Data detection method, device and equipment and computer storage medium

Abstract

The invention discloses a data detection method and device, equipment and a computer storage medium. The method comprises the following steps: acquiring target data in a cloud disk, and encrypting the target data by using an encryption algorithm to obtain encrypted data; and under the condition that the encrypted data is not matched with the data of the illegal content database, extracting key frame data from the video data. And inputting the key frame data into an image semantic segmentation model, extracting first feature information and semantic information in the key frame data by using the image semantic segmentation model to obtain semantic segmentation images, and classifying the semantic segmentation images by using the image semantic segmentation model to obtain a figure image set and a scene image set. And calculating the feature vector of the corresponding image set based on a classifier in the first target model to obtain a target category probability value. According to the invention, whether the video file is illegal is determined by extracting the key frame data from the video data, so that the time for detecting the video file is shortened, and the detection efficiency of the video file can be improved.

Description

Technical Field

[0001] The present invention relates to the field of cloud computing, and in particular to a method, device, equipment, computer storage medium and computer program product for data detection. Background Art

[0002] With the development of informatization, more and more users choose to store their private data in cloud disks.

[0003] At present, there is a phenomenon that some users store illegal video files in cloud disks, which makes illegal video files spread through cloud disk sharing, causing great impact on social security. In order to prevent the spread and preservation of illegal video files, in related technologies, manual detection is usually used to identify and block video files saved by users. However, it takes a long time to detect video illegal information using manual detection, resulting in low detection efficiency. Summary of the invention

[0004] The embodiments of the present disclosure provide a data detection method, apparatus, device, computer storage medium, and computer program product, which can improve the detection efficiency of video files.

[0005] In a first aspect, an embodiment of the present disclosure provides a method for data detection, the method comprising:

[0006] Obtain target data in the cloud disk, the target data includes video data;

[0007] Inputting the key frame data in the video data into the image semantic segmentation model, extracting the first feature information and semantic information in the key frame data using the image semantic segmentation model to obtain a semantic segmentation image, and classifying the semantic segmentation image using the classification module in the image semantic segmentation model to obtain a person image set and a scene image set;

[0008] Calculating the feature vector of the corresponding person image set or scene image set based on the classifier in the first target model to obtain a target category probability value;

[0009] When the target category probability value is greater than the target threshold, the video data is marked as having illegal video information.

[0010] In an achievable implementation, before inputting the key frame data in the video data into the image semantic segmentation model, the data detection method further includes:

[0011] Encrypt the target data using an encryption algorithm to obtain encrypted data;

[0012] extracting key frame data from the video data when the encrypted data does not match the data in the illegal content database;

[0013] Extract key frame data from video data, including:

[0014] Performing a frame interception operation on the video data to obtain a plurality of image frame data;

[0015] Based on the corresponding image entropies of the plurality of image frame data, an image frame corresponding to an image entropy whose difference between different image entropies is greater than a preset threshold or an image entropy that is an extreme value is selected to obtain candidate image frame data;

[0016] A differential operation is performed on the candidate image frame data to extract key frame data from the candidate image frame data.

[0017] In one practicable implementation, extracting first feature information and semantic information in key frame data using an image semantic segmentation model to obtain a semantic segmentation image includes:

[0018] Using a feature extraction module in an image semantic segmentation model, extract features from the key frame data to obtain first feature information;

[0019] Using a semantic extraction module in an image semantic segmentation model, extracting semantic boundaries from the first feature information to obtain semantic information;

[0020] The first feature information and the semantic information are fused using a fusion module in the image semantic segmentation model to obtain a semantic segmentation image.

[0021] In an achievable implementation, the first target model includes a plurality of sub-models, and the sub-models are used to identify different types of violations; the types include a first type, and the specific process of the sub-model corresponding to the first type identifying the first type of violation includes:

[0022] Using the feature extraction module in the sub-model corresponding to the first type, a feature extraction operation is performed on the person images in the person image set to obtain scale-invariant feature transform (SIFT) feature information;

[0023] Using a first preset algorithm, calculating optical flow feature vector information of a current frame and a next frame of a person image;

[0024] Calculate the image color feature information of the person image using a second preset algorithm;

[0025] The SIFT feature information, the optical flow feature vector information and the image color feature information are concatenated to obtain the second feature information;

[0026] The second feature information is converted into a feature vector.

[0027] In an achievable implementation, the first target model includes a plurality of sub-models, and the sub-models are used to identify different types of violations; the type includes a second type, and the sub-model corresponding to the second type identifies the second type of violation. The specific process includes:

[0028] Performing a downsampling operation on the scene images in the scene image set to obtain operation information;

[0029] Using the first convolution module in the sub-model corresponding to the second type to perform a convolution operation on the scene image to obtain convolution mapping information;

[0030] Fusing the operation information and the convolution mapping information to obtain fused information;

[0031] Using the second convolution module and the third convolution module in the sub-model corresponding to the second type to extract features from the fused information, obtaining third feature information;

[0032] The third feature information is converted into a feature vector.

[0033] In an achievable implementation, the target data further includes text data, and when the encrypted data does not match the data in the illegal content database, the method further includes:

[0034] Using the second target model to perform a truncation operation on the text of the text data to obtain a truncation text;

[0035] Based on the data information in the preset database, calculate the similarity information of the intercepted text;

[0036] When the number of intercepted texts whose similarity information is greater than the first threshold is greater than the second threshold, it is marked that illegal text information exists in the text data.

[0037] In an achievable implementation, when the second target model is a sensitive word model, using the second target model to perform a text interception operation on the text data to obtain the intercepted text includes:

[0038] Use the sensitive word model to segment the text data to obtain a word sequence;

[0039] Based on the data information in the preset database, similarity information of the intercepted text is calculated; when the number of the intercepted texts whose similarity information is greater than the first threshold is greater than the second threshold, marking the text data as having illegal text information includes:

[0040] Based on the data information in the preset sensitive word database, calculate the sensitive similarity information of the words in the word sequence;

[0041] When the number of words whose sensitive similarity information is greater than the first threshold is greater than the second threshold, it is marked that illegal text information exists in the text data.

