has a detailed record, including its encryption strength, transmission rate, resource consumption, applicable scenarios and other performance indicators. According to the importance of data security and transmission efficiency, weights are assigned to different performance indicators such as encryption strength, transmission rate, resource consumption, etc. The score of each transmission protocol in the database is calculated based on the selected attribute information and the set weight.

[0075] 3. Compare the scores of various data transmission protocols to determine the target data transmission protocol.

[0076] Compare the scores of all transport protocols and select the transport protocol with the highest score as the target data transmission protocol. If there are multiple protocols with the same or similar scores, further screening can be performed based on other factors.

[0077] S16. The cloud platform sends the secondary ciphertext to the mobile terminal according to the target data transmission protocol, so that the mobile terminal restores the original image data information according to the preset layered decryption framework. Step S16 also includes the following:

[0078] The secondary ciphertext is encapsulated according to the format corresponding to the target data transmission protocol and sent to the mobile terminal, so that the mobile terminal performs the first layer decryption and the second layer decryption respectively according to the preset layered decryption framework to obtain the original image data information.

[0079] The mobile terminal determines the decryption key according to the encryption algorithm type of the secondary ciphertext, and performs the first-level decryption based on the secondary ciphertext and the corresponding decryption key to obtain the primary ciphertext;

[0080] The mobile terminal determines the decryption key according to the encryption algorithm type of the first-level ciphertext, and performs the second-level decryption based on the first-level ciphertext and the corresponding decryption key to obtain the original image data information.

[0081] Specifically, the system encapsulates the secondary ciphertext in the format specified by the target data transmission protocol. Different transmission protocols have different encapsulation methods. For example, some protocols may require the addition of specific header information, checksums, etc. After the encapsulation is completed, the encapsulated secondary ciphertext is sent to the mobile terminal through the established transmission connection. In addition, the mobile terminal will store the received secondary ciphertext in a local temporary buffer area or a specific storage area for subsequent decryption operations. The mobile terminal pre-stores the key information corresponding to the encryption process. Before performing the decryption operation, you need to obtain the corresponding key first.

[0082] First layer decryption:

[0083] The mobile terminal calls the corresponding ABE decryption algorithm library according to the encryption algorithm type of the secondary ciphertext. Here, the mobile terminal has integrated the implementation code of the algorithm. The obtained secondary ciphertext and the corresponding ABE algorithm decryption key are passed to the decryption algorithm as input parameters. For example, the ABE decryption algorithm needs to perform complex bilinear pairing operations, calculate according to the elements in the key and the elements in the ciphertext, and restore the primary ciphertext or the intermediate data related to the primary ciphertext. After completing the first layer of decryption, the decryption result can be preliminarily verified.

[0084] Second layer decryption:

[0085] Call the corresponding decryption algorithm according to the encryption algorithm of the first-level ciphertext. For the AES algorithm, you need to set the correct key length, such as 128 bits or 256 bits, encryption mode, such as CBC mode and other parameters. Pass the first-level ciphertext and the corresponding symmetric encryption key as input parameters to the decryption algorithm. For example, for the AES-CBC mode, you need to first process the ciphertext with an initial vector, and then decrypt the ciphertext blocks one by one through a series of round function operations, and finally restore the original image feature area data. Similarly, after completing the second-level decryption, the decryption result needs to be verified again.

[0086] Finally, if the image data is specially processed during the encryption process, such as encrypting only the feature area as mentioned above, and then associating through key frames, etc., after restoring the data of each feature area, it is necessary to reassemble the data of these feature areas into the corresponding key frames to restore the complete original image data information. For example, the reorganization is performed based on the position information of the feature area in the key frame recorded during encryption to ensure the integrity and accuracy of the image data. After restoring the original image data information, the user can view, analyze, store and perform other subsequent operations through the mobile terminal.

