Content detection method and device based on artificial intelligence and computer equipment

Abstract

The invention provides a content detection method and device based on artificial intelligence and computer equipment. The method comprises the following steps: receiving input content at the current moment through edge equipment; detecting the content to be detected through the time sequence and the model; judging whether the detection result is interested content or not; if yes, comparing the input content at the current moment with the input content at the previous moment through a similarity comparison algorithm to obtain a similarity value; judging whether the similarity value is greater than a preset threshold value or not; if yes, marking the content as interested content, and outputting a detection result at the current moment; and detecting the input content at the next moment. According to the method, the time continuity of the input content is utilized, the defects of the model are complemented, the detection mechanism is optimized, and the reliability of content detection is improved. Through the relevance between the contents, the false detection and omission ratio of rapid change of the input contents is reduced. The calculation overhead is reduced on the edge device, the utilization rate of the calculation power of the edge device is improved, and the high-efficiency and accurate real-time content detection requirement is met.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence, and particularly to a content detection method, device and computer device based on artificial intelligence. Background Art

[0002] Nowadays, the application scenarios of artificial intelligence technology are more extensive, and the edge side is a typical application scenario, that is, target detection and recognition are carried out on edge devices, such as surveillance cameras, video live broadcasts, etc. However, when edge devices are applied, they are more complex than the server side. How to quickly adapt to various application scenarios and accurately identify and detect is an important research topic at present.

[0003] As a real-time perception system, edge devices are the main means for artificial intelligence to contact the real world, which also means that edge devices need to have strong adaptability to contact complex and changeable actual application scenarios. However, existing edge devices are still limited by their own computing power and storage capacity.

[0004] On the other hand, the traditional method for edge devices to detect and recognize targets is to first annotate the target content and construct training data, then train an artificial intelligence model, and use the model to perform inference on content detection and recognition. Finally, the detection and recognition ability of the model is improved by increasing the scale of training data. However, this method has low recognition accuracy, cannot adapt to rapidly changing application scenarios, recognize hidden image information, etc., and has problems such as slow effect, long time consumption, and high labor cost, and cannot meet the requirements of high-efficiency and accurate real-time content detection, and is prone to false detection and missed detection. Summary of the Invention

[0005] The present invention aims to solve the problems in the above-mentioned prior art that the existing target detection and recognition method has low recognition accuracy, cannot quickly adapt to multiple application scenarios, and has long time consumption and high labor cost, and provides a content detection method, device and computer device based on artificial intelligence.

[0006] The present invention provides a content detection method based on artificial intelligence, including the following steps:

[0007] Receiving input content at the current moment through a preset edge device to obtain content to be detected; wherein, the content to be detected includes but is not limited to video, picture, text and audio;

[0008] Detecting the content to be detected through a preset time series and model to obtain a detection result;

[0009] Judging whether the detection result is content of interest;

[0010] If so, compare the input content at the current moment with the input content at the previous moment through a preset similarity comparison algorithm to obtain a similarity value;

[0011] Determine whether the similarity value is greater than a preset threshold;

[0012] If so, mark it as interesting content and output the detection result at the current moment;

[0013] Detect the input content at the next moment.

[0014] Further, before the step of detecting the content to be detected through a preset time series and model to obtain a detection result, it includes:

[0015] Optimize the model to obtain an optimized model; wherein, the optimization methods include but are not limited to compressing the number of parameters, simplifying the structure, and adjusting the training strategy;

[0016] Deploy the optimized model on the edge device.

[0017] Further, in the step of detecting the content to be detected through a preset time series and model to obtain a detection result, it includes:

[0018] Detect the content to be detected simultaneously through multiple preset processors.

[0019] Further, in the step of detecting the content to be detected simultaneously through multiple preset processors, it includes:

[0020] Identify the overall information of the content to be detected; wherein, the overall information includes interesting content and uninteresting content.

[0021] Further, in the step of if so, compare the input content at the current moment with the input content at the previous moment through a preset similarity comparison algorithm to obtain a similarity value, it includes:

[0022] Extract the intermediate representation features of the overall information and grayscale the intermediate representation features;

[0023] Generate a code according to the intermediate representation features;

[0024] Perform an exclusive OR operation on each bit of the code to quantify the overall information at the current moment and the overall information at the previous moment, and subtract to obtain the similarity value.

