Intelligent community data processing method and system based on artificial intelligence

Abstract

The application embodiment provides an intelligent community data processing method and system based on artificial intelligence. The target image to be processed is divided according to the privacy levels of the basic stage and the derived stage, and the corresponding desensitization strategy is matched. Based on this, the image of different privacy levels is matched with the reasonable desensitization, and the relationship between data service and user privacy protection is better coordinated. When the target image is subjected to privacy level identification, the artificial intelligence model of the user data anonymous processing model is adopted, so that the efficient and accurate privacy level determination is ensured.

Description

Technical Field

[0001] This application relates to the field of data processing, particularly the field of image processing, and specifically to a smart community data processing method and system based on artificial intelligence. Background Technology

[0002] In smart communities, data collected and analyzed through ubiquitous IoT devices enables the interconnection of everything, facilitating residents' daily lives. For example, in smart senior living communities, data on the daily activities of elderly residents can be analyzed for fall detection, hazard warnings, and fraud prevention. However, while the Internet of Things brings convenience, it also poses risks of privacy breaches. For instance, cameras installed throughout community streets, internal roads, elevators, stairwells, and even indoor cameras used to monitor the elderly may capture images containing sensitive information such as faces and private parts. Leaks of this information could lead to varying degrees of loss and risk. Currently, data sensitivity processing in this area typically involves masking or removing descriptive information such as camera model, shooting date, location, and author, or adding adversarial noise to prevent AI recognition. Existing technologies for desensitizing community data focus on the final processing of sensitive information; there is still room for improvement in the content recognition of sensitive information. Summary of the Invention

[0003] The purpose of this invention is to provide a smart community data processing method and system based on artificial intelligence.

[0004] In a first aspect, embodiments of this application provide a smart community data processing method based on artificial intelligence, applied to a server. The method includes: acquiring a target image to be processed uploaded by a community camera device; extracting image information representation from the target image to obtain a target image information representation vector; extracting a basic-order privacy representation vector from the target image information representation vector to obtain a basic-order privacy representation vector; and identifying basic-order privacy information based on the basic-order privacy representation vector to obtain image privacy level probability variables corresponding to each basic-order privacy level; extracting a derived-order privacy representation vector from the target image information representation vector to obtain a derived-order privacy representation vector; and performing expression based on the basic-order privacy representation vector and the derived-order privacy representation vector. The interaction vectors are used to obtain the derivative-order interaction vectors. Then, based on the derivative-order interaction vectors, derivative-order privacy information is identified to obtain the image privacy level probability variables corresponding to each derivative-order privacy level. The privacy level classification accuracy corresponding to the derivative-order privacy level is higher than that corresponding to the basic-order privacy level. Based on the image privacy level probability variables corresponding to each basic-order privacy level and each derivative-order privacy level, the privacy level corresponding to the target image to be processed is obtained. The privacy level includes the target basic-order privacy level and the target derivative-order privacy level corresponding to the target basic-order privacy level. Based on the privacy level corresponding to the target image to be processed, the corresponding desensitization strategy is retrieved, and the target image to be processed is desensitized.

[0005] Optionally, the derivative-order interaction vector is obtained by performing expression vector interaction based on the basic-order private expression vector and the derived-order private expression vector, including: combining the basic-order private expression vector and the derived-order private expression vector to obtain a combined expression vector; extracting the jump error expression vector based on the combined expression vector and the derived-order private expression vector to obtain a jump error expression vector, and then integrating the jump error expression vector with feature elements to obtain a derived-order interaction vector; wherein, the extraction of the jump error expression vector based on the combined expression vector and the derived-order private expression vector to obtain the jump error expression vector includes: performing correlation projection calculation on the combined expression vector based on preset jump parameters to obtain a projected expression vector, and obtaining the vector sum value of the projected expression vector and the derived-order private expression vector to obtain the jump error expression vector.

[0006] Optionally, obtaining the privacy level of the target image to be processed based on the image privacy level probability variables corresponding to each basic privacy level and each derived privacy level includes: obtaining the image privacy level location; integrating the image privacy level probability variables corresponding to each basic privacy level and each derived privacy level based on the image privacy level location to obtain the location route probability variables; determining the target location route probability variable from the location route probability variables; and determining the privacy level of the target image to be processed based on the target location route probability variable; wherein, integrating the image privacy level probability variables corresponding to each basic privacy level and each derived privacy level based on the image privacy level location to obtain the location route probability variables includes: obtaining the image privacy level location; integrating the image privacy level probability variables corresponding to each basic privacy level and each derived privacy level based on the image privacy level location to obtain the location route probability variables; determining the target location route probability variable from the target location route probability variables; and determining the privacy level of the target image to be processed based on the target location route probability variable. The probability variables for each location route include: determining the image privacy level probability variable corresponding to the current basic privacy level from the image privacy level probability variables corresponding to each basic privacy level; obtaining the current derived privacy level corresponding to the current basic privacy level based on the image privacy level position, and determining the image privacy level probability variable corresponding to the current derived privacy level from the image privacy level probability variables corresponding to each derived privacy level; obtaining the multiplication result of the image privacy level probability variables corresponding to the current basic privacy level and the image privacy level probability variables corresponding to the current derived privacy level to obtain the current location route probability variable; and obtaining the location route probability variables by reviewing the image privacy level probability variables corresponding to each basic privacy level and the image privacy level probability variables corresponding to each derived privacy level.

[0007] Optionally, the method further includes: loading the target image to be processed into the target image desensitization processing model Model1; extracting image information representation from the target image to be processed according to the target image desensitization processing model Model1 to obtain the target image information representation vector; extracting the basic-order private representation vector from the target image information representation vector according to the target image desensitization processing model Model1 to obtain the basic-order private representation vector, and performing basic-order private information recognition based on the basic-order private representation vector to obtain the image privacy level probability variable corresponding to each basic-order privacy level; extracting the derived-order private representation vector from the target image information representation vector according to the target image desensitization processing model Model1 to obtain the derived-order private representation vector, performing representation vector interaction based on the basic-order private representation vector and the derived-order private representation vector to obtain the derived-order interaction vector, and then performing derived-order private information recognition based on the derived-order interaction vector to obtain the image privacy level probability variable corresponding to each derived-order privacy level.

[0008] Optionally, the target image desensitization processing model Model1 includes an image information representation extraction unit, a basic-level privacy information recognition unit, and a derived-level privacy information recognition unit. Loading the target image to be processed into the target image desensitization processing model Model1 includes: loading the target image to be processed into the image information representation extraction unit for image information representation extraction to obtain a target image information representation vector; loading the target image information representation vector into the basic-level privacy information recognition unit, extracting basic-level privacy representation vectors from the target image information representation vector based on the basic-level privacy information recognition unit to obtain basic-level privacy representation vectors, and obtaining basic-level privacy level support coefficients based on the basic-level privacy representation vectors to obtain image privacy level probability variables corresponding to each basic-level privacy level; loading the target image information representation vector and the basic-level privacy representation vector into the derived-level privacy information recognition unit, extracting derived-level privacy representation vectors from the target image information representation vector based on the derived-level privacy information recognition unit to obtain derived-level privacy representation vectors; and based on the basic-level privacy representation vectors... The basic-order privacy representation vector and the derived-order privacy representation vector interact to obtain the derived-order interaction vector. Then, based on the derived-order interaction vector, the derived-order privacy level support coefficient is obtained to obtain the image privacy level probability variable corresponding to each derived-order privacy level. The derived-order privacy information recognition unit includes a derived-order privacy representation vector extraction subunit, an expression vector interaction subunit, and a derived-order privacy information recognition subunit. The target image information representation vector and the basic-order privacy representation vector are loaded into the derived-order privacy information recognition unit, including: loading the target image information representation vector into the derived-order privacy representation vector extraction subunit to extract the derived-order privacy representation vector; loading the basic-order privacy representation vector and the derived-order privacy representation vector into the expression vector interaction subunit to interact to obtain the derived-order interaction vector; and loading the derived-order interaction vector into the derived-order privacy information recognition subunit to obtain the derived-order privacy level support coefficient to obtain the image privacy level probability variable corresponding to each derived-order privacy level.

[0009] Optionally, the method further includes: extracting intermediate-order privacy representation vectors from the target image information representation vector to obtain intermediate-order privacy representation vectors; performing representation vector interaction based on the basic-order privacy representation vector and the intermediate-order privacy representation vector to obtain intermediate-order interaction vectors; performing intermediate-order privacy level recognition based on the intermediate-order interaction vectors to obtain image privacy level probability variables corresponding to each intermediate-order privacy level; performing representation vector interaction between the basic-order privacy representation vector, the intermediate-order interaction vector, and the derived-order privacy representation vector to obtain the target derived-order interaction vector; and then performing derived-order privacy recognition based on the target derived-order interaction vector. Information recognition is performed to obtain the probability variables of the target image privacy level corresponding to each derived privacy level. The privacy level classification accuracy corresponding to the derived privacy level is higher than that corresponding to the intermediate privacy level, and the privacy level classification accuracy corresponding to the intermediate privacy level is higher than that corresponding to the basic privacy level. Based on the image privacy level probability variables corresponding to each basic privacy level, each intermediate privacy level, and each derived privacy level, the target privacy level of the target image to be processed is obtained. The target privacy level includes the target base-order privacy level, the target intermediate-order privacy level corresponding to the target base-order privacy level, and the target derived-order privacy level corresponding to the target intermediate-order privacy level. Specifically, an intermediate-order interaction vector is obtained by interacting the base-order privacy expression vector and the intermediate-order privacy expression vector, including: combining the base-order privacy expression vector and the intermediate-order privacy expression vector to obtain an intermediate-order combined expression vector; extracting the jump error expression vector based on the intermediate-order combined expression vector and the intermediate-order privacy expression vector to obtain an intermediate-order jump error expression vector; and then... The expression vectors are processed by feature element coordination to obtain the intermediate-order interaction vector; the basic-order private expression vector, the intermediate-order interaction vector, and the derived-order private expression vector are processed by expression vector interaction to obtain the target derived-order interaction vector, including: combining the basic-order private expression vector, the intermediate-order interaction vector, and the derived-order private expression vector to obtain the target combined expression vector; extracting the jump error expression vector based on the target combined expression vector and the derived-order private expression vector to obtain the target jump error expression vector; and processing the target jump error expression vector by feature element coordination to obtain the target derived-order interaction vector.

[0010] Optionally, the method further includes: loading the target image to be processed into the target image desensitization processing model Model2; extracting image information representation from the target image to be processed according to the target image desensitization processing model Model2 to obtain the target image information representation vector; extracting the basic-order privacy representation vector from the target image information representation vector according to the target image desensitization processing model Model2 to obtain the basic-order privacy representation vector, and performing basic-order privacy information recognition based on the basic-order privacy representation vector to obtain the image privacy level probability variable corresponding to each basic-order privacy level; extracting the intermediate-order privacy representation vector from the target image information representation vector according to the target image desensitization processing model Model2 to obtain the intermediate-order privacy representation vector, performing expression vector interaction based on the basic-order privacy representation vector and the intermediate-order privacy representation vector to obtain the intermediate-order interaction vector, and performing intermediate-order interaction vector recognition based on the intermediate-order interaction vector. The process involves identifying the privacy level of each intermediate-level privacy level to obtain the probability variables of the image privacy level corresponding to each intermediate-level privacy level. Based on the target image desensitization processing model Model2, the basic-level privacy expression vector, intermediate-level interaction vector, and derived-level privacy expression vector are interacted to obtain the target derived-level interaction vector. Then, based on the target derived-level interaction vector, derived-level privacy information is identified to obtain the probability variables of the target image privacy level corresponding to each derived-level privacy level. Finally, based on the probability variables of the image privacy level corresponding to each basic-level privacy level, each intermediate-level privacy level, and each derived-level privacy level, the target privacy level of the target image to be processed is obtained. The target privacy level includes the target basic-level privacy level, the target intermediate-level privacy level corresponding to the target basic-level privacy level, and the target derived-level privacy level corresponding to the target intermediate-level privacy level.