[0087] Embodiment 2

[0088] See also Figure 2 , the server uses a convolutional neural network model to extract the target feature area in the image data information, and also includes the following steps:

[0089] S21. Extracting target feature vectors of video data information and picture data information from image data information using a convolutional neural network model;

[0090] In this embodiment, the convolutional neural network (CNN) is composed of multiple convolutional layers, pooling layers and fully connected layers. The convolutional layer performs convolution operations by sliding the convolution kernel on the image data to extract local features; the pooling layer is used to downsample the output of the convolutional layer to reduce the data dimension while retaining the main features; the fully connected layer is used to comprehensively process the features extracted by the previous layers and finally output the target feature vector.

[0091] First, the acquired video stream is split into a series of picture frames in chronological order, and each picture frame is fed into the CNN model as independent input data. Assuming that the video frame rate is f (unit: frames / second) and the total video duration is T (unit: seconds), then the total number of video frames N = f×T. Here, the size of each picture frame is h×w (unit: pixel), where h is the height, w is the width, and each pixel has c color channels. Therefore, the video data information input to the CNN model is represented as a four-dimensional tensor X video =[N, h, w, c]. Next, the acquired image data is sent as input to the CNN model. Assuming that the image size is H×W (unit: pixel), and each pixel has c color channels, the image data information input to the CNN model can be represented as a three-dimensional tensor X image =[H, W, c].

[0092] In the convolution layer, the size of the convolution kernel is set to k×k (unit: pixel), the step size is s (unit: pixel), and the padding method is p. For the input tensor X, the calculation formula for the output tensor Y after the convolution layer is:

[0093]

[0094]

[0095] Where: h out and w out Represents the dimensions of the output tensor Y in the height and width directions respectively.

[0096] The calculation formula for the convolution operation is:

[0097]

[0098] in: is the element value of the output tensor Y of the l-th convolutional layer at position (i, j), The element value of the input tensor X at position (i+m, j+n) for the l-1th layer, is the element value of the l-th layer convolution kernel at position (m,n), b l is the bias term of the lth convolutional layer.

[0099] Through the continuous operation of multiple convolutional layers, local features at different levels in the image data information are continuously extracted, and finally the feature tensor Z processed by the convolutional layer is obtained. Similarly, for video data information, it is the feature tensor obtained for each picture frame, and for picture data information, it is the feature tensor obtained for the entire picture.

[0100] The pooling layer uses the maximum pooling method to downsample the feature tensor Z output by the convolution layer. The specific steps are similar to the above operation. Through the pooling operation, the data dimension is further reduced, the main features are retained, and the feature tensor P processed by the pooling layer is obtained. After being processed by multiple convolutional layers and pooling layers, the obtained feature tensor P is flattened into a one-dimensional vector and then sent to the fully connected layer. Assuming that the fully connected layer has n neurons, the output calculation formula of the fully connected layer is:

[0101]

[0102] Where: y j is the output value of the jth neuron, w ij is the weight connecting the i-th input neuron and the j-th output neuron, x i is the input value of the i-th input neuron, that is, the element value of the flattened feature tensor P, b j is the bias term of the jth fully connected layer neuron.

[0103] After the processing of the above fully connected layer, the target feature vector V is finally output. For video data information, the target feature vector V is obtained for each picture frame. For image data information, it is the target feature vector V obtained for the entire image image .

[0104] S22. Determine a key frame based on the target feature vector of the video data information, and determine the first feature area through the key frame; step S22 also includes the following:

[0105] A feature metric value is calculated based on a target feature vector of the video data information. When the feature metric value is greater than a set threshold, a key frame set is determined according to the feature metric value.

[0106] Specifically, a threshold T1 is set, which is used to determine whether each frame in the video data information is a key frame. For each frame f in the video, its corresponding target feature vector is calculate The characteristic metric value M f , the formula is as follows:

[0107]

[0108] Where: d is the dimension of the target feature vector.

[0109] If M f ≥T1, the frame f is determined to be a key frame. In this way, a key frame set F1 is screened out from the video data information.

[0110] For each key frame f∈F1, its corresponding target feature vector is use Calculate with the pre-trained position regression model to obtain the position information of the feature area of ​​key frame f. Suppose the input of the position regression model is The output is the location of the feature area (x start ,y start , x end ,y end ), where x start and start are the starting horizontal and vertical coordinates of the feature area in the key frame, x end and end The ending horizontal and vertical coordinates of the feature region in the key frame. In this way, the first feature region in each key frame f is determined

[0111] S23. Determine a second feature region based on the target feature vector of the image data information;

[0112] Specifically, a threshold T2 is also set. For the image data information, its target feature vector is V image , calculate V image The characteristic metric value M image , if M image ≥T2, then according to V image Calculate with the pre-trained position regression model to obtain the position information R of the feature area in the image image .