[0042] In an achievable implementation, when the second target model is a topic model, using the second target model to perform a text interception operation on the text data to obtain the intercepted text includes:

[0043] Use the topic model to extract the text topics of the text data to obtain the text topics;

[0044] Based on the data information in the preset database, similarity information of the intercepted text is calculated; when the number of the intercepted texts whose similarity information is greater than the first threshold is greater than the second threshold, marking the text data as having illegal text information includes:

[0045] Based on the data information in the preset illegal topic database, calculating the illegal topic similarity information of the text topic;

[0046] When the number of text topics whose illegal topic similarity information is greater than the first threshold is greater than the second threshold, it is marked that illegal text information exists in the text data.

[0047] In one possible implementation, the target data further includes image data, and when the encrypted data does not match the data in the illegal content database, the method further includes:

[0048] inputting the image data into a third target model;

[0049] Using the classifier in the third target model to calculate the feature vector of the image data, to obtain the image category probability value;

[0050] When the image category probability value is greater than the third threshold, the image data is marked as having illegal image information.

[0051] In an achievable implementation, the data detection method further includes:

[0052] When it is detected that the number of target data of any one of illegal video information, illegal text information and illegal image information is greater than a preset number threshold, the user corresponding to the target data is marked as an illegal user.

[0053] In a second aspect, an embodiment of the present disclosure provides a data detection device, the device comprising:

[0054] An acquisition module is used to acquire target data in the cloud disk, and the target data includes video data;

[0055] An input module is used to input the key frame data into the image semantic segmentation model, extract the first feature information and semantic information in the key frame data using the image semantic segmentation model to obtain a semantic segmentation image, and classify the semantic segmentation image using the classification module in the image semantic segmentation model to obtain a person image set and a scene image set;

[0056] A calculation module, used to calculate the feature vector of the corresponding person image set or scene image set based on the classifier in the first target model to obtain a target category probability value;

[0057] The marking module is used to mark the video data as having illegal video information when the target category probability value is greater than the target threshold.

[0058] In a third aspect, an embodiment of the present disclosure provides a data detection device, the device comprising a processor and a memory storing computer program instructions; the processor reads and executes the computer program instructions to implement a data detection method as described in any one of the first aspects.

[0059] In a fourth aspect, an embodiment of the present disclosure provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer program instructions, and when the computer program instructions are executed by a processor, a method for data detection as described in any one of the first aspects is implemented.

[0060] In a fifth aspect, an embodiment of the present disclosure provides a computer program product, including a computer program, which implements a data detection method as described in any one of the first aspects when the computer program is executed by a processor.

[0061] The embodiments of the present disclosure provide a method, apparatus, device, computer storage medium and computer program product for data detection. The method obtains target data in a cloud disk, and the target data includes video data, inputs key frame data in the video data into an image semantic segmentation model, uses the image semantic segmentation model to extract first feature information and semantic information in the key frame data to obtain a semantic segmentation image, and uses the classification module in the image semantic segmentation model to classify the semantic segmentation image to obtain a set of character images and a set of scene images. Based on the classifier in the first target model, the feature vector of the corresponding set of character images or the set of scene images is calculated to obtain a target category probability value. When the target category probability value is greater than the target threshold, the video data is marked as having illegal video information. It can be seen that the present disclosure reduces the time for detecting video files by extracting key frame data from video data to determine whether a video file is illegal, thereby improving the efficiency of detecting video files. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the technical solution of the embodiment of the present invention, the following is a brief introduction to the drawings required for use in the embodiment of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0063] Figure 1 is a flow chart of a data detection method provided by an embodiment of the present disclosure;

[0064] Figure 2 is a flowchart of a method for extracting key frame data provided by an embodiment of the present disclosure;

[0065] Figure 3 It is a flowchart of a method for obtaining a semantic segmentation image provided by an embodiment of the present disclosure;

[0066] Figure 4 is a flowchart of a method for detecting text data provided by an embodiment of the present disclosure;

[0067] Figure 5 is a flowchart of a method for detecting image data provided by an embodiment of the present disclosure;

[0068] Figure 6 is a structural schematic diagram of a data detection device provided by an embodiment of the present disclosure;

[0069] Figure 7 It is a structural diagram of a data detection device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0070] The features and exemplary embodiments of various aspects of the present invention will be described in detail below. In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present invention, rather than to limit the present invention. For those skilled in the art, the present invention can be implemented without the need for some of these specific details. The following description of the embodiments is only to provide a better understanding of the present invention by illustrating examples of the present invention.

[0071] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "include..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0072] Before describing the technical solutions provided by the embodiments of the present disclosure, in order to facilitate understanding of the embodiments of the present disclosure, the present disclosure specifically describes the problems existing in the related technologies:

[0073] The explosive growth of information on the Internet has opened up an era of Internet information explosion. If you want to have advanced productivity, you must rely on informatization, and the development of informatization is also the general trend of the times. Due to the rise of cloud disk (network disk) technology in recent years, cloud disks are more and more popular among users for their security, stability, massive storage, and the ability to share files anytime and anywhere.

[0074] At present, there is a phenomenon that some users store illegal video files in cloud disks, which makes illegal video files spread through cloud disk sharing, causing great impact on social security. In order to prevent the spread and preservation of illegal video files, in related technologies, manual detection is usually used to identify and block video files saved by users. However, it takes a long time to detect video illegal information using manual detection, resulting in low detection efficiency.

[0075] The embodiments of the present disclosure provide a method, apparatus, device, computer storage medium and computer program product for data detection, which can solve the above-mentioned technical problems existing in the related technologies.

[0076] The disclosed embodiment obtains target data in a cloud disk, and the target data includes video data. The key frame data in the video data is input into an image semantic segmentation model, and the image semantic segmentation model is used to extract the first feature information and semantic information in the key frame data to obtain a semantic segmentation image. The semantic segmentation image is classified using a classification module in the image semantic segmentation model to obtain a set of character images and a set of scene images. The feature vectors of the corresponding set of character images or set of scene images are calculated based on the classifier in the first target model to obtain a target category probability value. It can be seen that the disclosed embodiment reduces the time for detecting video files by extracting key frame data from video data to determine whether a video file is in violation, thereby improving the efficiency of detecting video files.

[0077] The following is an introduction to the data detection method provided by the embodiment of the present disclosure.

[0078] Figure 1 FIG. 1 is a flow chart of a method for data detection provided by an embodiment of the present disclosure. Figure 1 The method may include the following steps S110-S140.