[0011] Optionally, the method further includes: obtaining a target image example and its corresponding image privacy level annotation; loading the target image example into the original user data anonymization processing model Model1', extracting image information representation from the target image example based on the original user data anonymization processing model Model1', and obtaining an example target image information representation vector; extracting a basic-order privacy representation vector from the example target image information representation vector based on the original user data anonymization processing model Model1', obtaining an example basic-order privacy representation vector, identifying basic-order privacy information based on the example basic-order privacy representation vector, and obtaining the example image privacy level probability variable corresponding to each basic-order privacy level; extracting a derived-order privacy representation vector from the example target image information representation vector based on the original user data anonymization processing model Model1', obtaining an example derived-order privacy representation vector, and interacting the representation vectors based on the example basic-order privacy representation vector and the example derived-order privacy representation vector to obtain the example derived-order privacy representation vector. Interaction vectors are used, and then derivative-level privacy information is identified based on the example derivative-level interaction vectors to obtain the example image privacy level probability variable corresponding to each derivative-level privacy level. The privacy level classification accuracy corresponding to the derivative-level privacy level is higher than that corresponding to the basic-level privacy level. The error between the example image privacy level probability variable corresponding to each basic-level privacy level and the basic-level privacy level annotation in the image privacy level annotation is obtained to obtain the basic-level privacy level error information. The error between the example image privacy level probability variable corresponding to each derivative-level privacy level and the derivative-level privacy level annotation in the image privacy level annotation is also obtained to obtain the derivative-level privacy level error information. The original user data anonymization processing model Model1' is adjusted according to the basic-level privacy level error information and the derivative-level privacy level error information, and the steps of obtaining the target image example and the corresponding image privacy level annotation are repeated until the training cutoff condition is met to obtain the target image desensitization processing model Model1.

[0012] Optionally, the method further includes: loading the target image example into the original user data anonymization processing model Model2'; extracting image information representation from the target image example based on the original user data anonymization processing model Model2' to obtain the example target image information representation vector; extracting the basic-order privacy representation vector from the example target image information representation vector based on the original user data anonymization processing model Model2' to obtain the example basic-order privacy representation vector; identifying basic-order privacy information based on the example basic-order privacy representation vector to obtain the example image privacy level probability variable corresponding to each basic-order privacy level; and further... Model 2' extracts intermediate-order privacy representation vectors from the example target image information representation vector to obtain the example intermediate-order privacy representation vector. It then performs representation vector interaction based on the example basic-order privacy representation vector and the example intermediate-order privacy representation vector to obtain the example intermediate-order interaction vector. Based on the example intermediate-order interaction vector, it performs intermediate-order privacy level identification to obtain the example image privacy level probability variable corresponding to each intermediate-order privacy level. Following the original user data anonymization processing model Model 2', it performs representation vector interaction between the example basic-order privacy representation vector, the example intermediate-order interaction vector, and the example derived-order privacy representation vector to obtain the example target derived-order interaction vector. Then, based on the example... Derivative-order privacy information is identified using the target derived-order interaction vector, obtaining the privacy level probability variable of the example target image corresponding to each derived-order privacy level. The privacy level classification accuracy corresponding to the derived-order privacy level is higher than that corresponding to the intermediate-order privacy level, and the privacy level classification accuracy corresponding to the intermediate-order privacy level is higher than that corresponding to the basic-order privacy level. The error between the example image privacy level probability variable corresponding to each basic-order privacy level and the basic-order privacy level annotation in the image privacy level annotation is obtained to obtain the basic-order privacy level error information. The example image privacy level probability variable corresponding to each intermediate-order privacy level and the image privacy level are also obtained. The error between intermediate privacy level annotations in the basic privacy level annotation is used to obtain intermediate privacy level error information. The error between the example target image privacy level probability variable corresponding to each derived privacy level and the derived privacy level annotation in the image privacy level annotation is obtained to obtain derived privacy level target error information. Based on the basic privacy level error information, intermediate privacy level error information and derived privacy level target error information, the original user data anonymization processing model Model2' is adjusted, and the steps of returning to obtain target image examples and corresponding image privacy level annotations are repeated until the training cutoff condition is met to obtain target image desensitization processing model Model2.

[0013] Secondly, embodiments of this application also provide a smart community data processing system based on artificial intelligence, including a server and a community camera device that communicate with each other. The server includes a memory and a processor. The memory stores a computer program. When the processor runs the computer program, it implements the method of the first aspect. Beneficial effects

[0014] The AI-based smart community data processing method and system provided in this application classifies the target image into privacy levels based on fundamental and derived levels, and matches corresponding desensitization strategies. Based on this, reasonable desensitization is adapted for images with different privacy levels, better coordinating the relationship between data services and user privacy protection. When identifying the privacy level of the target image, an AI model based on a user data anonymization processing model is used, ensuring efficient and accurate determination of the privacy level. During the optimization training of the user data anonymization processing model, the target image example is loaded into the original user data anonymization processing model Model1'. Image information representation is extracted from the target image example to obtain the example target image information representation vector. The example target image information representation vector is then used to extract the basic-order private representation vector to obtain the example basic-order private representation vector. Based on the example basic-order private representation vector, basic-order private information is identified to obtain the example image privacy level probability variable corresponding to each basic-order privacy level. The example target image information representation vector is then used to extract the derived-order private representation vector to obtain the example derived-order private representation vector. Based on the example basic-order private representation vector and the example derived-order private representation vector, representation vector interaction is performed to obtain the example derived-order interaction vector. Then, based on the example derived-order interaction vector, derived-order private information is identified to obtain the example image privacy level probability variable corresponding to each derived-order privacy level. During model tuning, the example derived-order interaction vector is obtained by interacting the example basic-order private representation vector and the example derived-order private representation vector. Then, the derived-order private information is identified. The private level classification accuracy of the derived-order private level is higher than that of the basic-order private level. Therefore, the hierarchical information of the private level can be obtained during model tuning. Then, the basic-order private level error information and the derived-order private level error information are obtained. The model is adjusted based on the basic-order private level error information and the derived-order private level error information until the training cutoff condition is met, and the target image desensitization processing model Model1 is obtained to enhance the recognition ability of the target image desensitization processing model.

[0015] Other features will be described in part in the following description. These features will be partially discovered by those skilled in the art upon examination of the following content and figures, or may be learned through production or application. The features of the present application can be implemented and obtained by practice or use of various aspects of the methods, tools, and combinations listed in the detailed examples described below. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0017] The methods, systems, and / or procedures shown in the accompanying drawings will be further described with reference to exemplary embodiments. These exemplary embodiments will be described in detail with reference to the drawings. These exemplary embodiments are non-limiting exemplary embodiments, wherein reference numerals in the various views of the drawings represent similar mechanisms.

[0018] Figure 1 This is a schematic diagram illustrating the composition of an AI-based smart community data processing system according to some embodiments of this application.

[0019] Figure 2 This is a schematic diagram illustrating the hardware and software composition of a server according to some embodiments of this application.

[0020] Figure 3 This is a flowchart illustrating an artificial intelligence-based smart community data processing method according to some embodiments of this application.

[0021] Figure 4 This is a schematic diagram of the architecture of the data processing device provided in the embodiments of this application. Detailed Implementation

[0022] To better understand the above technical solutions, the technical solutions of this application will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of this application and the specific features in the embodiments are detailed descriptions of the technical solutions of this application, rather than limitations on the technical solutions of this application. In the absence of conflict, the embodiments of this application and the technical features in the embodiments can be combined with each other.

[0023] In the detailed description below, numerous specific details are illustrated with examples to provide a comprehensive understanding of the relevant guidance. However, it will be apparent to those skilled in the art that this application can be practiced without these details. In other instances, well-known methods, procedures, systems, components, and / or circuits have been described at a relatively high level without detail to avoid unnecessarily obscuring aspects of this application.

[0024] These and other characteristics, the functions disclosed in the current application, the methods of execution, the functions of related elements in the structure, the combination of components, and the economics of production, may become more apparent in consideration of the following description with reference to the accompanying drawings, all of which form part of this application. However, it should be clearly understood that the drawings are for illustrative and descriptive purposes only and are not intended to limit the scope of this application. It should be understood that these drawings are not drawn to scale.

[0025] This application uses flowcharts to illustrate the execution process performed by a system according to embodiments of this application. It should be clearly understood that the execution processes in the flowcharts may not be executed sequentially. Instead, these execution processes may be executed in reverse order or simultaneously. Additionally, at least one other execution process may be added to the flowchart. One or more execution processes may be deleted from the flowchart.

[0026] Figure 1 This is a schematic diagram of the composition of an AI-based smart community data processing system 400 according to some embodiments of this application. The AI-based smart community data processing system 400 includes a server 100 and community camera equipment 300 that are interconnected via a network 200.

[0027] In some embodiments, please refer to Figure 2 This is a schematic diagram of the architecture of server 100, which includes a data processing device 110, a memory 120, a processor 130, and a communication unit 140. The memory 120, processor 130, and communication unit 140 are electrically connected directly or indirectly to each other to achieve data transmission or interaction. For example, these components can be electrically connected to each other through one or more communication buses or signal lines. The data processing device 110 includes at least one software function module that can be stored in the memory 120 in the form of software or firmware or embedded in the operating system (OS) of server 100. The processor 130 is used to execute executable modules stored in the memory 120, such as the software function modules and computer programs included in the data processing device 110.

[0028] The memory 120 may be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), etc. The memory 120 stores programs, which are executed by the processor 130 upon receiving execution instructions. The communication unit 140 establishes a communication connection between the server 100 and the front-end camera device 200 via a network and sends and receives data over the network.

[0029] The processor may be an integrated circuit chip with signal processing capabilities. The aforementioned processor can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application-Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor can be a microprocessor or any conventional processor.

[0030] Understandable. Figure 2 The structure shown is for illustrative purposes only; server 100 may also include components such as... Figure 2 The more or fewer components shown, or having the same Figure 2 The different configurations shown. Figure 2 The components shown can be implemented using hardware, software, or a combination thereof.

[0031] Figure 3 This is a flowchart illustrating an artificial intelligence-based smart community data processing method according to some embodiments of this application. The method is applied to... Figure 1 The server 100 in the diagram may specifically include the following steps 101-105. Based on the following steps, some optional embodiments will be described. These embodiments should be understood as examples and should not be construed as essential technical features for implementing this solution.

[0032] 101. Obtain the target image to be processed uploaded by the community camera equipment, extract the image information representation from the target image to obtain the target image information representation vector.

[0033] In this embodiment, the community camera equipment includes various types of cameras installed in different locations, such as spherical cameras along community roads, road cameras inside residential areas, and indoor surveillance cameras. The target image to be processed is a target image that requires desensitization processing, such as human images, vehicle images, and indoor images. Sensitive information involved includes, for example, faces, sensitive parts of the human body, license plates, and identification documents. The target image to be processed can be an image captured by the camera equipment, or a target video frame extracted from a video captured by the camera equipment. It can be captured in real time by the community camera equipment, or it can be selected from a target database stored by the camera equipment. The target image information representation vector is vector information describing the semantics of the target image to be processed, which can be extracted based on machine learning models such as CNN, DNN, RNN, and LSTM.