[0025] Further, in the step of extracting the intermediate representation features of the overall information and grayscaling the intermediate representation features, it includes:

[0026] Extract the low-frequency components of the overall information to obtain a set of low-frequency components.

[0027] Further, after the step of extracting the low-frequency components of the overall information to obtain a set of low-frequency components, it includes:

[0028] Calculate the mean value of the low-frequency components;

[0029] Compare each element of the set of low-frequency components with the mean value respectively and generate a binary code.

[0030] Further, in the step of determining whether the similarity value is greater than a preset threshold, it includes:

[0031] If not, mark it as uninteresting content, and calculate the similarity value between the input content at the current moment and the input content at the next moment through the similarity comparison algorithm;

[0032] Determine whether the similarity value is greater than the threshold;

[0033] If so, no detection is performed;

[0034] If not, detect again through the model.

[0035] The present invention also provides a content detection device based on artificial intelligence, including:

[0036] A receiving module, configured to receive the input content at the current moment through a preset edge device;

[0037] A first detection module, detecting the content to be detected through a preset time series and model to obtain a detection result;

[0038] A first judgment module, configured to judge whether the detection result is interesting content;

[0039] A comparison module, configured to, if so, compare the input content at the current moment with the input content at the previous moment through a preset similarity comparison algorithm to obtain a similarity value;

[0040] A second judgment module, configured to judge whether the similarity value is greater than a preset threshold;

[0041] A marking module, configured to, if so, mark it as interesting content and output the detection result at the current moment;

[0042] A second detection module, configured to detect the input content at the next moment.

[0043] The present invention also provides a computer device, including a memory and a processor, where the memory stores a computer program, and the processor executes the computer program to implement the steps in any one of the above methods.

[0044] The present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of any one of the above methods are implemented.

[0045] The present invention provides a content detection method, device and computer device based on artificial intelligence, having the following beneficial effects:

[0046] The present invention utilizes the characteristics of temporal continuity and similarity of the input content, effectively compensates for the deficiencies of the model, and meets the high-real-time application requirements. Optimize the detection mechanism to improve the effectiveness and reliability of content detection. Through the relevance between contents, reduce the false detection and missed detection rates of rapidly changing contents. It can reduce the computational overhead of content inference by the model on edge devices, improve the utilization rate of edge device computing power, meet the requirements of high-efficiency and accurate real-time content detection, and avoid false detection or missed detection caused by insufficient model capabilities.

[0047] The present invention uses an optimized model to detect temporal input content on edge devices, which can make full use of the computing power of edge devices while reducing unnecessary overhead, thereby improving the real-time requirements of detection and ensuring the high accuracy of overall input content detection. Based on the detection results of the input content at the previous moment, and using a similarity comparison algorithm to discriminate the input content of the front and rear time series, so as to avoid the detection of approximate content, which can not only improve the real-time detection performance, but also improve the detection accuracy. The present invention can detect interesting information in real time according to the temporal relationship of the input content. By comparing the similarity of the overall information of the previous moment and the next moment, it reduces the computational overhead and storage access overhead of the edge device for identifying interesting content. It can replace the traditional object detection algorithm with a similarity comparison algorithm, and can dynamically switch the detection algorithm to avoid the large computational overhead caused by a single algorithm, thereby meeting the real-time requirements of different application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a schematic diagram of the method steps of a content detection method based on artificial intelligence in the present invention;

[0049] Figure 2 It is a block diagram of the device structure of a content detection device based on artificial intelligence in the present invention;

[0050] Figure 3 It is a block diagram of the structure of a computer device of the present invention;

[0051] Figure 4 It is a schematic diagram of the steps of an embodiment of a content detection method based on artificial intelligence in the present invention;

[0052] Figure 5Flow chart of detection in a content detection method based on artificial intelligence according to the present invention;

[0053] Figure 6 Algorithm flow chart of similarity comparison algorithm in a content detection method based on artificial intelligence according to the present invention.

[0054] Marking description: receiving module 10, first detection module 20, first judgment module 30, comparison module 40, second judgment module 50, marking module 60, second detection module 70. Specific implementation manner

[0055] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0056] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0057] Existing detection methods cannot accurately detect the content at each moment, while the present application can discriminate according to the type of input content and classify and process it, which can reduce unnecessary computational overhead, optimize the detection process by skipping approximate content in the process, so as to adapt to various real-time application scenarios and can effectively identify interesting content and uninteresting content.