[0079] S110: Acquire target data in the cloud disk.

[0080] In one embodiment, the system obtains the target data saved by the user from the storage space in the cloud disk according to the recorded storage path information for detecting the target data and determining whether the user is an illegal user.

[0081] The target data represents the data to be detected, and the target data may include video data.

[0082] S120: Input the key frame data in the video data into the image semantic segmentation model, use the image semantic segmentation model to extract the first feature information and semantic information in the key frame data to obtain a semantic segmentation image, and use the classification module in the image semantic segmentation model to classify the semantic segmentation image to obtain a person image set and a scene image set.

[0083] After extracting the key frame data, the system inputs the key frame data into the image semantic segmentation model through the interface, and uses the extraction module in the image semantic segmentation model to extract the first feature information and semantic information in the key frame data to obtain a semantic segmentation image. The semantic segmentation image assigns a category label to each pixel in the image, which is used to parse the content in the image and facilitate the subsequent classification of the semantic segmentation image. According to the first feature information and semantic information in the semantic segmentation image, the classification model in the image semantic segmentation model is used to classify the semantic segmentation image to obtain a set of person images and a set of scene images. By classifying the semantic segmentation images into different image sets, each image set can be detected for a fixed detection subject, thereby improving the detection results.

[0084] S130: Calculate the feature vector of the corresponding person image set or scene image set based on the classifier in the first target model to obtain a target category probability value.

[0085] After obtaining the character image set and the scene image set, the system uses the corresponding feature extraction module to extract features from the character image set and the scene image set to obtain feature vectors of the character image set and the scene image set. After obtaining the feature vectors of the character image set and the scene image set, the feature vectors of the character image set and the scene image set are input into the first target model, and the feature vectors of the character image set or the scene image set are calculated using the classifier in the first target model to obtain the target category probability value corresponding to the character image set or the scene image set. By calculating the target category probability value corresponding to the character image set or the scene image set, the possibility of violation of the image set can be analyzed according to the target category probability value.

[0086] The target category probability value represents the probability that a person image set or a scene image set contains illegal information.

[0087] In one example, the classifier may include a binary classifier, and the binary classifier may be constructed based on the Elman algorithm.

[0088] S140: When the target category probability value is greater than the target threshold, marking the video data as having illegal video information.

[0089] The system reads the preset target threshold and compares the preset target threshold with the calculated target category probability threshold through the comparison module to obtain a comparison result. If the comparison result indicates that the calculated target category probability value is greater than the preset target threshold, the video data is marked as having illegal video information. By obtaining the target category probability value of the image set and comparing it with the target threshold, the method of identifying illegal video data is standardized, which facilitates the determination of illegal video data.

[0090] In one embodiment, the classifier in the first target model calculates the feature vector of the person image set and obtains a target category probability threshold of 0.8, which is greater than a preset target threshold of 0.6, and then marks the video data as having illegal video information.

[0091] The disclosed embodiment obtains target data in a cloud disk, and the target data includes video data. The key frame data in the video data is input into an image semantic segmentation model, and the image semantic segmentation model is used to extract the first feature information and semantic information in the key frame data to obtain a semantic segmentation image. The semantic segmentation image is classified using a classification module in the image semantic segmentation model to obtain a set of character images and a set of scene images. The feature vectors of the corresponding set of character images or set of scene images are calculated based on the classifier in the first target model to obtain a target category probability value. It can be seen that the disclosed embodiment reduces the time for detecting video files by extracting key frame data from video data to determine whether a video file is in violation of regulations, thereby improving the efficiency of detecting video files.

[0092] In one embodiment, Figure 2 As shown, before the key frame data in the video data is input into the image semantic segmentation model, the data detection method further includes step S150 and step S160.

[0093] S150: Encrypt the target data using an encryption algorithm to obtain encrypted data.

[0094] The system encrypts the target data based on the data information of the target data according to the preset encryption algorithm in the encryption component to obtain encrypted data corresponding to the data information of the target data. By encrypting the target data, encrypted data reflecting the data characteristics of the target data can be obtained, which is convenient for subsequent comparison to see whether the target data contains illegal information.

[0095] The encrypted data represents the encrypted value, and each target data can be encrypted according to its own data information to obtain a corresponding encrypted data, namely, an encrypted value.

[0096] In one example, the encryption algorithm may include an MD5 encryption algorithm, and the encrypted data may include an MD5 value.

[0097] S160: When the encrypted data does not match the data in the illegal content database, extract key frame data from the video data.

[0098] After obtaining the encrypted data corresponding to the target data, the system reads the encrypted data of the illegal content database through the interface, and matches the encrypted data with the encrypted data of the illegal content database to determine whether the target data contains illegal information. In the case that the encrypted data corresponding to the target data does not match the data in the illegal content database, the system uses the image frame extraction component to extract key frame data from the video data to facilitate further detection.

[0099] The key frame data represents the image frame of the video data.

[0100] In one example, the illegal content database may include multiple illegal data, each illegal data having corresponding encrypted data such as "9871", "2721" and so on.

[0101] The disclosed embodiment can reduce the time for determining whether video data contains illegal information and improve the review efficiency by matching the encrypted data corresponding to the target data with the data in the illegal content database.

[0102] In one embodiment, extracting key frame data from video data may include steps S161-S163:

[0103] S161: Perform frame capture operation on the video data to obtain multiple image frame data.

[0104] The system uses an image frame capture tool to perform frame capture operations on video data to obtain multiple continuous image frame data, which is convenient for subsequent acquisition of alternative image frames.

[0105] S162: Based on the corresponding image entropies of the plurality of image frame data, select image frames whose image entropies have a difference value between different image entropies greater than a preset threshold or whose image entropies are extreme values, to obtain candidate image frame data.

[0106] The image entropy calculation formula is used to calculate the image entropy corresponding to the data of multiple image frames. According to the image entropy corresponding to the data of multiple image frames, the image entropy whose difference between different image entropies is greater than the preset threshold or the image entropy corresponding to the extreme value is selected to obtain the candidate image frame, wherein the candidate image frame is most likely to contain the illegal information. By selecting the candidate image frame, it is possible to effectively identify whether the video data contains illegal information.