[0034] 102. Extract the basic-order privacy expression vector from the target image information expression vector to obtain the basic-order privacy expression vector, and perform basic-order privacy information recognition based on the basic-order privacy expression vector to obtain the image privacy level probability variable corresponding to each basic-order privacy level.

[0035] The base level is the level with the smallest granularity (or degree of subdivision) within the privacy level, i.e., the highest level, and therefore covers the widest range of images. In practical applications, the privacy information involved in the target image is divided into different privacy levels. For example, for personal privacy information, the base level may include exposed subject I, derived levels of exposed subject I include face II1 and private parts II2, derived levels of face II1 include facial features III1 and contours III2, derived levels of facial features III1 include eyes IV1, and so on. Depending on the target image, different privacy level divisions can be performed, and the granularity and standards of the division are not limited. In this embodiment, by dividing each privacy level into different levels, a tree-like privacy level hierarchy is obtained, where the base level is the level of the main level in the tree-like privacy level hierarchy. The closer the level in the tree-like privacy level hierarchy is to the base level, the lower the accuracy of the corresponding image privacy level division. Different tree-like privacy level hierarchies can be formed by different base level privacy levels, and each base level privacy level corresponds to a tree-like privacy level hierarchy. The basic-order privacy representation vector can be extracted based on the convolutional neural network corresponding to the basic-order level. This vector is used to analyze the privacy level corresponding to the basic-order level, which indicates the privacy level corresponding to the basic-order level in the tree-structured privacy level hierarchy. The image privacy level probability variable represents the probability that the corresponding privacy level corresponds to the privacy level of the target image to be processed. The larger the image privacy level probability variable, the higher the confidence level that the corresponding privacy level corresponds to the privacy level of the target image to be processed. The image privacy level probability variable corresponding to the basic-order privacy level represents the probability that the corresponding basic-order privacy level corresponds to the privacy level of the target image to be processed. The larger the probability, the higher the confidence level that the privacy level of the target image to be processed corresponds to the basic-order privacy level. Optionally, high-level semantic information of the target image information representation vector can be extracted using the convolutional neural network corresponding to the basic-order level to obtain the basic-order privacy representation vector. Then, the basic-order privacy information recognition parameters are used to identify the basic-order privacy information in the basic-order privacy representation vector to obtain the image privacy level probability variable corresponding to each basic-order privacy level.

[0036] 103. The target image information expression vector is extracted by the derivative-order private expression vector to obtain the derivative-order private expression vector. The expression vectors are interacted based on the basic-order private expression vector and the derivative-order private expression vector to obtain the derivative-order interaction vector. Then, the derivative-order private information is identified based on the derivative-order interaction vector to obtain the image privacy level probability variable corresponding to each derivative-order privacy level. The privacy level classification accuracy corresponding to the derivative-order privacy level is higher than that corresponding to the basic-order privacy level.

[0037] The derived order refers to the order within the privacy hierarchy with finer granularity, thus involving the smallest target image region. For example, the derived order is the order corresponding to a branch privacy level in a tree-like privacy hierarchy. The derived order privacy representation vector indicates the image features obtained through the convolutional neural network corresponding to the derived order. In a tree-like privacy hierarchy, different orders are adapted to different convolutional neural networks for extracting high-level semantic information. Based on the derived order privacy representation vector, the privacy level corresponding to the derived order can be analyzed and identified. The derived order interaction vector is obtained by interacting with the basic order privacy representation vector and the derived order privacy representation vector. This ensures that information from the basic order is considered when determining the privacy level of the derived order, resulting in a more accurate determination. The derived order privacy level is the privacy level corresponding to the derived order in a tree-like privacy hierarchy. Higher privacy level classification precision results in a more refined privacy level. Privacy levels of the same order within a tree-like privacy hierarchy have the same classification precision. The privacy level classification accuracy of the derivative-order privacy level is higher than that of the basic-order privacy level in the basic-order. The probability variable of the privacy level of the image corresponding to the privacy level of the derivative-order represents the probability that the privacy level of the target image to be processed is the same as the privacy level of the derivative-order privacy level. The higher the probability, the higher the confidence level that the privacy level of the target image to be processed is the privacy level of the derivative-order privacy level.

[0038] Optionally, a derivative-order private representation vector can be extracted from the target image information representation vector using a derivative-order convolutional neural network to obtain a derivative-order private representation vector. Then, the basic-order private representation vector and the derivative-order private representation vector are combined to obtain a combined representation vector. This combined representation vector is then mined to obtain a derivative-order interaction vector. Finally, derivative-order private information is identified from the derivative-order interaction vector using derivative-order private information recognition parameters to obtain the image privacy level probability variable corresponding to each derivative-order privacy level.

[0039] 104. Based on the image privacy level probability variables corresponding to each basic privacy level and each derived privacy level, the privacy level of the target image to be processed is obtained. The privacy level includes the target basic privacy level and the target derived privacy level corresponding to the target basic privacy level.

[0040] The target basic privacy level indicates the privacy level of the basic level in the tree-like privacy level hierarchy corresponding to the target image to be processed. The target derived privacy level indicates the privacy level of the derived level in the tree-like privacy level hierarchy corresponding to the target image to be processed. The target basic privacy level and the target derived privacy level are the privacy levels of the same tree-like privacy level hierarchy. In other words, the privacy level of the trunk corresponding to the hierarchy where the target derived privacy level is located is the target basic privacy level. Optionally, a branch chain from the basic privacy level of the trunk to the corresponding derived privacy level of the branch can be determined by a preset privacy level hierarchy (tree-like privacy level hierarchy). The image privacy level probability variable corresponding to the basic privacy level and the image privacy level probability variable corresponding to the derived privacy level are obtained on each branch chain. The image privacy level probability variable of each branch chain is obtained. Then, the branch chain with the largest image privacy level probability variable is determined. The basic privacy level and derived privacy level included in the branch chain with the largest image privacy level probability variable are taken as the target basic privacy level and the target derived privacy level corresponding to the target basic privacy level. In this way, the privacy level corresponding to the target image to be processed is obtained.

[0041] In an optional implementation, the basic-level privacy level corresponding to the maximum image privacy level probability variable can be determined from the image privacy level probability variables corresponding to each basic-level privacy level to obtain the target basic-level privacy level. Then, the derived-level privacy level corresponding to the target basic-level privacy level can be obtained according to the preset privacy level. The derived-level privacy level corresponding to the maximum image privacy level probability variable can be determined from the image privacy level probability variables of each derived-level privacy level corresponding to the target basic-level privacy level to obtain the target derived-level privacy level.

[0042] 105. Based on the privacy level of the target image to be processed, retrieve the corresponding desensitization strategy and perform desensitization processing on the target image to be processed.

[0043] Desensitization strategies include existing desensitization operations such as metadata removal, mosaic masking, and Fawkes-based adversarial noise addition. It can be understood that different privacy levels can correspond to different desensitization strategies to meet different desensitization needs.

[0044] This application embodiment involves extracting a basic-level private representation vector from the target image information representation vector to obtain a basic-level private representation vector. Then, basic-level private information recognition is performed to obtain the image privacy level probability variable corresponding to each basic-level privacy level. Next, a derived-level private representation vector is extracted from the target image information representation vector to obtain a derived-level private representation vector. Finally, a derived-level interaction vector is obtained by interacting the basic-level and derived-level private representation vectors. Then, derived-level private information recognition is performed based on the derived-level interaction vector to obtain the image privacy level probability variable corresponding to each derived-level privacy level. This process utilizes high-level semantic information from the basic level. Information is used to determine the privacy level of the derived order. Since the privacy level classification accuracy of the derived order is higher than that of the basic order, the order information of the privacy level can be taken into account, making the obtained image privacy level probability variables corresponding to each derived order privacy level more accurate. Based on the image privacy level probability variables corresponding to each basic order privacy level and each derived order privacy level, the privacy level corresponding to the target image to be processed is obtained. The privacy level covers the target basic order privacy level and the target derived order privacy level corresponding to the target basic order privacy level, thus improving the accuracy of the obtained privacy level.

[0045] In an optional implementation, step 103 involves performing expression vector interaction based on the basic-order private expression vector and the derived-order private expression vector to obtain the derived-order interaction vector. Specifically, this includes: combining the basic-order private expression vector and the derived-order private expression vector to obtain a combined expression vector; extracting the jump error expression vector based on the combined expression vector and the derived-order private expression vector to obtain a jump error expression vector; and then integrating the jump error expression vector with feature elements to obtain the derived-order interaction vector.

[0046] The combined expression vector is obtained by combining the base-order private expression vector and the derived-order private expression vector, for example, by concatenating the first and second parts. The jump error expression vector indicates the image features obtained through the residual model (ResNet, using residual block jump concatenation), which can reduce the occurrence of gradient vanishing. Optionally, the combined expression vector can be obtained by combining the base-order private expression vector and the derived-order private expression vector, and then the jump error expression vector can be extracted from the combined expression vector and the derived-order private expression vector according to the residual model. Then, the jump error expression vector is subjected to feature element coordination algorithm to obtain the derived-order interaction vector. The operation process of the feature element coordination algorithm is to normalize the jump error expression vector to a numerical range, complete the data standardization, and achieve the purpose of gradient optimization.

[0047] In an optional implementation, the jump error expression vector is extracted based on the combined expression vector and the derived private expression vector to obtain the jump error expression vector. Specifically, this includes: performing correlation projection calculation on the combined expression vector based on preset jump parameters to obtain the projected expression vector, and obtaining the vector sum of the projected expression vector and the derived private expression vector to obtain the jump error expression vector.

[0048] The jump parameters include weight variables and bias variables, used for residual operations. The projected expression vector indicates the vector calculated by associating the combined expression vector with a linear mapping. Optionally, the combined expression vector can be weighted based on the weight variables in the preset jump parameters and then added to the bias variables in the preset jump parameters to obtain the projected expression vector. The projected expression vector is then summed with the derived-order private expression vector to obtain the jump error expression vector. This application combines the base-order private expression vector and the derived-order private expression vector to obtain the combined expression vector. The jump error expression vector is then extracted based on the combined expression vector and the derived-order private expression vector. Finally, the jump error expression vector is processed by feature element aggregation to obtain the derived-order interaction vector. This improves the reliability of the obtained derived-order interaction vector and reduces gradient vanishing.

[0049] In an optional implementation, step 104 obtains the privacy level corresponding to the target image to be processed based on the image privacy level probability variables corresponding to each basic privacy level and each derived privacy level, specifically including: 1041, retrieve the location of the image's privacy level.

[0050] 1042. Based on the image privacy level location, the probability variables of image privacy level corresponding to each basic privacy level and the probability variables of image privacy level corresponding to each derived privacy level are integrated to obtain the route probability variables at each location.

[0051] The image privacy level location indicates the position of each preset privacy level's hierarchy. Different privacy levels can correspond to the same privacy level hierarchy, or they can correspond to different privacy level hierarchys. The image privacy level location can also be obtained through a tree-like privacy level hierarchy. The location route probability variable represents the confidence level of the privacy level included in the hierarchy route corresponding to the target image to be processed. The hierarchy route can indicate the branch chain from the basic privacy level to the derived privacy level in the tree-like privacy level hierarchy, that is, the route from the privacy level of the main branch to the privacy level of the branch is a branch chain. Optionally, the image privacy level location can be obtained from a pre-set storage space, and then the image privacy level probability variables corresponding to the basic privacy level and the image privacy level probability variables corresponding to the same hierarchy route of the basic privacy level and the derived privacy level can be integrated based on the image privacy level location to obtain the location route probability variables.