[0113] S24. Determine a target feature region based on the first feature region and the second feature region.

[0114] All the first feature areas determined from the video data information are and a second feature region R determined from the image data information image Integrate and sort these feature regions according to the order of appearance in the original data, and finally obtain the target feature region set R target , R target It contains all the feature areas that need to be focused on and processed, which are filtered out from the video and image data information.

[0115] The above embodiment uses a convolutional neural network model to extract target feature vectors from video and image data information, and further determines key frames, different types of feature areas and the final target feature area, providing an accurate and important data foundation for subsequent data processing.

[0116] Embodiment 3

[0117] An embodiment of an image security processing system based on deep learning in the present invention includes the following steps:

[0118] The target feature region extraction unit is used for the server to obtain the image data information captured by the surveillance camera in real time, and to extract the target feature region in the image data information using a convolutional neural network model;

[0119] The first-level ciphertext generation and transmission unit is used for the server to encrypt the data of the target feature area through the AES encryption algorithm and transmit the generated first-level ciphertext to the cloud platform;

[0120] The secondary ciphertext generation unit is used for the cloud platform to encrypt the received primary ciphertext using the ABE algorithm to obtain the secondary ciphertext;

[0121] An access request judgment unit, configured to, when receiving an access request from a mobile terminal, cause the cloud platform to judge whether the access request from the mobile terminal satisfies an access condition;

[0122] A target data transmission protocol determination unit, configured to, when the access request of the mobile terminal satisfies the access condition, select attribute information according to the access request, and determine the target data transmission protocol based on the selected attribute information;

[0123] The image data information restoration unit is used for the cloud platform to send the secondary ciphertext to the mobile terminal according to the target data transmission protocol, so that the mobile terminal can restore the original image data information according to the preset hierarchical decryption framework.

[0124] Embodiment 4

[0125] The present invention provides a computer device, comprising a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor. When the processor executes the computer-readable instructions, the steps of the above method are implemented.

[0126] Those of ordinary skill in the art will appreciate that the units of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition of each example has been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0127] In the embodiments provided by the present invention, it should be understood that the division of units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored. In addition, each functional unit in each embodiment of the present invention can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units.

[0128] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-0nlyMemory), random access memory (RAM, RandomAccessMemory), mobile hard disk, magnetic disk or optical disk, etc., which can store program code.

[0129] It can be understood that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein by equivalents. These modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and specification of the present invention.

Claims

1. A deep learning-based image security processing method, characterized in that: include: The server obtains image data information captured in real time by the surveillance camera, and uses a convolutional neural network model to extract target feature areas in the image data information; The server encrypts the data of the target feature area by using an AES encryption algorithm, and transmits the generated primary ciphertext to the cloud platform; The cloud platform encrypts the received first-level ciphertext using the ABE algorithm to obtain a second-level ciphertext; When receiving an access request from a mobile terminal, the cloud platform determines whether the access request from the mobile terminal satisfies an access condition; If so, the cloud platform selects attribute information according to the access request, and determines a target data transmission protocol based on the selected attribute information; The cloud platform sends the secondary ciphertext to the mobile terminal according to the target data transmission protocol, so that the mobile terminal restores the original image data information according to a preset hierarchical decryption framework.

2. The image security processing method based on deep learning according to claim 1 is characterized in that: The server obtains image data information captured in real time by the surveillance camera, and uses a convolutional neural network model to extract target feature areas in the image data information, including: Extracting target feature vectors of video data information and picture data information in the image data information using a convolutional neural network model; Determine a key frame based on the target feature vector of the video data information, and determine a first feature area through the key frame; Determine a second feature area based on the target feature vector of the image data information; A target feature region is determined according to the first feature region and the second feature region.