[0058] Refer to the attached Figure 1 , a content detection method based on artificial intelligence in an embodiment of the present invention, includes:

[0059] S1. Receive the input content at the current moment through a preset edge device to obtain the content to be detected; wherein, the content to be detected includes but is not limited to video, picture, text, and audio;

[0060] S2. Detect the content to be detected through a preset time series and model to obtain a detection result;

[0061] S3. Judge whether the detection result is interesting content;

[0062] S4. If so, compare the input content at the current moment with the input content at the previous moment through a preset similarity comparison algorithm to obtain a similarity value;

[0063] S5. Judge whether the similarity value is greater than a preset threshold;

[0064] S6, if so, mark it as the content of interest and output the detection result at the current moment;

[0065] S7, detect the input content at the next moment.

[0066] In the above steps, first, the input content at the current moment is received through a preset edge device to obtain the content to be detected; the content to be detected includes but is not limited to video, picture, text, and audio; then, the content to be detected is detected through a preset time series and model to obtain the detection result. In a specific embodiment, the time series is in ascending order. The present application detects the content to be detected in the manner of the time series, which can avoid the situations of low real-time rate, missed detection, or false detection of the content. Then, it is judged whether the detection result is the content of interest. If so, the input content at the current moment is compared with the input content at the previous moment through a preset similarity comparison algorithm to obtain a similarity value. Then, it is judged whether the similarity value is greater than a preset threshold. If so, it is marked as the content of interest, and the detection result at the current moment is output. Then, the input content at the next moment is detected. In a specific embodiment, the previous moment, the current moment, and the next moment are consecutive moments. In another specific embodiment, if the received current moment is the 2nd second, the previous moment is the 1st second, and the next moment is the 3rd second. Based on the detection result of the input content at the previous moment, and by using the similarity comparison algorithm to discriminate the input content in the front and back time series, the present application avoids detecting approximate content and skips the detection operation of approximate content in the process, which can not only avoid unnecessary calculation and storage overhead, but also improve the detection accuracy.

[0067] Specifically, as Figure 4 shown, an optimized model is deployed on the edge device. After receiving the input content, the input content is detected through the model and multiple processors. Since the input content has timeliness, the similarity degree of the input content in different time series is discriminated through the similarity comparison algorithm, and finally it is divided into the content of interest and the content of no interest.

[0068] In an embodiment, before the step of detecting the content to be detected through a preset time series and model to obtain the detection result, it includes:

[0069] Optimize the model to obtain the optimized model; wherein, the optimization methods include but are not limited to compressing the number of parameters, simplifying the structure, and adjusting the training strategy;

[0070] Deploy the optimized model on the edge device.

[0071] In this embodiment, before detecting the content to be detected through the model, the model is optimized. The optimization methods include reducing the number of parameters, simplifying the structure, adjusting the training strategy, and improving the inference speed. It can reduce the computing and storage requirements while improving the model accuracy, and enhance the robustness and adaptability of the model. In a specific embodiment, the model includes an object detection model or an image classification model. In another specific embodiment, the model is optimized, such as the object detection model after quantization.

[0072] In one embodiment, in the step of obtaining the detection result by detecting the content to be detected through a preset time series and the model, it includes:

[0073] Detecting the content to be detected simultaneously through multiple preset processors.

[0074] In this embodiment, detecting the content to be detected simultaneously through multiple processors can reduce the computing burden on the edge device.

[0075] In one embodiment, in the step of detecting the content to be detected simultaneously through multiple preset processors, it includes:

[0076] Identifying the overall information of the content to be detected; wherein, the overall information includes the content of interest and the content not of interest.

[0077] In this embodiment, detecting the content to be detected simultaneously through multiple processors, wherein the processor can identify the overall information of the content to be detected, and the overall information includes the content of interest and the content not of interest.

[0078] In one embodiment, if so, in the step of comparing the input content at the current moment with the input content at the previous moment through a preset similarity comparison algorithm to obtain a similarity value, it includes:

[0079] Extracting the intermediate representation features of the overall information and graying the intermediate representation features;

[0080] Generating a code according to the intermediate representation features;

[0081] Performing an exclusive OR operation on each bit of the code to quantify the overall information at the current moment and the overall information at the previous moment, and subtracting to obtain a similarity value.