[0052] 1043. Determine the target location route probability variable from the route probability variables of each location, and determine the privacy level corresponding to the target image to be processed based on the target location route probability variable.

[0053] Optionally, the magnitudes of the probability variables of each location route can be compared, and the largest target among the location route probability variables can be determined as the target location route probability variable. The basic order privacy level and the derived order privacy level in the route corresponding to the target location route probability variable can be used as the privacy level corresponding to the target image to be processed.

[0054] This application integrates the probability variables of image privacy levels corresponding to each basic privacy level and each derived privacy level based on the obtained image privacy level location to obtain the probability variables of each location route. Then, it determines the probability variable of the target location route from the probability variables of each location route, and determines the privacy level corresponding to the target image to be processed based on the probability variables of the target location route. In this way, the order structure of the privacy level corresponding to the target image to be processed is more accurate, and the situation where the basic privacy level and the derived privacy level do not correspond to the same order route is eliminated.

[0055] In an optional implementation, step 1042 integrates the image privacy level probability variables corresponding to each basic privacy level and each derived privacy level based on the image privacy level location to obtain the route probability variables for each location, specifically including: 10421. Determine the image privacy level probability variable corresponding to the current basic privacy level from the image privacy level probability variables corresponding to each basic privacy level.

[0056] 10422, based on the image privacy level position, obtain the current derived level privacy level corresponding to the current basic level privacy level, and determine the image privacy level probability variable corresponding to the current derived level privacy level from the image privacy level probability variables corresponding to each derived level privacy level.

[0057] The current basic privacy level indicates the basic privacy level to be integrated and obtained at this time. The current derived privacy level indicates the derived privacy level that needs to be integrated with the basic privacy level among all derived privacy levels of the current basic privacy level. Optionally, the current basic privacy level can be determined sequentially from each basic privacy level, and the image privacy level probability variable corresponding to the current basic privacy level can be obtained from the image privacy level probability variables corresponding to each basic privacy level. Then, based on the image privacy level position, all derived privacy levels covered by the current basic privacy level can be obtained. After that, the derived privacy level can be used as the current derived privacy level sequentially, and the image privacy level probability variable corresponding to the current derived privacy level can be obtained from the image privacy level probability variables corresponding to each derived privacy level.

[0058] 10423, obtain the multiplication result of the image privacy level probability variable corresponding to the current basic privacy level and the image privacy level probability variable corresponding to the current derived privacy level, and get the current location route probability variable.

[0059] 10424. By reviewing the image privacy level probability variables corresponding to each basic privacy level and each derived privacy level, the probability variables of each location route can be obtained.

[0060] The current location route probability variable indicates the confidence level of the current base-order privacy level and the current derived-order privacy level as the corresponding privacy level of the target image to be processed. Optionally, the current location route probability variable can be obtained by multiplying the image privacy level probability variable corresponding to the current base-order privacy level with the image privacy level probability variable corresponding to the current derived-order privacy level. The location route probability variable for each route is obtained by multiplying the image privacy level probability variable corresponding to the base-order privacy level with the image privacy level probability variable corresponding to the derived-order privacy level.

[0061] For example, if an image has a basic privacy level of 1 and 2, and each basic privacy level has two derived privacy levels (11, 12, 21, and 22), then there are four possible routes. By determining 1 and 11 as the current basic and derived privacy levels, and considering all basic and derived privacy levels, and obtaining the probability variables for routes corresponding to 2 and 22, four probability variables for each route are obtained. The route with the highest probability variable is then identified, and the privacy level in its corresponding route is determined as the privacy level of the target image to be processed. This application obtains the current derived privacy level corresponding to the current basic privacy level based on the image privacy level position, then multiplies the image privacy level probability variable corresponding to the current basic privacy level and the image privacy level probability variable corresponding to the current derived privacy level to obtain the current location route probability variable. Then, it reviews the image privacy level probability variables corresponding to each basic privacy level and each derived privacy level to obtain the route probability variables for each location, thereby improving the accuracy of the route probability variables for each location.

[0062] In an optional implementation, the AI-based smart community data processing method provided in this application may further include the following steps: 110. Load the target image to be processed into the target image desensitization processing model Model1.

[0063] 120. Based on the target image desensitization processing model Model1, the image information representation of the target image to be processed is extracted to obtain the target image information representation vector.

[0064] The target image desensitization processing model is an optimized machine learning model for determining the multi-level privacy level of an image. This determination includes at least two levels of privacy level identification. The target image desensitization processing model (Model1) includes basic-level privacy information recognition and derived-level privacy information recognition. Optionally, the model can be optimized using optimized examples (sample examples for model debugging). When determining the privacy level of a target image, the target image to be processed is used as input to the target image desensitization processing model (Model1). The model extracts image information representation from the target image to obtain a target image information representation vector, for example, through feature extraction using a convolutional neural network.

[0065] 130. Based on the target image desensitization processing model Model1, the target image information expression vector is extracted into a basic-order private expression vector to obtain a basic-order private expression vector. Based on the basic-order private expression vector, basic-order private information is identified to obtain the image privacy level probability variable corresponding to each basic-order privacy level.

[0066] 140. Based on the target image desensitization processing model Model1, the target image information expression vector is extracted by the derivative-level private expression vector to obtain the derivative-level private expression vector. The expression vectors are interacted based on the basic-level private expression vector and the derivative-level private expression vector to obtain the derivative-level interaction vector. Then, the derivative-level private information is identified based on the derivative-level interaction vector to obtain the image privacy level probability variable corresponding to each derivative-level privacy level.

[0067] Optionally, the target image information representation vector can be extracted based on the target image desensitization processing model Model1, and then basic-level privacy information recognition and derived-level privacy information recognition can be performed simultaneously. This involves extracting basic-level privacy representation vectors from the target image information representation vectors, and then performing basic-level privacy information recognition based on these basic-level privacy representation vectors to obtain the image privacy level probability variables corresponding to each basic-level privacy level. Next, derived-level privacy representation vectors are extracted from the target image information representation vectors. The basic-level and derived-level privacy representation vectors are then interacted to obtain derived-level interaction vectors. Finally, derived-level privacy information recognition is performed based on these derived-level interaction vectors to obtain the image privacy level probability variables corresponding to each derived-level privacy level.

[0068] This application uses the target image desensitization processing model Model1 to determine the privacy level of the target image at multiple levels, obtains the image privacy level probability variables corresponding to each basic privacy level and the image privacy level probability variables corresponding to each derived privacy level, thereby improving the efficiency of privacy level determination.

[0069] In an optional implementation, the target image desensitization processing model Model1 includes an image information representation extraction unit, a basic-level private information recognition unit, and a derived-level private information recognition unit. Loading the target image to be processed into the target image desensitization processing model Model1 in step 110 may specifically include: 111. The target image to be processed is loaded into the image information expression extraction unit for image information expression extraction to obtain the target image information expression vector.

[0070] 112. Load the target image information representation vector into the basic-order privacy information recognition unit. Extract the basic-order privacy representation vector from the target image information representation vector based on the basic-order privacy information recognition unit to obtain the basic-order privacy representation vector. Obtain the basic-order privacy level support coefficient based on the basic-order privacy representation vector to obtain the image privacy level probability variable corresponding to each basic-order privacy level.

[0071] 113. The target image information representation vector and the basic-order privacy representation vector are loaded into the derived-order privacy information recognition unit. The derived-order privacy representation vector is extracted from the target image information representation vector by the derived-order privacy information recognition unit to obtain the derived-order privacy representation vector. The representation vectors are interacted based on the basic-order privacy representation vector and the derived-order privacy representation vector to obtain the derived-order interaction vector. Then, the derived-order privacy level support coefficient is obtained based on the derived-order interaction vector to obtain the image privacy level probability variable corresponding to each derived-order privacy level.

[0072] The image information extraction unit can be a convolutional neural network, such as VGG, used for image information extraction. It is the main unit for determining the privacy level. Of course, the image information extraction unit can also be other forms of deep neural networks. The basic-level privacy information recognition unit is a subordinate unit that determines the privacy level for each preset basic-level privacy level. It is also built based on deep networks such as convolutional neural networks. The derived-level privacy information recognition unit is a subordinate unit used to determine the privacy level for each preset derived-level privacy level. It is also built based on deep networks such as convolutional neural networks. Optionally, image information extraction can be performed first by the image information extraction unit, and then the target image information expression vector can be loaded into the two subordinate units (the basic-level privacy information recognition unit and the derived-level privacy information recognition unit). When determining the privacy level, the derived-level privacy information recognition unit obtains the basic-level privacy expression vector extracted by the basic-level privacy information recognition unit, performs expression vector interaction between the basic-level privacy expression vector and the derived-level privacy expression vector to obtain the derived-level interaction vector, and determines the privacy level based on the derived-level interaction vector. This application extracts feature information based on the image information expression extraction unit, and then determines the privacy level based on the basic-level privacy information recognition unit and the derived-level privacy information recognition unit, thereby enhancing the speed and accuracy of privacy level determination.

[0073] In an optional implementation, the derivative-level private information recognition unit includes a derivative-level private expression vector extraction subunit, an expression vector interaction subunit, and a derivative-level private information recognition subunit. For 103, the target image information representation vector and the basic-order private representation vector are loaded into the derived-order private information recognition unit. Specifically, this may include: loading the target image information representation vector into the derived-order private representation vector extraction subunit to extract the derived-order private representation vector, thereby obtaining the derived-order private representation vector; loading the basic-order private representation vector and the derived-order private representation vector into the representation vector interaction subunit to interact with the representation vector, thereby obtaining the derived-order interaction vector; and loading the derived-order interaction vector into the derived-order private information recognition subunit to obtain the derived-order privacy level support coefficient, thereby obtaining the image privacy level probability variable corresponding to each derived-order privacy level.

[0074] The derived-order private representation vector extraction subunit is used to extract derived-order private representation vectors. This extraction subunit may be based on a convolutional neural network (CNN), which can be a non-sparse neural network. The representation vector interaction subunit integrates and interacts features from different orders. The derived-order private information recognition subunit determines the privacy level of each derived-order privacy level. Optionally, the derived-order private representation vector can be extracted by the derived-order private representation vector extraction subunit, and then integrated with the basic-order private representation vector by the representation vector interaction subunit to obtain the derived-order interaction vector. This interaction vector is then used by the derived-order private information recognition subunit to determine the privacy level of the derived-order private level, thus improving the accuracy of the image privacy level probability variables corresponding to each derived-order privacy level.

[0075] In an optional implementation, the AI-based smart community data processing method provided in this application may further include the following steps: 100, the intermediate-order privacy representation vector is extracted from the target image information representation vector to obtain the intermediate-order privacy representation vector.

[0076] 200. Based on the basic-order privacy representation vector and the intermediate-order privacy representation vector, the representation vectors interact to obtain the intermediate-order interaction vector. Based on the intermediate-order interaction vector, the intermediate-order privacy level is identified to obtain the image privacy level probability variable corresponding to each intermediate-order privacy level.

[0077] In the above steps, the intermediate level refers to the preset privacy level, which lies between the basic and derived levels; the number of intermediate levels is not limited. The intermediate-level privacy representation vector indicates the image features obtained through the convolutional neural network corresponding to the intermediate level. The intermediate-level privacy level is identified using the intermediate-level privacy representation vector. The intermediate-level interaction vector indicates the feature vector obtained by integrating the basic and intermediate-level privacy representation vectors, so that information from the previous level can be considered when determining the privacy level of the intermediate level. The intermediate-level privacy level indicates the privacy level corresponding to the level between the basic and derived levels; the number of intermediate-level privacy levels is not limited. Optionally, high-level semantic information of the target image information representation vector can be extracted through intermediate-level depth extraction parameters to obtain the intermediate-level privacy representation vector. The basic and intermediate-level privacy representation vectors are then combined to obtain the combined representation vector. The combined representation vector is then mined to obtain the intermediate-level interaction vector. Finally, the intermediate-level privacy level is identified based on the intermediate-level interaction vector to obtain the image privacy level probability variable corresponding to each intermediate-level privacy level.