[0107] In one example, after calculating the image entropy corresponding to the data of multiple image frames, if the image entropy difference between the first image frame and the 24th image frame is 50, which is greater than the preset threshold value of 20, the first image frame and the 24th image frame are used as candidate image frames. If the image entropy of the 30th image frame is a maximum value of 100, the 30th image frame is used as a candidate image frame.

[0108] In one example, the expression for calculating image entropy is shown in formula (1):

[0109] E(f i ) = - ∑ j pf i (j)log pf i (j) (1)

[0110] Among them, i and j are indicators; E(f i ) is f i Image entropy of the frame; pf i (j) is f i Probability density function of the frame.

[0111] S163: performing a differential operation on the candidate image frame data to extract key frame data from the candidate image frame data.

[0112] After obtaining the candidate image frame data, the system uses a differential operation component to extract key frame data in the candidate image frame according to local frame differences of the candidate image frame data, so that the key frame data can be used as a representative of the candidate image frame data.

[0113] The disclosed embodiment reduces the number of image frames used for detection by intercepting video data and screening and performing differential extraction operations on image frames according to image entropy, thereby reducing the time for data detection and improving the efficiency of data detection.

[0114] In one embodiment, Figure 3 As shown, using an image semantic segmentation model to extract first feature information and semantic information in key frame data to obtain a semantic segmentation image may include steps S121-S123.

[0115] S121: Utilize the feature extraction module in the image semantic segmentation model to perform feature extraction on the key frame data to obtain first feature information.

[0116] The first feature information is used to form a semantic feature image.

[0117] The system utilizes the feature extraction module in the image semantic segmentation model to extract the first feature information from the image corresponding to the key frame data.

[0118] The first feature information may include an original image feature map, the feature extraction module may include a shape flow module, and the shape flow module is constructed based on the MoIFT_CLD algorithm.

[0119] S122: Utilize the semantic extraction module in the image semantic segmentation model to perform semantic boundary extraction on the first feature information to obtain semantic information.

[0120] The semantic extraction module in the image semantic segmentation model is used to perform semantic boundary extraction on the first feature information, and semantic information is extracted from the first feature information. The semantic information is used to form a semantic feature image.

[0121] The semantic extraction module may include a shape flow module, and the semantic information may include semantic boundary information.

[0122] S123: Using a fusion module in the image semantic segmentation model, the first feature information and the semantic information are fused to obtain a semantic segmentation image.

[0123] After the first feature information and the semantic information are acquired, the first feature information and the semantic information are fused using a fusion module in the image semantic segmentation model to obtain a semantic segmentation image.

[0124] The disclosed embodiment can obtain an accurate semantic segmentation image by extracting feature information and semantic information of key frame data, thereby facilitating subsequent data detection.

[0125] In one embodiment, the first target model may include multiple sub-models, and the sub-models are used to identify different types of violations; the type may include a first type, and the specific process of the sub-model corresponding to the first type identifying the first type of violation may include steps S210-S250.

[0126] S210: Using the feature extraction module in the sub-model corresponding to the first type, perform a feature extraction operation on the person images in the person image set to obtain SIFT feature information.

[0127] Through the feature extraction module in the sub-model corresponding to the first type, feature extraction operations are performed on the character images in the character image set to extract the Scale Invariant Feature Transform (SIFT) feature information in the character images, which is convenient for subsequent fusion with other feature information to obtain the second feature information.

[0128] Among them, the sub-model corresponding to the first type can be created by the MoIFT_CLD-BOVW-Elman algorithm.

[0129] S220: Calculate optical flow feature vector information of the current frame and the next frame of the character image using a first preset algorithm.

[0130] Through the first preset algorithm, based on the feature points on the character image, the optical flow feature vector information of the current frame and the next frame of the character image are calculated respectively, so as to facilitate the subsequent fusion with other feature information to obtain the second feature information.

[0131] Among them, the first preset algorithm may include an optical flow method.

[0132] S230: Calculate the image color feature information of the person image using a second preset algorithm.

[0133] The image color feature information of the character image is calculated and extracted through the second preset algorithm, so as to facilitate subsequent fusion with other feature information to obtain the second feature information.

[0134] The second preset algorithm may include a color layout descriptor (Color Layout Descriptor, CLD) algorithm.

[0135] S240: concatenating the SIFT feature information, the optical flow feature vector information, and the image color feature information to obtain second feature information.

[0136] After obtaining SIFT feature information, optical flow feature vector information and image color feature information, the SIFT feature information, optical flow feature vector information and image color feature information are spliced ​​in a preset splicing order to obtain second feature information. The second feature information reflects multiple feature information of the image, which facilitates subsequent identification of whether the video data includes illegal information based on the multiple feature information of the image.

[0137] The second feature information may include the MoSIFT_CLD feature of the character image of the current frame.

[0138] S250: Convert the second feature information to obtain a feature vector.

[0139] After the second feature information is acquired, the second feature information is converted into a feature vector using a preset conversion method so that the format type of the feature vector is suitable for the sub-model corresponding to the first type.

[0140] In one embodiment, the feature clustering module in the sub-model corresponding to the first type may be used to perform feature clustering on the second feature information to obtain a feature vector.

[0141] The disclosed embodiment obtains different feature information respectively and fuses them to obtain target feature information, thereby determining a feature vector.

[0142] In one embodiment, the first target model may include multiple sub-models, and the sub-models are used to identify different types of violations; the type may include a second type, and the sub-model corresponding to the second type identifies the second type of violation. The specific process may include steps S2060-S2100.

[0143] S2060: Perform a downsampling operation on the scene images in the scene image set to obtain operation information.

[0144] The sub-model corresponding to the second type is used to perform a down-sampling operation on the scene images in the scene image set to obtain operation information corresponding to the down-sampling operation, so as to facilitate subsequent fusion with other information to obtain fused information.

[0145] S2070: Use the first convolution module in the sub-model corresponding to the second type to perform a convolution operation on the scene image to obtain convolution mapping information.

[0146] A convolution operation is performed on the scene image through the first convolution module in the sub-model corresponding to the second type to obtain convolution mapping information corresponding to the convolution operation, so as to facilitate subsequent fusion with other information to obtain fused information.

[0147] Among them, the sub-model corresponding to the second type can be created by the LFMCNN algorithm.