[0082] In this embodiment, when comparing the input content at the current moment with that at the previous moment, an intermediate representation information extraction method is adopted to extract the intermediate representation features in the overall information, and the correlation between the sequential contents is calculated using the representation features. The specific process is as follows: after extracting the intermediate representation features of the overall information, the intermediate representation features are grayscaled, and then codes are generated according to the intermediate representation features. Bitwise exclusive OR operations are performed on the codes to quantify the overall information at the current moment and the overall and information at the previous moment, and the overall information at the current moment is subtracted from the overall information at the previous moment to obtain a similarity value.

[0083] Specifically, as Figure 6 shown, based on the detection result of the input content at time T1, a similarity comparison algorithm is used to determine whether the input content at time T2 is the content of interest. If so, the similarity comparison algorithm is enabled to detect the continuous input content, and the input content greater than the threshold is marked as the content of interest, and the input content less than the threshold is marked as the content not of interest, and the input content at the next moment is calculated through the similarity comparison algorithm. If it is still not similar to the input content at the next moment, the detection continues. It can avoid detecting similar contents, significantly improve the detection efficiency, and enhance the accuracy of input content detection.

[0084] In a specific embodiment, the similarity comparison algorithm is the perceptual hashing algorithm. Specifically, two images are compared, the hash values of the two images are calculated, and the Hamming distance between the two images is calculated, and then it is determined whether the two images are similar. Among them, the Hamming distance refers to the number of different characters at the corresponding positions between two equal-length strings and is used to quantify the similarity of the two images. The greater the Hamming distance, the smaller the similarity of the images; the smaller the Hamming distance, the greater the similarity of the images.

[0085] In an embodiment, in the steps of extracting the intermediate representation features of the overall information and grayscaling the intermediate representation features, it includes:

[0086] Extract the low-frequency components of the overall information to obtain a low-frequency component set.

[0087] After the step of extracting the low-frequency components of the overall information to obtain a low-frequency component set, it includes:

[0088] Calculate the mean value of the low-frequency components;

[0089] Compare each element of the low-frequency component set with the mean value respectively and generate a binary code.

[0090] In this embodiment, first, the low-frequency components of the overall information are extracted through a preset discrete cosine algorithm to obtain a low-frequency component set; then the mean value of the low-frequency components is calculated, and then the elements of the low-frequency component set are compared with the mean value respectively to generate a binary code.

[0091] Specifically, the similarity comparison algorithm compresses high-dimensional visual information into a 64-bit code by extracting the visual features of an image, and quantifies the difference between two images through the Hamming distance.

[0092] More specifically, as Figure 5 shown, after extracting the intermediate representation features, first, the input image is grayscaled to simplify the multi-channel image into a single-channel grayscale image, thereby reducing the computational complexity and retaining the main visual features. Then the grayscale image is scaled to a fixed size to further reduce the data dimension, serving as the basis for subsequent frequency-domain transformation. In the feature transformation stage, the low-frequency components of the image are extracted through the discrete cosine transform to obtain a set of low-frequency components, which represent the overall visual features of the image. Among them, the discrete cosine transform is DCT, Discrete Cosine Transform, a mathematical operation related to the Fourier transform. In the Fourier series expansion, if the function to be expanded is a real even function, then only cosine terms are included in its Fourier series, and its discretization can lead to the cosine transform. The Fourier transform is a mathematical tool that represents a function satisfying certain conditions as a linear combination of trigonometric functions or their integrals. The Fourier series is the predecessor of the Fourier transform and can decompose a periodic function into a sum of a series of sine and cosine functions. Then, the mean value of the low-frequency components is calculated, and each element of the set of low-frequency components is compared with the mean value to generate a 64-bit binary code, which has the advantages of efficient storage and fast calculation. Finally, by performing an exclusive OR operation bit by bit on the codes of the two images, the Hamming distance is counted to quantify the difference between the overall information at the previous moment and the overall information at the current moment.

[0093] In one embodiment, in the step of determining whether the similarity value is greater than a preset threshold, it includes:

[0094] If not, it is marked as uninteresting content, and the similarity value between the input content at the current moment and the input content at the next moment is calculated through the similarity comparison algorithm;

[0095] Determine whether the similarity value is greater than the threshold;

[0096] If so, no detection is performed;

[0097] If not, the detection is performed again through the model.