[0078] 300, the basic-order private expression vector, intermediate-order interaction vector and derived-order private expression vector are interacted to obtain the target derived-order interaction vector. Then, based on the target derived-order interaction vector, derived-order private information is identified to obtain the probability variable of the target image private level corresponding to each derived-order private level. The private level classification accuracy corresponding to the derived-order private level is higher than that corresponding to the intermediate-order private level, and the private level classification accuracy corresponding to the intermediate-order private level is higher than that corresponding to the basic-order private level.

[0079] When the target derived-order interaction vector indicates the existence of an intermediate-order privacy level, it is the feature vector obtained by integrating the basic-order privacy expression vector, the intermediate-order interaction vector, and the derived-order privacy expression vector. The target derived-order interaction vector includes information on the privacy level determination for the basic, intermediate, and derived orders. The privacy level classification accuracy corresponding to the intermediate-order privacy level is lower than that corresponding to the derived-order privacy level, but higher than that corresponding to the basic-order privacy level. The privacy level classification accuracy corresponding to the intermediate-order privacy level is between that of the basic-order privacy level and the derived-order privacy level. When the privacy level of the target image is determined at the intermediate-order, the privacy level of the target image at the derived-order is determined by interacting the expression vectors based on the basic-order privacy expression vector, the intermediate-order interaction vector, and the derived-order privacy expression vector to obtain the target derived-order interaction vector. The basic-order privacy expression vector, the intermediate-order interaction vector, and the derived-order privacy expression vector can be combined, and the jump error expression vector is extracted from the combined expression vector. Then, the feature elements of the jump error expression vector are integrated to obtain the target derived-order interaction vector. Then, the privacy level of the derived order is determined based on the target derived order interaction vector. Privacy level determination can be performed using multiple privacy level determination parameters, obtaining the target image privacy level probability variable corresponding to each derived order privacy level. Determining the privacy level of the derived order based on the target derived order interaction vector, by fusing the feature information from the basic and intermediate orders during privacy level determination, improves the accuracy of privacy information recognition for the derived order.

[0080] 400. Based on the probability variables of image privacy levels corresponding to each basic privacy level, the probability variables of image privacy levels corresponding to each intermediate privacy level, and the probability variables of target image privacy levels corresponding to each derived privacy level, the target privacy level corresponding to the target image to be processed is obtained. The target privacy level includes the target basic privacy level, the target intermediate privacy level corresponding to the target basic privacy level, and the target derived privacy level corresponding to the target intermediate privacy level.

[0081] The target intermediate privacy level indicates the intermediate privacy level of the tree-like privacy level hierarchy corresponding to the target image to be processed. Optionally, a branch chain can be determined based on a preset privacy level hierarchy, i.e., a tree-like privacy level hierarchy, from the basic privacy level corresponding to the main branch to the intermediate privacy level corresponding to the middle branch, and then to the derived privacy level corresponding to the branch. The image privacy level probability variables corresponding to the basic privacy level, the intermediate privacy level, and the derived privacy level on each branch chain are multiplied to obtain the image privacy level probability variables of each branch chain. Then, the branch chain with the largest image privacy level probability variable is determined. The basic privacy level, intermediate privacy level, and derived privacy level included in the branch chain with the largest image privacy level probability variable are taken as the target basic privacy level, the target intermediate privacy level corresponding to the target basic privacy level, and the target derived privacy level corresponding to the target intermediate privacy level, thus obtaining the target privacy level corresponding to the target image to be processed. In an optional implementation, the basic-level privacy level corresponding to the maximum image privacy level probability variable can be determined from the image privacy level probability variables corresponding to each basic-level privacy level to obtain the target basic-level privacy level. Then, each intermediate-level privacy level corresponding to the target basic-level privacy level is obtained according to a preset privacy level. Next, the intermediate-level privacy level corresponding to the maximum image privacy level probability variable is determined from the image privacy level probability variables corresponding to each intermediate-level privacy level corresponding to the target basic-level privacy level to obtain the target intermediate-level privacy level. Then, each derived-level privacy level corresponding to the target intermediate-level privacy level is obtained based on a preset privacy level. Finally, the derived-level privacy level corresponding to the maximum image privacy level probability variable is determined from the image privacy level probability variables corresponding to each derived-level privacy level corresponding to the target basic-level privacy level to obtain the target derived-level privacy level. This application determines the intermediate privacy level based on the target image information representation vector, and makes the obtained target privacy level include the target basic privacy level, the target intermediate privacy level corresponding to the target basic privacy level, and the target derived privacy level corresponding to the target intermediate privacy level, thereby improving the accuracy of the obtained target privacy level.

[0082] In an optional implementation, step 400 involves performing expression vector interaction based on the basic-order private expression vector and the intermediate-order private expression vector to obtain an intermediate-order interaction vector. Specifically, this includes: combining the basic-order private expression vector and the intermediate-order private expression vector to obtain an intermediate-order combined expression vector; extracting a jump error expression vector based on the intermediate-order combined expression vector and the intermediate-order private expression vector to obtain an intermediate-order jump error expression vector; and then integrating the intermediate-order jump error expression vector with feature elements to obtain an intermediate-order interaction vector.

[0083] The intermediate-order combined expression vector indicates the expression vector obtained by combining the basic-order private expression vector and the intermediate-order private expression vector, which is the combined expression vector obtained when the intermediate-order expression vectors interact. The intermediate-order jump error expression vector indicates the jump error expression vector mined when the intermediate-order expression vectors interact. The intermediate-order interaction vector indicates the integrated vector mined when the intermediate-order expression vectors interact. Optionally, the intermediate-order combined expression vector can be obtained by combining the basic-order private expression vector and the intermediate-order private expression vector; then, the jump error expression vector can be extracted from the intermediate-order combined expression vector and the intermediate-order private expression vector according to the residual model to obtain the intermediate-order jump error expression vector; then, the intermediate-order jump error expression vector can be subjected to feature element coordination based on the feature element coordination algorithm to obtain the intermediate-order interaction vector; when extracting the jump error expression vector, for example, the intermediate-order combined expression vector can be subjected to nonlinear projection transformation to obtain the intermediate-order projected expression vector; then, the vector sum of the intermediate-order projected expression vector and the intermediate-order private expression vector can be obtained to obtain the intermediate-order jump error expression vector. This application proposes to improve the accuracy of intermediate-level privacy level determination by integrating the basic-level privacy representation vector with the intermediate-level high-level semantic information before determining the intermediate-level privacy level.

[0084] In an optional implementation, the process of performing expression vector interaction between the basic-order private expression vector, the intermediate-order interaction vector, and the derived-order private expression vector in step 300 to obtain the target derived-order interaction vector may specifically include: combining the basic-order private expression vector, the intermediate-order interaction vector, and the derived-order private expression vector to obtain the target combined expression vector; extracting the jump error expression vector based on the target combined expression vector and the derived-order private expression vector to obtain the target jump error expression vector; and integrating the target jump error expression vector with feature elements to obtain the target derived-order interaction vector.

[0085] The target combined expression vector indicates the expression vector obtained by combining the base-order private expression vector, the intermediate-order interaction vector, and the derived-order private expression vector when it has an intermediate order. In other words, the target combined expression vector is the combined expression vector obtained by the derived order when the expression vectors interact. The target jump error expression vector indicates the jump error expression vector mined by the derived order when the expression vectors interact when it has an intermediate order. The target combined expression vector indicates the fusion vector mined by the derived order when the expression vectors interact when it has an intermediate order. Optionally, the target combined expression vector can be obtained by concatenating the base-order private expression vector, the intermediate-order interaction vector, and the derived-order private expression vector in sequence. Then, the jump error expression vector can be extracted from the target combined expression vector and the derived-order private expression vector according to the residual model to obtain the target jump error expression vector. Finally, the target jump error expression vector can be subjected to feature element coordination based on the feature element coordination algorithm to obtain the target derived-order interaction vector. This application integrates the high-level semantic information of all levels to obtain the target derived level interaction vector during the process of identifying derived level private information, thereby improving the accuracy of the obtained target derived level interaction vector. Then, the privacy level of the derived level is determined based on the target derived level interaction vector, making the determination of the privacy level of the derived level more accurate.

[0086] In an optional implementation, the AI-based smart community data processing method provided in this application may further include the following steps: (1) Load the target image to be processed into the target image desensitization processing model Model2, and extract the image information expression of the target image to be processed according to the target image desensitization processing model Model2 to obtain the target image information expression vector.

[0087] (2) Based on the target image desensitization processing model Model2, the target image information expression vector is extracted by the basic level private expression vector to obtain the basic level private expression vector, and the basic level private information is identified based on the basic level private expression vector to obtain the image privacy level probability variable corresponding to each basic level privacy level.

[0088] The target image desensitization processing model Model2 indicates a target image desensitization processing model that includes basic-level privacy information recognition, intermediate-level privacy level recognition, and derived-level privacy information recognition. In an optional implementation, the target image desensitization processing model may include basic-level privacy information recognition, multiple intermediate-level privacy level recognitions, and derived-level privacy information recognitions. The privacy level determination parameters for basic-level privacy information recognition in the target image desensitization processing model Model2 may be consistent with or inconsistent with those in the target image desensitization processing model Model1, depending on the preset privacy level of the basic level. Optionally, the privacy level of the basic level can be determined based on the target image desensitization processing model Model2; that is, a basic-level privacy expression vector is extracted based on a high-level semantic information extraction network, and then basic-level privacy information recognition is performed based on the basic-level privacy expression vector by a basic-level privacy level determination network to obtain the image privacy level probability variables corresponding to each basic-level privacy level.

[0089] (3) Based on the target image desensitization processing model Model2, the target image information expression vector is extracted into an intermediate-level privacy expression vector to obtain an intermediate-level privacy expression vector. The expression vectors are interacted based on the basic-level privacy expression vector and the intermediate-level privacy expression vector to obtain an intermediate-level interaction vector. The intermediate-level privacy level is identified based on the intermediate-level interaction vector to obtain the image privacy level probability variable corresponding to each intermediate-level privacy level.

[0090] Optionally, the intermediate privacy level can be determined based on the target image desensitization processing model Model2. That is, the intermediate privacy expression vector is extracted based on the intermediate high-level semantic information extraction network, and then the basic-level privacy expression vector and the intermediate-level privacy expression vector are interacted based on the intermediate-level expression vector interaction network to obtain the intermediate-level interaction vector. Then, the intermediate privacy level determination network is used to identify the intermediate privacy level based on the intermediate-level interaction vector to obtain the image privacy level probability variable corresponding to each intermediate privacy level.

[0091] (4) Based on the target image desensitization processing model Model2, the basic-order private expression vector, the intermediate-order interaction vector and the derived-order private expression vector are interacted to obtain the target derived-order interaction vector. Then, based on the target derived-order interaction vector, the derived-order private information is identified to obtain the target image private level probability variable corresponding to each derived-order private level.

[0092] Optionally, the privacy level of the derived order can be determined based on the target image desensitization processing model Model2. That is, the derived order privacy expression vector is extracted based on the high-level semantic information extraction network of the derived order, and then the expression vector interaction network of the derived order interacts the basic order privacy expression vector, the intermediate order depth vector and the value derived order privacy expression vector to obtain the target derived order interaction vector. Then, the derived order privacy level determination neural network identifies the derived order privacy information based on the target derived order interaction vector to obtain the image privacy level probability variable corresponding to each derived order privacy level.