3. The image security processing method based on deep learning according to claim 2 is characterized in that: The step of determining a key frame based on the target feature vector of the video data information, and determining a first feature area through the key frame includes: Calculating a feature metric value based on a target feature vector of the video data information, and when the feature metric value is greater than or equal to a set threshold, determining a key frame set according to the feature metric value; The feature metric value M f : Among them: f is the key frame, is the target feature vector, and d is the dimension of the target feature vector.

4. The image security processing method based on deep learning according to claim 1 is characterized in that: The cloud platform uses the ABE algorithm to encrypt the received first-level ciphertext to obtain the second-level ciphertext, including: Set the attribute information, master public key and master private key of the ABE algorithm, and build the access structure; The first-level ciphertext received by the cloud platform is encrypted according to the attribute information, the master public key, the master private key and the access structure to obtain the second-level ciphertext.

5. The image security processing method based on deep learning according to claim 1, characterized in that: When receiving an access request from a mobile terminal, the cloud platform determines whether the access request from the mobile terminal meets the access conditions, including: Extracting user-related attributes and device attributes of the mobile terminal based on the access request; The user-related attributes and the device attributes of the mobile terminal are compared one by one with the preset access conditions, and the preset access conditions are a series of access conditions set according to the security policy and business requirements of the system.

6. The image security processing method based on deep learning according to claim 1, characterized in that: If so, the cloud platform selects attribute information according to the access request, and determines a target data transmission protocol based on the selected attribute information, including: When the access request of the mobile terminal satisfies the access condition, selecting a relevant attribute subset based on the access request of the mobile terminal; weights are set according to the subset of related attributes, and scores of various data transmission protocols in the constructed database are calculated based on the set weights. The scores of various data transmission protocols are compared to determine the target data transmission protocol.

7. The image security processing method based on deep learning according to claim 1 is characterized in that: The cloud platform sends the secondary ciphertext to the mobile terminal according to the target data transmission protocol, so that the mobile terminal restores the original image data information according to a preset hierarchical decryption framework, including: The secondary ciphertext is encapsulated according to the format corresponding to the target data transmission protocol and then sent to the mobile terminal, so that the mobile terminal performs first-layer decryption and second-layer decryption respectively according to a preset layered decryption framework to obtain the original image data information.

8. The image security processing method based on deep learning according to claim 7 is characterized in that: The step of causing the mobile terminal to perform first-layer decryption and second-layer decryption respectively according to a preset layered decryption framework to obtain original image data information includes: The mobile terminal determines a decryption key according to the encryption algorithm type of the secondary ciphertext, and performs a first-level decryption based on the secondary ciphertext and the corresponding decryption key to obtain a primary ciphertext; The mobile terminal determines a decryption key according to the encryption algorithm type of the first-level ciphertext, and performs a second-level decryption based on the first-level ciphertext and the corresponding decryption key to obtain original image data information.

9. An image security processing system based on deep learning, characterized in that: include: A target feature region extraction unit is used for the server to obtain image data information captured in real time by a surveillance camera, and to extract the target feature region in the image data information using a convolutional neural network model; A first-level ciphertext generation and transmission unit, used for the server to encrypt the data of the target feature area through the AES encryption algorithm and transmit the generated first-level ciphertext to the cloud platform; A secondary ciphertext generating unit, used for the cloud platform to encrypt the received primary ciphertext using an ABE algorithm to obtain a secondary ciphertext; An access request judging unit, configured to, when receiving an access request from a mobile terminal, cause the cloud platform to judge whether the access request from the mobile terminal satisfies an access condition; a target data transmission protocol determining unit, configured to, when the access request of the mobile terminal satisfies the access condition, cause the cloud platform to select attribute information according to the access request, and determine the target data transmission protocol based on the selected attribute information; The image data information restoration unit is used for the cloud platform to send the secondary ciphertext to the mobile terminal according to the target data transmission protocol, so that the mobile terminal can restore the original image data information according to a preset hierarchical decryption framework.

10. A computer device comprising a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor, characterized in that: When the processor executes the computer-readable instructions, the steps of the deep learning-based image security processing method as described in any one of claims 1 to 8 are implemented.

Images