[0148] S2080: Fuse the operation information and the convolution mapping information to obtain fused information.

[0149] The acquired operation information and convolution mapping information are fused using the fusion module in the sub-model corresponding to the second type to obtain fused information.

[0150] In one example, the fusion information may include space, color, contour, and other information of the scene image.

[0151] S2090: Use the second convolution module and the third convolution module in the sub-model corresponding to the second type to perform feature extraction on the fused information to obtain third feature information. The third feature information reflects multiple feature information of the image, which facilitates subsequent identification of whether the video data includes illegal information based on the multiple feature information of the image.

[0152] After the fusion information is obtained, a feature extraction operation is performed on the fusion information through the second convolution module and the third convolution module in the sub-model corresponding to the second type to extract feature information of the fusion information and obtain third feature information.

[0153] S2100: Convert the third feature information to obtain a feature vector.

[0154] After obtaining the third feature information, the system converts the third feature information into a feature vector through a preset conversion method in the conversion module, so that the format type of the feature vector is suitable for the sub-model corresponding to the second type.

[0155] The disclosed embodiment performs downsampling and convolution operations on the scene image, fuses the operation information with the convolution mapping information, obtains fusion information and performs feature extraction, thereby determining the feature vector.

[0156] In one embodiment, the target data may also include text data. In the case where the encrypted data does not match the data in the illegal content database, the data detection method may further include steps S310-S330. Figure 4 shown.

[0157] S310: Using the second target model to perform a truncation operation on the text of the text data to obtain a truncation text.

[0158] By using the preset interception method in the second target model, the text of the text data is intercepted to obtain the intercepted text corresponding to the interception operation. By intercepting the text of the text data, each text can be detected to improve the accuracy of the detection.

[0159] In one example, a preset interception method can be used to intercept the text "Xiao Li sings, Xiao Wang runs, and Xiao Hong reads" into three intercepted texts of "Xiao Li sings", "Xiao Wang runs", and "Xiao Hong reads".

[0160] S320: Calculate similarity information of the intercepted text based on the data information in the preset database.

[0161] After the intercepted text is obtained, based on the data information stored in the database, a preset similarity calculation method is used to calculate the similarity information between the intercepted text and the data information stored in the database, so as to facilitate subsequent comparison.

[0162] S330: When the number of intercepted texts whose similarity information is greater than the first threshold is greater than the second threshold, it is marked that the text data contains illegal text information.

[0163] After calculating the similarity information of the intercepted text, the similarity information is compared with the first threshold and the number of intercepted texts with similarity information greater than the first threshold is calculated. If the number of intercepted texts with similarity information greater than the first threshold is greater than the second threshold, the text data composed of the intercepted text is marked as having illegal text information.

[0164] The values ​​of the first threshold and the second threshold are not fixed.

[0165] The disclosed embodiment can effectively determine whether the text data contains illegal text information by intercepting the text data and calculating the similarity information of the text data.

[0166] In one embodiment, when the second target model is a sensitive word model, performing a clipping operation on the text data using the second target model to obtain the clipped text may include step S311.

[0167] S311: Use the sensitive word model to perform word segmentation on the text data to obtain a word sequence.

[0168] By performing a word segmentation operation on the text data through the sensitive word model, a word sequence including multiple words is obtained. By performing a word segmentation operation on the text data, each word can be detected to improve the accuracy of the detection.

[0169] In one embodiment, a sentiment calculation method may be used to select keywords in a word sequence to obtain a keyword sequence, that is, a new word sequence.

[0170] In one example, the expression of the sentiment calculation method is shown in formula (2):

[0171]

[0172] Among them, P c is a keyword sequence, c and t are both indicator parameters, is the sentiment content value, is the emotional parameter, Z is the key parameter, is the natural constant e Second power, is the natural constant e Power.

[0173] In one embodiment, based on the data information in a preset database, similarity information of the intercepted text is calculated; when the number of intercepted texts whose similarity information is greater than a first threshold is greater than a second threshold, marking the text data for the presence of illegal text information may include step S321 and step S331.

[0174] S321: Based on the data information in the preset sensitive word database, calculate the sensitive similarity information of the words in the word sequence.

[0175] After obtaining the word sequence, based on the sensitive word data information stored in the preset sensitive word database, a preset sensitive similarity calculation method is used to calculate the corresponding sensitive similarity information between each word in the sensitive word sequence and the data information in the sensitive word database, so as to facilitate subsequent comparison.

[0176] In one example, the expression for calculating sensitive similarity information is shown in formula (3):

[0177]

[0178] Among them, P is the set of sensitive words in the sensitive word database, P i is the i-th sensitive word in the sensitive word set, S i is the sensitivity coefficient, P ci is the current keyword, Similarity(P ci ,P i ) and Similarity(Pci ) is the sensitive similarity function, count(P) is the total number of sensitive words in the sensitive word set P.

[0179] S331: When the number of words whose sensitive similarity information is greater than the first threshold is greater than the second threshold, it is marked that illegal text information exists in the text data.

[0180] After calculating the sensitive similarity information of each word in the word sequence, the sensitive similarity information is compared with the first threshold and the number of words whose sensitive similarity information is greater than the first threshold is calculated. If the number of words whose sensitive similarity information is greater than the first threshold is greater than the second threshold, the text data composed of the marked words contains illegal text information.

[0181] The disclosed embodiment can effectively determine whether there is illegal text information related to sensitive words in the text data by performing word segmentation processing on the text data and calculating the sensitive similarity information of the words.

[0182] In one embodiment, when the second target model is a topic model, performing a truncation operation on text data using the second target model to obtain truncation text may include step S312.

[0183] S312: Using the topic model to perform text topic extraction processing on the text of the text data to obtain the text topic.

[0184] The text of the text data is extracted and processed based on a preset extraction method through a topic model to obtain multiple text topics. By performing extraction and processing operations on the text data, each text topic can be detected to improve the accuracy of detection.

[0185] Among them, the topic model may include a BTM (Biterm) topic model.

[0186] In one embodiment, based on the data information in a preset database, similarity information of the intercepted text is calculated; when the number of intercepted texts whose similarity information is greater than a first threshold is greater than a second threshold, marking the text data for the presence of illegal text information may include step S322 and step S332.

[0187] S322: Based on the data information in the preset illegal topic database, calculate the illegal topic similarity information of the text topic.