[0093] (5) Based on the probability variables of image privacy level corresponding to each basic privacy level, the probability variables of image privacy level corresponding to each intermediate privacy level, and the probability variables of target image privacy level corresponding to each derived privacy level, the target privacy level corresponding to the target image to be processed is obtained. The target privacy level includes the target basic privacy level, the target intermediate privacy level corresponding to the target basic privacy level, and the target derived privacy level corresponding to the target intermediate privacy level.

[0094] Optionally, based on the target image desensitization processing model Model2, the probability variables of image privacy levels corresponding to each basic privacy level, each intermediate privacy level, and each derived privacy level can be obtained to acquire the probability variables of image privacy levels for each route. Then, the basic, intermediate, and derived privacy levels in the route corresponding to the maximum image privacy level probability variable are determined. This yields the privacy level of the target image output by the target image desensitization processing model Model2, namely the target basic privacy level, the target intermediate privacy level corresponding to the target basic privacy level, and the target derived privacy level corresponding to the target intermediate privacy level. Using the target image desensitization processing model Model2 further improves the accuracy of privacy level determination.

[0095] In an optional implementation, this application also provides an optimized training process for the user data anonymization model, including: (A) Obtain the target image example and the corresponding image privacy level annotation, load the target image example into the original user data anonymization processing model Model1', extract the image information representation of the target image example based on the original user data anonymization processing model Model1', and obtain the example target image information representation vector.

[0096] The target image example is the target image used in the model optimization process. The image privacy level annotation includes indication information (such as labels) of the privacy level of the order. The image privacy level annotation includes the basic order privacy level annotation and the corresponding derived order privacy level annotation. The original user data anonymization model Model1' is the user data anonymization model Model1 with pre-configured model parameters. The user data anonymization model Model1 includes basic order privacy information recognition and derived order privacy information recognition. The example target image information representation vector indicates the representation vector corresponding to the target image example.

[0097] (B) Based on the original user data anonymization processing model Model1', the example target image information expression vector is extracted by the basic-order private expression vector to obtain the example basic-order private expression vector. Based on the example basic-order private expression vector, basic-order private information is identified to obtain the example image privacy level probability variable corresponding to each basic-order privacy level.

[0098] The example basic-order privacy representation vector is the basic-order privacy representation vector obtained during the optimization process. The example image privacy level probability variable indicates the confidence level of the corresponding privacy level obtained during the optimization process as the privacy level corresponding to the target image example. Optionally, the privacy level of the basic order can be determined by the basic-order privacy information identification unit in the original user data anonymization processing model Model1'. The basic-order privacy information identification unit includes an advanced semantic information mining module and a privacy level determination module. The privacy level is determined based on the advanced semantic information mining module and the privacy level determination module to obtain the example image privacy level probability variable corresponding to each basic-order privacy level.

[0099] (C) Based on the original user data anonymization processing model Model1', the example target image information expression vector is extracted by the derivative-order private expression vector to obtain the example derivative-order private expression vector. The expression vectors are interacted based on the example basic-order private expression vector and the example derivative-order private expression vector to obtain the example derivative-order interaction vector. Then, based on the example derivative-order interaction vector, the derivative-order private information is identified to obtain the example image privacy level probability variable corresponding to each derivative-order privacy level. The privacy level classification accuracy corresponding to the derivative-order privacy level is higher than that corresponding to the basic-order privacy level.

[0100] The example derived-order private representation vector indicates the derived-order private representation vector obtained during the optimization process, and the example derived-order interaction vector indicates the integrated feature vector obtained by the original user data anonymization processing model Model1' during the optimization process. Optionally, the privacy level of the derived order can be determined based on the derived-order private information identification unit in the original user data anonymization processing model Model1'. The derived-order private information identification unit includes an advanced semantic information mining module, an expression vector interaction module, and a privacy level determination module. By determining the privacy level through the advanced semantic information mining module, the expression vector interaction module, and the privacy level determination module, the example image privacy level probability variable corresponding to each derived-order privacy level is obtained.

[0101] (D) Obtain the error between the example image privacy level probability variable corresponding to each basic privacy level and the basic privacy level annotation in the image privacy level annotation to obtain basic privacy level error information, and obtain the error between the example image privacy level probability variable corresponding to each derived privacy level and the derived privacy level annotation in the image privacy level annotation to obtain derived privacy level error information.

[0102] The basic-order privacy level error information represents the error between the example image privacy level probability variable corresponding to the optimized basic-order privacy level and the basic-order privacy level annotation in the image privacy level annotation. The derived-order privacy level error information represents the error between the example image privacy level probability variable corresponding to the optimized derived-order privacy level and the derived-order privacy level annotation in the image privacy level annotation. Optionally, the error between the example image privacy level probability variable corresponding to each basic-order privacy level and the basic-order privacy level annotation in the image privacy level annotation can be obtained through a privacy level determination error function to obtain the basic-order privacy level error information. Similarly, the error between the example image privacy level probability variable corresponding to each derived-order privacy level and the derived-order privacy level annotation in the image privacy level annotation can be obtained to obtain the derived-order privacy level error information. The privacy level determination error function can be, for example, a cross-entropy error function.

[0103] (E) Adjust the original user data anonymization processing model Model1' based on the basic privacy level error information and the derived privacy level error information, and repeat the steps of returning to obtain the target image example and the corresponding image privacy level annotation until the training cutoff condition is met, and obtain the target image desensitization processing model Model1.

[0104] The training cutoff condition indicates the conditions for obtaining the target image anonymization processing model Model1, such as model convergence (meeting the required number of iterations, error within a set range, etc.). The target image anonymization processing model Model1 is the optimized training model used to determine the multi-level privacy level of the target image. Optionally, the training cutoff condition can be determined by obtaining the sum of errors of the basic-level privacy level error information and the derived-level privacy level error information. For example, it can be determined whether the value of the sum of errors is less than a preset value. If not, the original user data anonymization processing model Model1' is adjusted by the sum of errors to obtain the adjusted original user data anonymization processing model Model1'. The adjusted original user data anonymization processing model Model1' is then used as the original user data anonymization processing model Model1', and the steps of obtaining the target image example and the corresponding image privacy level annotation are repeated until the training cutoff condition is met, thus obtaining the target image anonymization processing model Model1.

[0105] This application involves loading a target image example into the original user data anonymization processing model Model1' to extract image information representation from the target image example, obtaining an example target image information representation vector. The example target image information representation vector is then used to extract a basic-order private representation vector, resulting in an example basic-order private representation vector. Based on the example basic-order private representation vector, basic-order private information is identified to obtain the example image privacy level probability variable corresponding to each basic-order privacy level. The example target image information representation vector is then used to extract a derived-order private representation vector, resulting in an example derived-order private representation vector. Based on the example basic-order private representation vector and the example derived-order private representation vector, representation vector interaction is performed to obtain an example derived-order interaction vector. Finally, based on the example derived-order interaction vector, derived-order private information is identified to obtain the example image privacy level probability variable corresponding to each derived-order privacy level. Based on the interaction of the example base-order private representation vector and the example derived-order private representation vector during training, the example derived-order interaction vector is obtained. Then, the derived-order private information is identified. Since the private level classification accuracy corresponding to the derived-order private level is higher than that corresponding to the base-order private level, the hierarchical information of the private level can be learned. Then, the error information of the base-order private level and the error information of the derived-order private level are obtained. Based on the error information of the base-order private level and the error information of the derived-order private level, the model is optimized until the training cutoff condition is met, and the target image desensitization processing model Model1 is obtained. The obtained target image desensitization processing model Model1 has higher accuracy.

[0106] In an optional implementation, the optimization training process of the user data anonymization processing model provided in this application may further include the following processes: (F) Load the target image example into the original user data anonymization processing model Model2', and extract the image information representation of the target image example based on the original user data anonymization processing model Model2' to obtain the example target image information representation vector.

[0107] The original user data anonymization model, Model2', is a second user data anonymization model with pre-configured model parameters. This second model includes basic-level privacy information recognition, intermediate-level privacy level recognition, and derived-level privacy information recognition. Optionally, a target image example can be loaded into the original user data anonymization model, Model2', and the image information representation of the target image example can be extracted using the representation vector extraction network within Model2', resulting in the example target image information representation vector.

[0108] (G) Based on the original user data anonymization processing model Model2', the example target image information expression vector is extracted by the basic-order private expression vector to obtain the example basic-order private expression vector. Based on the example basic-order private expression vector, the basic-order private information is identified to obtain the example image privacy level probability variable corresponding to each basic-order privacy level.

[0109] (H) Based on the original user data anonymization processing model Model2', the intermediate privacy expression vector of the example target image information expression vector is extracted to obtain the example intermediate privacy expression vector. The expression vectors are interacted based on the example basic privacy expression vector and the example intermediate privacy expression vector to obtain the example intermediate interaction vector. The intermediate privacy level is identified based on the example intermediate interaction vector to obtain the example image privacy level probability variable corresponding to each intermediate privacy level.

[0110] The example intermediate privacy representation vector indicates the intermediate privacy representation vector obtained during the training of the original user data anonymization model Model2'. The example intermediate interaction vector is the integrated feature vector obtained during the intermediate optimization process. Optionally, the privacy level of the intermediate privacy level can be determined by the intermediate privacy level recognition network in the original user data anonymization model Model2'. The intermediate privacy level recognition network includes a high-level semantic information mining module, an expression vector interaction module, and a privacy level determination module. By using the high-level semantic information mining module, expression vector interaction module, and privacy level determination module included in the intermediate privacy level recognition network, the intermediate privacy level is identified, and the example image privacy level probability variable corresponding to each intermediate privacy level is obtained. For example, the expression vector interaction module combines and concatenates the input expression vectors and then extracts the jump error expression vector based on the residual module network.

[0111] (I) Based on the original user data anonymization processing model Model2', the example basic-level private expression vector, the example intermediate-level interaction vector, and the example derived-level private expression vector are used to perform expression vector interaction to obtain the example target derived-level interaction vector. Then, based on the example target derived-level interaction vector, derived-level private information is identified to obtain the example target image privacy level probability variable corresponding to each derived-level privacy level. The privacy level classification accuracy corresponding to the derived-level privacy level is higher than that corresponding to the intermediate-level privacy level, and the privacy level classification accuracy corresponding to the intermediate-level privacy level is higher than that corresponding to the basic-level privacy level.

[0112] The example target derived-order interaction vector is the target derived-order interaction vector obtained during the optimization training of the original user data anonymization model Model2'. Optionally, the privacy level of the derived-order privacy level can be determined based on the derived-order privacy information recognition unit in the original user data anonymization model Model2'. This derived-order privacy information recognition unit includes an advanced semantic information mining module, an expression vector interaction module, and a privacy level determination module. Using the advanced semantic information mining module, expression vector interaction module, and privacy level determination module included in the derived-order privacy information recognition unit, the privacy level of the derived order is identified, and the example image privacy level probability variable corresponding to each derived-order privacy level is obtained.

[0113] (J) Obtain the error between the example image privacy level probability variable corresponding to each basic privacy level and the basic privacy level annotation in the image privacy level annotation to obtain basic privacy level error information. Obtain the error between the example image privacy level probability variable corresponding to each intermediate privacy level and the intermediate privacy level annotation in the image privacy level annotation to obtain intermediate privacy level error information. Obtain the error between the example target image privacy level probability variable corresponding to each derived privacy level and the derived privacy level annotation in the image privacy level annotation to obtain derived privacy level target error information.