[0098] In this embodiment, if the similarity value between the input content at the current moment and the input content at the previous moment is less than the preset threshold, it is marked as uninteresting content, and the similarity value between the input content at the current moment and the input content at the next moment is calculated through the similarity comparison algorithm. Then, it is determined whether the similarity value is greater than the threshold. If so, no detection is performed. If not, the detection is performed again through the model.

[0099] In summary, in specific implementation, after the edge device receives the input content at the current moment, it first performs detection through the provided model. If the detection result is the content of interest, it calculates the similarity value between the input content at the current moment and the input content at the previous moment through the similarity comparison algorithm. If it is greater than the threshold, it indicates that the detection results of the input content at the current moment and the input content at the previous moment are the same; if it is less than the threshold, there is no correlation between the input content at the current moment and the input content at the previous moment. At this time, the model is called to continue the detection. In the similarity comparison algorithm, the intermediate representation features at the current moment and the previous moment are extracted, then the intermediate representation features are grayscaled and transformed, and finally the input content is encoded into a 64-bit binary value, and the Hamming distance is used to calculate the similarity value between the input content at the current moment and the input content at the previous moment. If they are similar, there is no need to perform detection, thus reducing the computational overhead.

[0100] Reference appendix Figure 2 , an artificial intelligence-based content detection device, comprising:

[0101] A receiving module 10, configured to receive the input content at the current moment through a preset edge device to obtain the content to be detected;

[0102] A first detection module 20, configured to detect the content to be detected through a preset time series and model to obtain a detection result;

[0103] A first judgment module 30, configured to judge whether the detection result is the content of interest;

[0104] A comparison module 40, configured to, if so, compare the input content at the current moment with the input content at the previous moment through a preset similarity comparison algorithm to obtain a similarity value;

[0105] A second judgment module 50, configured to judge whether the similarity value is greater than a preset threshold;

[0106] A marking module 60, configured to, if so, mark it as the content of interest and output the detection result at the current moment;

[0107] A second detection module 70, configured to detect the input content at the next moment;

[0108] An optimization module, configured to optimize the model to obtain an optimized model;

[0109] A detection unit, configured to simultaneously detect the content to be detected through multiple preset processors;

[0110] An identification subunit, configured to identify the overall information of the content to be detected;

[0111] An extraction unit, configured to extract the intermediate representation features of the overall information and grayscale the intermediate representation features;

[0112] A generating unit, configured to generate a code according to intermediate representation features;

[0113] A calculating unit, configured to perform an exclusive OR operation on each bit of the code to quantify the overall information at the current moment and the overall information at the previous moment, and subtract them to obtain a similarity value;

[0114] An extracting subunit, configured to extract low-frequency components of the overall information to obtain a set of low-frequency components;

[0115] A calculating subunit, configured to calculate the mean value of the low-frequency components;

[0116] A comparing subunit, configured to compare each element of the set of low-frequency components with the mean value respectively and generate a binary code;

[0117] A marking unit, configured to, if not, mark it as content of no interest, and calculate the similarity value between the input content at the current moment and the input content at the next moment through a similarity comparison algorithm;

[0118] A judging unit, configured to judge whether the similarity value is greater than a threshold;

[0119] An activating unit, configured to, if not, detect again through the model.

[0120] In this embodiment, first, the receiving module 10 receives the input content at the current moment through a preset edge device to obtain the content to be detected. In addition, the receiving module 10 can receive the input content at the previous moment and the next moment. Then, the optimization module optimizes the model to obtain an optimized model. Next, the first detection module 20 detects the content to be detected through a preset time series and the model to obtain a detection result. During this period, the detection unit simultaneously detects the content to be detected through multiple processors, that is, the recognition subunit recognizes the overall information of the content to be detected. Then, the first judgment module 30 judges whether the detection result is the content of interest. If so, the comparison module 40 compares the input content at the current moment with the input content at the previous moment through a preset similarity comparison algorithm to obtain a similarity value. Specifically, the extraction unit extracts the intermediate representation features of the overall information and grayscales the intermediate representation features. Then, the generation unit generates a code according to the intermediate representation features. Next, the calculation unit performs an exclusive OR operation on each bit of the code to quantify the overall information at the current moment and the overall information at the previous moment, and subtracts them to obtain a similarity value. More specifically, the extraction subunit extracts the low-frequency components of the overall information to obtain a set of low-frequency components. Then, the calculation subunit calculates the mean value of the low-frequency components. Finally, the comparison subunit compares each element of the set of low-frequency components with the mean value respectively and generates a binary code. Then, the second judgment module 50 judges whether the similarity value is greater than a preset threshold. If so, the marking module 60 marks it as the content of interest and outputs the detection result at the current moment. If not, the marking unit marks it as the content of no interest and calculates the similarity value between the input content at the current moment and the input content at the next moment through the similarity comparison algorithm. Then, the judgment unit judges whether the similarity value is greater than the threshold. When the similarity value between the input content at the current moment and the input content at the next moment is less than the preset threshold, the activation unit activates the second detection module 70 to detect the input content at the next moment.