[0188] After obtaining the text topic, based on the illegal topic data information stored in the preset illegal topic database, the preset topic similarity calculation method is used to calculate the illegal topic similarity information corresponding to each text topic and the data information in the illegal topic database, so as to facilitate subsequent comparison.

[0189] S332: When the number of text topics whose similarity information of illegal topics is greater than the first threshold is greater than the second threshold, it is marked that illegal text information exists in the text data.

[0190] After calculating the illegal topic similarity information of the text topic, the illegal topic similarity information is compared with the first threshold and the number of text topics whose illegal topic similarity information is greater than the first threshold is calculated. If the number of text topics whose illegal topic similarity information is greater than the first threshold is greater than the second threshold, the text data composed of the marked text topic contains illegal text information.

[0191] The embodiment of the present disclosure can effectively determine whether there is illegal text information related to the illegal topic in the text data by extracting and processing the text data and calculating the illegal topic similarity information of the text topic.

[0192] In one embodiment, the target data may also include image data. In the case where the encrypted data does not match the data in the illegal content database, the data detection method further includes steps S410-S430. Figure 5 shown.

[0193] S410: Input the image data into the third target model.

[0194] After the image data is acquired, the image data is input into the third object model.

[0195] S420: Utilize the classifier in the third target model to calculate the feature vector of the image data to obtain an image category probability value.

[0196] After the feature vector of the image data is obtained by using the feature extraction module in the third target model, the feature vector of the image data is calculated by the classifier in the third target model to obtain the image category probability value corresponding to the image data, which is convenient for subsequent comparison.

[0197] Among them, the image category probability value represents the probability that the image data contains illegal information, and the third target model can be constructed based on the NSFW algorithm.

[0198] S430: When the image category probability value is greater than a third threshold, mark the image data as having illegal image information.

[0199] The image category probability is compared with the third threshold value. If the image category probability is greater than the third threshold value, the image data is marked as having illegal image information.

[0200] The disclosed embodiment inputs the image data into the third target model, obtains the image category probability value and compares it with the third threshold value, which can effectively determine whether there is illegal image information in the image data.

[0201] In one embodiment, the data detection method further includes step S170.

[0202] S170: When it is detected that the number of target data of any one of illegal video information, illegal text information and illegal image information is greater than a preset number threshold, mark the user corresponding to the target data as an illegal user.

[0203] After marking the target data for any one of the illegal video information, illegal text information and illegal image information, the number of illegal target data is counted. If the number of illegal target data detected is greater than the preset number threshold, the user corresponding to the illegal target data is marked as an illegal user and corresponding processing is performed.

[0204] The disclosed embodiment detects the number of illegal target data and marks users whose number is greater than a certain threshold, thereby accurately identifying illegal users and facilitating the subsequent reduction of the spread of illegal data.

[0205] like Figure 6 As shown, the embodiment of the present disclosure further provides a data detection device 600, and the data detection device 600 includes:

[0206] The acquisition module 601 is used to acquire target data in the cloud disk, where the target data includes video data;

[0207] An input module 602 is used to input key frame data in the video data into an image semantic segmentation model, extract first feature information and semantic information in the key frame data using the image semantic segmentation model to obtain a semantic segmentation image, and classify the semantic segmentation image using a classification module in the image semantic segmentation model to obtain a person image set and a scene image set;

[0208] A calculation module 603 is used to calculate the feature vector of the corresponding person image set or scene image set based on the classifier in the first target model to obtain a target category probability value;

[0209] The marking module 604 is used to mark the video data as having illegal video information when the target category probability value is greater than the target threshold.

[0210] In some embodiments, before inputting the key frame data in the video data into the image semantic segmentation model, the data detection method further includes:

[0211] Encrypt the target data using an encryption algorithm to obtain encrypted data;

[0212] extracting key frame data from the video data when the encrypted data does not match the data in the illegal content database;

[0213] Extract key frame data from video data, including:

[0214] Performing a frame interception operation on the video data to obtain a plurality of image frame data;

[0215] Based on the corresponding image entropies of the plurality of image frame data, an image frame corresponding to an image entropy whose difference between different image entropies is greater than a preset threshold or an image entropy that is an extreme value is selected to obtain candidate image frame data;

[0216] A differential operation is performed on the candidate image frame data to extract key frame data from the candidate image frame data.

[0217] In some embodiments, extracting first feature information and semantic information from key frame data using an image semantic segmentation model to obtain a semantic segmentation image includes:

[0218] Using a feature extraction module in an image semantic segmentation model, extract features from the key frame data to obtain first feature information;

[0219] Using a semantic extraction module in an image semantic segmentation model, extracting semantic boundaries from the first feature information to obtain semantic information;

[0220] The first feature information and the semantic information are fused using a fusion module in the image semantic segmentation model to obtain a semantic segmentation image.

[0221] In some embodiments, the first target model includes multiple sub-models, and the sub-models are used to identify different types of violations; the type includes a first type, and the specific process of the sub-model corresponding to the first type identifying the first type of violation includes:

[0222] Using the feature extraction module in the sub-model corresponding to the first type, a feature extraction operation is performed on the person images in the person image set to obtain scale-invariant feature transform (SIFT) feature information;

[0223] Using a first preset algorithm, calculating optical flow feature vector information of a current frame and a next frame of a person image;

[0224] Calculate the image color feature information of the person image using a second preset algorithm;

[0225] The SIFT feature information, the optical flow feature vector information and the image color feature information are concatenated to obtain the second feature information;

[0226] The second feature information is converted into a feature vector.

[0227] In some embodiments, the first target model includes multiple sub-models, and the sub-models are used to identify different types of violations; the type includes the second type, and the sub-model corresponding to the second type identifies the second type of violation. The specific process includes:

[0228] Performing a downsampling operation on the scene images in the scene image set to obtain operation information;

[0229] Using the first convolution module in the sub-model corresponding to the second type to perform a convolution operation on the scene image to obtain convolution mapping information;

[0230] Fusing the operation information and the convolution mapping information to obtain fused information;

[0231] Using the second convolution module and the third convolution module in the sub-model corresponding to the second type to extract features from the fused information, obtaining third feature information;

[0232] The third feature information is converted into a feature vector.