[0114] The image privacy level annotation includes indication information corresponding to the intermediate privacy level. The intermediate privacy level error information represents the error between the example image privacy level probability variable corresponding to the optimized intermediate privacy level and the intermediate privacy level annotation in the image privacy level annotation. Optionally, the basic-order privacy level error information, intermediate-order privacy level error information, and derived-order privacy level error information can be obtained through a privacy level determination error function, such as the cross-entropy error function.

[0115] (K) Adjust the original user data anonymization processing model Model2' based on the basic privacy level error information, intermediate privacy level error information and derived privacy level target error information, and return the steps of obtaining target image examples and corresponding image privacy level annotations in a loop until the training cutoff condition is met, and obtain the target image desensitization processing model Model2.

[0116] Optionally, the training cutoff condition can be determined by obtaining the sum of errors in the basic privacy level error information, intermediate privacy level error information, and derived privacy level target error information, and then judging whether the training cutoff condition is met. If not, the original user data anonymization processing model Model2' is adjusted based on the error sum to obtain the adjusted original user data anonymization processing model Model2'. The adjusted original user data anonymization processing model Model2' is used as the original user data anonymization processing model Model2', and the steps of obtaining the target image example and the corresponding image privacy level annotation are repeated until the training cutoff condition is met, thus obtaining the target image desensitization processing model Model2. In an optional implementation, the second user data anonymization processing model may include user data anonymization processing models for basic privacy information recognition, multiple intermediate privacy level recognition, and derived privacy information recognition. The optimized target image anonymization processing model Model2, when determining the privacy level of the target image to be identified, can perform basic-level privacy information recognition, multiple intermediate-level privacy level recognition, and derived-level privacy information recognition. This yields the target basic-level privacy level, the target intermediate-level privacy level corresponding to the target basic-level privacy level, and the target derived-level privacy level corresponding to the last target intermediate-level privacy level. This application obtains the basic-level privacy level error information, intermediate-level privacy level error information, and derived-level privacy level error information, and then adjusts the original user data anonymization processing model Model2' based on these error information until the training cutoff condition is met, thus obtaining the target image anonymization processing model Model2 and improving the accuracy of the obtained target image anonymization processing model Model2.

[0117] As an example of an embodiment of this application, a smart community data processing method based on artificial intelligence is provided, including: (10) Obtain the target image to be processed, load the target image to be processed into the target image desensitization processing model Model2, extract the image information expression of the target image to be processed according to the image information expression extraction unit in the target image desensitization processing model Model2, and obtain the target image information expression vector.

[0118] (20) Based on the basic-level privacy information recognition unit in the target image desensitization processing model Model2, the target image information expression vector is extracted by the basic-level privacy expression vector to obtain the basic-level privacy expression vector, and the basic-level privacy information is recognized based on the basic-level privacy expression vector to obtain the image privacy level probability variable corresponding to each basic-level privacy level.

[0119] (30) Based on the intermediate privacy level determination network in the target image desensitization processing model Model2, the target image information expression vector is extracted into an intermediate privacy expression vector to obtain an intermediate privacy expression vector. The expression vectors are interacted based on the basic privacy expression vector and the intermediate privacy expression vector to obtain an intermediate interaction vector. The intermediate privacy level is identified based on the intermediate interaction vector to obtain the image privacy level probability variable corresponding to each intermediate privacy level.

[0120] (40) Based on the derivative-level private information recognition unit in the target image desensitization processing model Model2, the target image information expression vector is extracted into a derivative-level private expression vector to obtain a derivative-level private expression vector. The basic-level private expression vector, the intermediate-level interaction vector and the derivative-level private expression vector are interacted to obtain the target derivative-level interaction vector. Then, based on the target derivative-level interaction vector, derivative-level private information is recognized to obtain the target image privacy level probability variable corresponding to each derivative-level privacy level.

[0121] (50) Obtain the image privacy level position; determine the image privacy level probability variable corresponding to the current basic privacy level from the image privacy level probability variables corresponding to each basic privacy level, obtain the current intermediate privacy level corresponding to the current basic privacy level based on the image privacy level position, and obtain the current derived privacy level corresponding to the current intermediate privacy level.

[0122] (60) Determine the image privacy level probability variable corresponding to the current intermediate privacy level from the image privacy level probability variables corresponding to each intermediate privacy level, and determine the image privacy level probability variable corresponding to the current derived privacy level from the target image privacy level probability variables corresponding to each derived privacy level. Obtain the multiplication result of the image privacy level probability variables corresponding to the current basic privacy level, the current intermediate privacy level, and the current derived privacy level to obtain the current location route probability variable.

[0123] (70) By reviewing the image privacy level probability variables corresponding to each basic privacy level, each intermediate privacy level, and each derived privacy level, the probability variables of each location route are obtained. The target location route probability variable is determined from the probability variables of each location route. Based on the target location route probability variable, the target privacy level corresponding to the target image to be processed is determined. The target privacy level includes the target basic privacy level, the target intermediate privacy level corresponding to the target basic privacy level, and the target derived privacy level corresponding to the target intermediate privacy level.

[0124] Please refer to Figure 4 This is a functional module architecture diagram of the data processing device 110 provided in an embodiment of the present invention. The data processing device 110 can be used to execute a smart community data processing method based on artificial intelligence. The data processing device 110 includes: Image acquisition module 111 is used to acquire the target image to be processed uploaded by the community camera device, extract image information representation from the target image to obtain the target image information representation vector; The vector extraction module 112 is used to extract the basic-order privacy expression vector from the target image information expression vector to obtain the basic-order privacy expression vector, and to perform basic-order privacy information recognition based on the basic-order privacy expression vector to obtain the image privacy level probability variable corresponding to each basic-order privacy level. The vector interaction module 113 is used to extract the derived-order private expression vector from the target image information expression vector to obtain the derived-order private expression vector, perform expression vector interaction based on the basic-order private expression vector and the derived-order private expression vector to obtain the derived-order interaction vector, and then perform derived-order private information recognition based on the derived-order interaction vector to obtain the image privacy level probability variable corresponding to each derived-order privacy level. The privacy level classification accuracy corresponding to the derived-order privacy level is higher than the privacy level classification accuracy corresponding to the basic-order privacy level. The level determination module 114 is used to obtain the privacy level corresponding to the target image to be processed based on the image privacy level probability variables corresponding to each basic privacy level and the image privacy level probability variables corresponding to each derived privacy level. The privacy level includes the target basic privacy level and the target derived privacy level corresponding to the target basic privacy level. The desensitization processing module 115 is used to retrieve the corresponding desensitization strategy according to the privacy level of the target image to be processed, and to perform desensitization processing on the target image to be processed.

[0125] Since the data processing method for smart communities based on artificial intelligence provided in the above embodiments has been described in detail, and the principle of the data processing device 110 is the same as that method, the execution principle of each module of the data processing device 110 will not be described again here.

[0126] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative; for example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0127] In addition, the functional modules in the various embodiments of the present invention can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0128] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, an IoT data server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0129] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0130] It should be understood that for the technical terms for which no explanations are provided above, those skilled in the art can deduce their meanings without doubt based on the disclosed content. The content disclosed in the embodiments of this application is clear and complete to those skilled in the art. It should be understood that the process by which those skilled in the art deduce and analyze the unexplained technical terms based on the disclosed content is based on the content recorded in this application; therefore, the above content is not a judgment of the inventiveness of the overall solution.

[0131] The basic concepts have been described above. Obviously, for those skilled in the art, the detailed disclosure above is merely illustrative and does not constitute a limitation of this application. Although not explicitly stated herein, those skilled in the art can make various modifications, improvements, and corrections to this application. Such modifications, improvements, and corrections are suggested in this application, and therefore remain within the spirit and scope of the exemplary embodiments of this application.

[0132] It should also be understood that, in order to simplify the description disclosed in this application and thus aid in the understanding of at least one embodiment of the invention, multiple features may sometimes be grouped into a single embodiment, drawing, or description thereof in the foregoing description of the embodiments of this application. However, this method of disclosure does not imply that the subject matter of this application requires more features than those mentioned in the claims. In fact, the embodiments contain fewer features than all the features of the single embodiments disclosed above.

Claims

1. A smart community data processing method based on artificial intelligence, characterized in that, Applied to a server, the method includes: The target image to be processed is uploaded by the community camera equipment. Image information representation is extracted from the target image to obtain the target image information representation vector. The target image information expression vector is extracted by extracting the basic-order private expression vector to obtain the basic-order private expression vector, and the basic-order private information is identified based on the basic-order private expression vector to obtain the image privacy level probability variable corresponding to each basic-order privacy level. The target image information representation vector is used to extract the derived-order private representation vector to obtain the derived-order private representation vector. The representation vectors are then interacted based on the basic-order private representation vector and the derived-order private representation vector to obtain the derived-order interaction vector. Then, derived-order private information is identified based on the derived-order interaction vector to obtain the image privacy level probability variable corresponding to each derived-order privacy level. The privacy level classification accuracy corresponding to the derived-order privacy level is higher than that corresponding to the basic-order privacy level. The privacy level classification accuracy is used to characterize the classification granularity or subdivision degree between different privacy levels. The basic order is the order with the smallest division granularity in the tree-like privacy level hierarchy, i.e., the highest-level order. The derived order is the order of the privacy level with a finer and larger division granularity in the tree-like privacy level hierarchy. Based on the image privacy level probability variables corresponding to each basic privacy level and the image privacy level probability variables corresponding to each derived privacy level, the privacy level corresponding to the target image to be processed is obtained. The privacy level includes the target basic privacy level and the target derived privacy level corresponding to the target basic privacy level. Based on the privacy level of the target image to be processed, the corresponding desensitization strategy is retrieved, and the target image to be processed is desensitized. The step of obtaining the derived-order interaction vector by performing expression vector interaction based on the basic-order private expression vector and the derived-order private expression vector includes: The basic-order private representation vector and the derived-order private representation vector are combined to obtain the combined representation vector; Based on the combined expression vector and the derived order private expression vector, the jump error expression vector is extracted to obtain the jump error expression vector. The jump error expression vector is an image feature obtained by jumping and connecting the residual blocks of the residual model. The jump error expression vector is then subjected to feature element coordination to obtain the derived order interaction vector. Feature element coordination involves normalizing the jump error expression vector to a numerical range to achieve data standardization and gradient optimization. The step of extracting the jump error expression vector based on the combined expression vector and the derived-order private expression vector to obtain the jump error expression vector includes: performing correlation projection calculation on the combined expression vector based on preset jump parameters to obtain the projected expression vector, and obtaining the vector sum value of the projected expression vector and the derived-order private expression vector to obtain the jump error expression vector. The step of obtaining the privacy level of the target image to be processed based on the image privacy level probability variables corresponding to each basic privacy level and the image privacy level probability variables corresponding to each derived privacy level includes: Location of image privacy level; Based on the image privacy level location, the image privacy level probability variables corresponding to each basic privacy level and the image privacy level probability variables corresponding to each derived privacy level are integrated to obtain the route probability variables at each location. Determine the target location route probability variable from the location route probability variables, and determine the privacy level of the target image to be processed based on the target location route probability variable; The step of integrating the image privacy level probability variables corresponding to each basic privacy level and each derived privacy level based on the image privacy level location to obtain the location route probability variables includes: determining the image privacy level probability variable corresponding to the current basic privacy level from the image privacy level probability variables corresponding to each basic privacy level; obtaining the current derived privacy level corresponding to the current basic privacy level based on the image privacy level location, and determining the image privacy level probability variable corresponding to the current derived privacy level from the image privacy level probability variables corresponding to each derived privacy level; obtaining the multiplication result of the image privacy level probability variables corresponding to the current basic privacy level and the image privacy level probability variables corresponding to the current derived privacy level to obtain the current location route probability variable; and reviewing the image privacy level probability variables corresponding to each basic privacy level and the image privacy level probability variables corresponding to each derived privacy level to obtain the location route probability variables.