[0233] In some embodiments, the target data further includes text data, and in the case where the encrypted data does not match the data in the illegal content database, the method further includes:

[0234] Using the second target model to perform a truncation operation on the text of the text data to obtain a truncation text;

[0235] Based on the data information in the preset database, calculate the similarity information of the intercepted text;

[0236] When the number of intercepted texts whose similarity information is greater than the first threshold is greater than the second threshold, it is marked that illegal text information exists in the text data.

[0237] In some embodiments, when the second target model is a sensitive word model, using the second target model to perform a text interception operation on the text data to obtain the intercepted text includes:

[0238] Use the sensitive word model to segment the text data to obtain a word sequence;

[0239] Based on the data information in the preset database, similarity information of the intercepted text is calculated; when the number of the intercepted texts whose similarity information is greater than the first threshold is greater than the second threshold, marking the text data as having illegal text information includes:

[0240] Based on the data information in the preset sensitive word database, calculate the sensitive similarity information of the words in the word sequence;

[0241] When the number of words whose sensitive similarity information is greater than the first threshold is greater than the second threshold, it is marked that illegal text information exists in the text data.

[0242] In some embodiments, when the second target model is a topic model, using the second target model to perform a truncation operation on text of the text data to obtain the truncation text includes:

[0243] Use the topic model to extract the text topics of the text data to obtain the text topics;

[0244] Based on the data information in the preset database, similarity information of the intercepted text is calculated; when the number of the intercepted texts whose similarity information is greater than the first threshold is greater than the second threshold, marking the text data as having illegal text information includes:

[0245] Based on the data information in the preset illegal topic database, calculating the illegal topic similarity information of the text topic;

[0246] When the number of text topics whose illegal topic similarity information is greater than the first threshold is greater than the second threshold, it is marked that illegal text information exists in the text data.

[0247] In some embodiments, the target data further includes image data, and in the case where the encrypted data does not match the data in the illegal content database, the method further includes:

[0248] inputting the image data into a third target model;

[0249] Using the classifier in the third target model to calculate the feature vector of the image data, to obtain the image category probability value;

[0250] When the image category probability value is greater than the third threshold, the image data is marked as having illegal image information.

[0251] In some embodiments, the method further comprises:

[0252] When it is detected that the number of target data of any one of illegal video information, illegal text information and illegal image information is greater than a preset number threshold, the user corresponding to the target data is marked as an illegal user.

[0253] Figure 6 Each module / unit in the device shown has the function of realizing Figures 1 to 5 The functions of each step executed by the application system in the present invention can achieve the corresponding technical effects. For the sake of brevity, they will not be repeated here.

[0254] Figure 7 A schematic diagram of the hardware structure of data detection provided by an embodiment of the present disclosure is shown.

[0255] The data detection device may include a processor 701 and a memory 702 storing computer program instructions.

[0256] Specifically, the processor 701 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0257] The memory 702 may include a large capacity memory for data or instructions. By way of example and not limitation, the memory 702 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive or a combination of two or more of these. In appropriate cases, the memory 702 may include a removable or non-removable (or fixed) medium. In appropriate cases, the memory 702 may be inside or outside the integrated gateway disaster recovery device. In a specific embodiment, the memory 702 is a non-volatile solid-state memory.

[0258] The memory 702 may include a read-only memory (ROM), a random access memory (RAM), a magnetic disk storage medium device, an optical storage medium device, a flash memory device, an electrical, optical or other physical / tangible memory storage device. Therefore, generally, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method according to one aspect of the present disclosure.

[0259] The processor 701 implements any one of the data detection methods in the above embodiments by reading and executing computer program instructions stored in the memory 702 .

[0260] In one example, the data detection device may further include a communication interface 703 and a bus 704. Figure 7 As shown, the processor 701, the memory 702, and the communication interface 703 are connected via a bus 704 and communicate with each other.

[0261] The communication interface 703 is mainly used to implement communication between various modules, devices, units and / or equipment in the embodiments of the present application.

[0262] Bus 704 includes hardware, software or both, and couples the components of online data flow billing equipment to each other. For example, but not limitation, the bus may include Accelerated Graphics Port (AGP) or other graphics bus, Enhanced Industry Standard Architecture (EISA) bus, Front Side Bus (FSB), Hyper Transport (HT) interconnection, Industry Standard Architecture (ISA) bus, InfiniBand interconnection, Low Pin Count (LPC) bus, Memory bus, Micro Channel Architecture (MCA) bus, Peripheral Component Interconnect (PCI) bus, PCI-Express (PCI-X) bus, Serial Advanced Technology Attachment (SATA) bus, Video Electronics Standards Association Local (VLB) bus or other suitable bus or two or more of these combinations. In appropriate cases, bus 704 may include one or more buses. Although the embodiment of the present invention describes and illustrates a specific bus, the present invention considers any suitable bus or interconnection. In addition, in conjunction with the method for data detection in the above-mentioned embodiment, the present application embodiment also provides a computer storage medium to realize. The computer storage medium stores computer program instructions; when the computer program instructions are executed by the processor, any one of the data detection methods in the above embodiments is implemented.

[0263] An embodiment of the present application also provides a computer program product, including a computer program, which implements any one of the data detection methods in the above embodiments when the computer program is executed by a processor.

[0264] It should be clear that the present invention is not limited to the specific configuration and processing described above and shown in the figures. For the sake of simplicity, a detailed description of the known method is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps after understanding the spirit of the present invention.

[0265] The functional blocks shown in the above structural block diagram can be implemented as hardware, software, firmware or a combination thereof. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), appropriate firmware, plug-in, function card, etc. When implemented in software, the elements of the present invention are programs or code segments used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or communication link by a data signal carried in a carrier. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, read-only memory (Read-Only Memory, ROM), flash memory, erasable read-only memory (Erasable ReadOnly Memory, EROM), floppy disks, compact disc read-only memory (Compact Disc Read-Only Memory, CD-ROM), optical discs, hard disks, optical fiber media, radio frequency (Radio Frequency, RF) links, etc. The code segment can be downloaded via a computer network such as the Internet, an intranet, etc.

[0266] It should also be noted that the exemplary embodiments mentioned in the present invention describe some methods or systems based on a series of steps or devices. However, the present invention is not limited to the order of the above steps, that is, the steps can be performed in the order mentioned in the embodiments, or in a different order from the embodiments, or several steps can be performed simultaneously.

[0267] Aspects of the present disclosure are described above with reference to the flowchart and / or block diagram of the method, device (system) and computer program product according to the embodiment of the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of each box in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device to produce a machine so that these instructions executed by the processor of the computer or other programmable data processing device enable the implementation of the function / action specified in one or more boxes of the flowchart and / or block diagram. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field programmable logic circuit. It can also be understood that each box in the block diagram and / or flowchart and the combination of boxes in the block diagram and / or flowchart can also be implemented by dedicated hardware that performs a specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions.

[0268] The above is only a specific implementation of the present invention. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the system, module and unit described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here. It should be understood that the protection scope of the present invention is not limited to this. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be covered within the protection scope of the present invention.

Claims

1. A method for data detection, characterized in that: include: Obtain target data in the cloud disk, the target data includes video data; Inputting the key frame data in the video data into an image semantic segmentation model, extracting the first feature information and semantic information in the key frame data using the image semantic segmentation model to obtain a semantic segmentation image, and classifying the semantic segmentation image using a classification module in the image semantic segmentation model to obtain a person image set and a scene image set; Calculating the feature vector of the corresponding person image set or scene image set based on the classifier in the first target model to obtain a target category probability value; When the target category probability value is greater than the target threshold, the video data is marked as having illegal video information.

2. The method according to claim 1, characterized in that Before inputting the key frame data in the video data into the image semantic segmentation model, the method further includes: Encrypting the target data using an encryption algorithm to obtain encrypted data; extracting key frame data from the video data when the encrypted data does not match the data in the illegal content database; The step of extracting key frame data from the video data comprises: Performing a frame interception operation on the video data to obtain a plurality of image frame data; Based on the corresponding image entropies of the plurality of image frame data, selecting image frames corresponding to image entropies whose difference between different image entropies is greater than a preset threshold or image entropies that are extreme values, to obtain candidate image frame data; A differential operation is performed on the candidate image frame data to extract key frame data from the candidate image frame data.

3. The method according to claim 1, characterized in that The step of extracting the first feature information and the semantic information in the key frame data by using the image semantic segmentation model to obtain a semantic segmentation image includes: Using a feature extraction module in the image semantic segmentation model, extract features from the key frame data to obtain the first feature information; Using a semantic extraction module in the image semantic segmentation model, extracting semantic boundaries from the first feature information to obtain the semantic information; The first feature information and the semantic information are fused using a fusion module in the image semantic segmentation model to obtain the semantic segmentation image.

4. The method according to claim 1, characterized in that The first target model includes a plurality of sub-models, and the sub-models are used to identify different types of violations; the types include a first type, and the specific process of the sub-model corresponding to the first type identifying the first type of violation includes: Using the feature extraction module in the sub-model corresponding to the first type, a feature extraction operation is performed on the person images in the person image set to obtain scale-invariant feature transform (SIFT) feature information; Calculating optical flow feature vector information of a current frame and a next frame of the person image using a first preset algorithm; Calculating the image color feature information of the character image by using a second preset algorithm; The SIFT feature information, the optical flow feature vector information and the image color feature information are concatenated to obtain second feature information; The second feature information is converted to obtain the feature vector.

5. The method according to claim 1, characterized in that The first target model includes a plurality of sub-models, and the sub-models are used to identify different types of violations; the type includes a second type, and the sub-model corresponding to the second type identifies the specific process of the second type of violation, including: Performing a downsampling operation on the scene images in the scene image set to obtain operation information; Using the first convolution module in the sub-model corresponding to the second type to perform a convolution operation on the scene image to obtain convolution mapping information; Fusing the operation information and the convolution mapping information to obtain fused information; Using the second convolution module and the third convolution module in the sub-model corresponding to the second type to perform feature extraction on the fusion information to obtain third feature information; The third feature information is converted to obtain the feature vector.

6. The method according to claim 2, characterized in that The target data also includes text data, and in the case where the encrypted data does not match the data in the illegal content database, the method further includes: Using the second target model to perform a truncation operation on the text of the text data to obtain a truncation text; Based on the data information in the preset database, calculating the similarity information of the intercepted text; When the number of the intercepted texts whose similarity information is greater than the first threshold is greater than the second threshold, it is marked that the text data contains illegal text information.

7. The method according to claim 6, characterized in that In the case where the second target model is a sensitive word model, the using the second target model to perform a interception operation on the text of the text data to obtain the intercepted text includes: Using the sensitive word model to perform word segmentation on the text of the text data to obtain a word sequence; The method of calculating the similarity information of the intercepted text based on the data information in the preset database; and marking the existence of illegal text information in the text data when the number of the intercepted texts whose similarity information is greater than the first threshold is greater than the second threshold, includes: Based on the data information in the preset sensitive word database, calculating the sensitive similarity information of the words in the word sequence; When the number of words whose sensitive similarity information is greater than the first threshold is greater than the second threshold, it is marked that the text data contains the illegal text information.

8. The method according to claim 6, characterized in that In the case where the second target model is a topic model, the using the second target model to perform a interception operation on the text of the text data to obtain the intercepted text includes: Using the topic model to perform text topic extraction processing on the text of the text data to obtain text topics; The method of calculating the similarity information of the intercepted text based on the data information in the preset database; and marking the existence of illegal text information in the text data when the number of the intercepted texts whose similarity information is greater than the first threshold is greater than the second threshold, includes: Based on the data information in a preset illegal topic database, calculating the illegal topic similarity information of the text topic; When the number of text topics whose illegal topic similarity information is greater than the first threshold is greater than the second threshold, it is marked that illegal text information exists in the text data.

9. The method according to claim 2, characterized in that: The target data also includes image data, and in the case where the encrypted data does not match the data in the illegal content database, the method further includes: inputting the image data into a third object model; Calculating the feature vector of the image data using the classifier in the third target model to obtain an image category probability value; When the image category probability value is greater than a third threshold, the image data is marked as having illegal image information.

10. The method according to any one of claims 1 to 9, characterized in that: The method further comprises: When it is detected that the number of target data of any one of the illegal video information, illegal text information and illegal image information is greater than a preset number threshold, the user corresponding to the target data is marked as an illegal user.

Images