2. The data processing method for smart communities based on artificial intelligence as described in claim 1, characterized in that, The method further includes: The target image to be processed is loaded into the target image desensitization processing model Model1; Based on the target image desensitization processing model Model1, the target image to be processed is extracted to obtain the target image information expression vector; Based on the target image desensitization processing model Model1, the target image information expression vector is extracted into a basic-level private expression vector to obtain a basic-level private expression vector. Based on the basic-level private expression vector, basic-level private information is identified to obtain the image privacy level probability variable corresponding to each basic-level privacy level. Based on the target image desensitization processing model Model1, the target image information expression vector is extracted by the derivative-level private expression vector to obtain the derivative-level private expression vector. The expression vectors are interacted based on the basic-level private expression vector and the derivative-level private expression vector to obtain the derivative-level interaction vector. Then, the derivative-level private information is identified based on the derivative-level interaction vector to obtain the image privacy level probability variable corresponding to each derivative-level privacy level.

3. The data processing method for smart communities based on artificial intelligence as described in claim 2, characterized in that, The target image desensitization processing model Model1 includes an image information expression extraction unit, a basic-level private information recognition unit, and a derived-level private information recognition unit. Loading the target image to be processed into the target image desensitization processing model Model1 includes: loading the target image to be processed into the image information expression extraction unit to extract image information, obtaining a target image information expression vector; loading the target image information expression vector into the basic-level private information recognition unit, extracting a basic-level private expression vector from the target image information expression vector based on the basic-level private information recognition unit, obtaining a basic-level private expression vector; and extracting a basic-level private expression vector based on the basic-level private expression vector. The system obtains the basic-level privacy level support coefficients to get the image privacy level probability variables corresponding to each basic-level privacy level; it loads the target image information expression vector and the basic-level privacy expression vector into the derived-level privacy information recognition unit, and extracts the derived-level privacy expression vector from the target image information expression vector based on the derived-level privacy information recognition unit to get the derived-level privacy expression vector; it performs expression vector interaction based on the basic-level privacy expression vector and the derived-level privacy expression vector to get the derived-level interaction vector, and then obtains the derived-level privacy level support coefficients based on the derived-level interaction vector to get the image privacy level probability variables corresponding to each derived-level privacy level; The derived-order privacy information recognition unit includes a derived-order privacy expression vector extraction subunit, an expression vector interaction subunit, and a derived-order privacy information recognition subunit. The step of loading the target image information expression vector and the basic-order privacy expression vector into the derived-order privacy information recognition unit includes: loading the target image information expression vector into the derived-order privacy expression vector extraction subunit to extract the derived-order privacy expression vector, obtaining the derived-order privacy expression vector; loading the basic-order privacy expression vector and the derived-order privacy expression vector into the expression vector interaction subunit to interact with the expression vector, obtaining the derived-order interaction vector; and loading the derived-order interaction vector into the derived-order privacy information recognition subunit to obtain the derived-order privacy level support coefficient, obtaining the image privacy level probability variable corresponding to each derived-order privacy level.

4. The smart community data processing method based on artificial intelligence as described in claim 3, characterized in that, The method further includes: The target image information representation vector is extracted by the intermediate-order private representation vector to obtain the intermediate-order private representation vector; Based on the basic-order privacy expression vector and the intermediate-order privacy expression vector, the expression vectors are interacted to obtain the intermediate-order interaction vector. Based on the intermediate-order interaction vector, the intermediate-order privacy level is identified to obtain the image privacy level probability variable corresponding to each intermediate-order privacy level. The basic-order private expression vector, the intermediate-order interaction vector, and the derived-order private expression vector are interacted to obtain the target derived-order interaction vector. Then, derived-order private information is identified based on the target derived-order interaction vector to obtain the target image private level probability variable corresponding to each derived-order private level. The private level classification accuracy corresponding to the derived-order private level is higher than that corresponding to the intermediate-order private level, and the private level classification accuracy corresponding to the intermediate-order private level is higher than that corresponding to the basic-order private level. The target privacy level of the target image to be processed is obtained based on the probability variables of the image privacy level corresponding to each basic privacy level, the probability variables of the image privacy level corresponding to each intermediate privacy level, and the probability variables of the target image privacy level corresponding to each derived privacy level. The target privacy level includes the target basic privacy level, the target intermediate privacy level corresponding to the target basic privacy level, and the target derived privacy level corresponding to the target intermediate privacy level. The step of performing expression vector interaction based on the basic-order private expression vector and the intermediate-order private expression vector to obtain the intermediate-order interaction vector includes: combining the basic-order private expression vector and the intermediate-order private expression vector to obtain the intermediate-order combined expression vector; extracting the jump error expression vector based on the intermediate-order combined expression vector and the intermediate-order private expression vector to obtain the intermediate-order jump error expression vector; and organizing the intermediate-order jump error expression vector by feature elements to obtain the intermediate-order interaction vector. The step of performing expression vector interaction on the basic-order private expression vector, the intermediate-order interaction vector, and the derived-order private expression vector to obtain a target derived-order interaction vector includes: combining the basic-order private expression vector, the intermediate-order interaction vector, and the derived-order private expression vector to obtain a target combined expression vector; extracting a jump error expression vector based on the target combined expression vector and the derived-order private expression vector to obtain a target jump error expression vector; and performing feature element aggregation on the target jump error expression vector to obtain the target derived-order interaction vector.

5. The smart community data processing method based on artificial intelligence as described in claim 4, characterized in that, The method further includes: The target image to be processed is loaded into the target image desensitization processing model Model2; Based on the target image desensitization processing model Model2, the target image to be processed is extracted to obtain the target image information expression vector; Based on the target image desensitization processing model Model2, the target image information expression vector is extracted by basic-level private expression vector to obtain basic-level private expression vector, and basic-level private information is identified based on the basic-level private expression vector to obtain the image privacy level probability variable corresponding to each basic-level privacy level. Based on the target image desensitization processing model Model2, the target image information expression vector is extracted into an intermediate-order privacy expression vector to obtain an intermediate-order privacy expression vector. The expression vectors are interacted based on the basic-order privacy expression vector and the intermediate-order privacy expression vector to obtain an intermediate-order interaction vector. The intermediate-order privacy level is identified based on the intermediate-order interaction vector to obtain the image privacy level probability variable corresponding to each intermediate-order privacy level. Based on the target image desensitization processing model Model2, the basic-order private expression vector, the intermediate-order interaction vector, and the derived-order private expression vector are interacted to obtain the target derived-order interaction vector. Then, based on the target derived-order interaction vector, derived-order private information is identified to obtain the target image privacy level probability variable corresponding to each derived-order privacy level. The target privacy level of the target image to be processed is obtained based on the probability variables of the image privacy level corresponding to each basic privacy level, the probability variables of the image privacy level corresponding to each intermediate privacy level, and the probability variables of the target image privacy level corresponding to each derived privacy level. The target privacy level includes the target basic privacy level, the target intermediate privacy level corresponding to the target basic privacy level, and the target derived privacy level corresponding to the target intermediate privacy level.

6. The smart community data processing method based on artificial intelligence as described in claim 5, characterized in that, The method further includes: Obtain a sample target image and its corresponding privacy level annotation; The target image example is loaded into the original user data anonymization processing model Model1', and the image information representation of the target image example is extracted according to the original user data anonymization processing model Model1' to obtain the example target image information representation vector; Based on the original user data anonymization processing model Model1', the example target image information expression vector is extracted by basic-level private expression vector to obtain example basic-level private expression vector. Based on the example basic-level private expression vector, basic-level private information is identified to obtain the example image privacy level probability variable corresponding to each basic-level privacy level. Based on the original user data anonymization processing model Model1', the example target image information expression vector is used to extract the derived-order private expression vector to obtain the example derived-order private expression vector. The expression vectors are interacted based on the example basic-order private expression vector and the example derived-order private expression vector to obtain the example derived-order interaction vector. Then, based on the example derived-order interaction vector, derived-order private information is identified to obtain the example image privacy level probability variable corresponding to each derived-order privacy level. The privacy level classification accuracy corresponding to the derived-order privacy level is higher than the privacy level classification accuracy corresponding to the basic-order privacy level. Obtain the error between the example image privacy level probability variable corresponding to each basic privacy level and the basic privacy level annotation in the image privacy level annotation to obtain basic privacy level error information; and obtain the error between the example image privacy level probability variable corresponding to each derived privacy level and the derived privacy level annotation in the image privacy level annotation to obtain derived privacy level error information. The original user data anonymization processing model Model1' is adjusted based on the basic privacy level error information and the derived privacy level error information, and the steps of returning to obtain the target image example and the corresponding image privacy level annotation are repeated until the training cutoff condition is met, thus obtaining the target image desensitization processing model Model1.

7. The smart community data processing method based on artificial intelligence as described in claim 6, characterized in that, The method further includes: The target image example is loaded into the original user data anonymization processing model Model2', and the image information representation of the target image example is extracted according to the original user data anonymization processing model Model2' to obtain the example target image information representation vector; Based on the original user data anonymization processing model Model2', the basic-order private expression vector of the example target image information expression vector is extracted to obtain the example basic-order private expression vector. Based on the example basic-order private expression vector, basic-order private information is identified to obtain the example image privacy level probability variable corresponding to each basic-order privacy level. Based on the original user data anonymization processing model Model2', the intermediate-order privacy expression vector of the example target image information expression vector is extracted to obtain the example intermediate-order privacy expression vector. Based on the example basic-order privacy expression vector and the example intermediate-order privacy expression vector, the expression vectors are interacted to obtain the example intermediate-order interaction vector. Based on the example intermediate-order interaction vector, the intermediate-order privacy level is identified to obtain the example image privacy level probability variable corresponding to each intermediate-order privacy level. Based on the original user data anonymization processing model Model2', the example basic-level private expression vector, the example intermediate-level interaction vector, and the example derived-level private expression vector are interacted to obtain the example target derived-level interaction vector. Then, based on the example target derived-level interaction vector, derived-level private information is identified to obtain the example target image privacy level probability variable corresponding to each derived-level privacy level. The privacy level classification accuracy corresponding to the derived-level privacy level is higher than that corresponding to the intermediate-level privacy level, and the privacy level classification accuracy corresponding to the intermediate-level privacy level is higher than that corresponding to the basic-level privacy level. The error between the example image privacy level probability variable corresponding to each basic privacy level and the basic privacy level annotation in the image privacy level annotation is obtained to obtain basic privacy level error information. The error between the example image privacy level probability variable corresponding to each intermediate privacy level and the intermediate privacy level annotation in the image privacy level annotation is obtained to obtain intermediate privacy level error information. The error between the example target image privacy level probability variable corresponding to each derived privacy level and the derived privacy level annotation in the image privacy level annotation is obtained to obtain derived privacy level target error information. The original user data anonymization processing model Model2' is adjusted based on the basic privacy level error information, the intermediate privacy level error information, and the derived privacy level target error information. The steps of returning to obtain the target image example and the corresponding image privacy level annotation are repeated until the training cutoff condition is met, thus obtaining the target image desensitization processing model Model2.

8. A smart community data processing system based on artificial intelligence, characterized in that, The system includes servers and community camera equipment that communicate with each other. The servers include a memory and a processor. The memory stores a computer program that, when the processor runs the computer program, implements the method as described in any one of claims 1 to 7.

Images