[0121] See the attached Figure 3 , in the embodiment of the present application, a computer device is further provided. The computer device may be a server, and its internal structure may be as Figure 3As shown. The computer device includes a processor, a memory, a network interface, and a database. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system, computer program, and database in the non-volatile storage medium. The database of the computer device is used to store data such as templates, tables, and preset fields. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements a content detection method based on artificial intelligence, including the following steps:

[0122] Receiving the input content at the current moment through a preset edge device to obtain the content to be detected; among them, the content to be detected includes but is not limited to video, picture, text, and audio;

[0123] Detecting the content to be detected through a preset time series and model to obtain a detection result;

[0124] Judging whether the detection result is content of interest;

[0125] If so, comparing the input content at the current moment with the input content at the previous moment through a preset similarity comparison algorithm to obtain a similarity value;

[0126] Judging whether the similarity value is greater than a preset threshold;

[0127] If so, marking it as content of interest and outputting the detection result at the current moment;

[0128] Detecting the input content at the next moment.

[0129] In one embodiment, before the step of detecting the content to be detected through a preset time series and model to obtain a detection result, it includes:

[0130] Optimizing the model to obtain an optimized model; among them, the optimization methods include but are not limited to compressing the number of parameters, simplifying the structure, and adjusting the training strategy;

[0131] Deploying the optimized model on the edge device.

[0132] In one embodiment, in the step of detecting the content to be detected through a preset time series and model to obtain a detection result, it includes:

[0133] Detecting the content to be detected simultaneously by multiple preset processors.

[0134] In one embodiment, in the step of detecting the content to be detected simultaneously by multiple preset processors, it includes:

[0135] Identify the overall information of the content to be detected; among them, the overall information includes the content of interest and the content of no interest.

[0136] In one embodiment, if so, in the step of comparing the input content at the current moment with the input content at the previous moment through a preset similarity comparison algorithm to obtain a similarity value, it includes:

[0137] Extract the intermediate representation features of the overall information and grayscale the intermediate representation features;

[0138] Generate a code according to the intermediate representation features;

[0139] Perform an exclusive OR operation on each bit of the code to quantify the overall information at the current moment and the overall information at the previous moment, and subtract to obtain a similarity value.

[0140] In one embodiment, in the step of extracting the intermediate representation features of the overall information and grayscaling the intermediate representation features, it includes:

[0141] Extract the low-frequency components of the overall information to obtain a set of low-frequency components.

[0142] In one embodiment, after the step of extracting the low-frequency components of the overall information to obtain a set of low-frequency components, it includes:

[0143] Calculate the mean value of the low-frequency components;

[0144] Compare each element of the set of low-frequency components with the mean value respectively and generate a binary code.

[0145] In one embodiment, in the step of determining whether the similarity value is greater than a preset threshold, it includes:

[0146] If not, mark it as the content of no interest, and calculate the similarity value between the input content at the current moment and the input content at the next moment through the similarity comparison algorithm;

[0147] Determine whether the similarity value is greater than the threshold;

[0148] If so, do not perform detection;

[0149] If not, detect again through the model.

[0150] Those skilled in the art can understand that Figure 3 The structure shown in is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied.

[0151] An embodiment of the present application further provides a computer storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements a content detection method based on artificial intelligence, including the following steps:

[0152] Receiving the input content at the current moment through a preset edge device to obtain the content to be detected; wherein, the content to be detected includes but is not limited to video, picture, text, and audio;

[0153] Detecting the content to be detected through a preset time series and model to obtain a detection result;

[0154] Judging whether the detection result is content of interest;

[0155] If so, comparing the input content at the current moment with the input content at the previous moment through a preset similarity comparison algorithm to obtain a similarity value;

[0156] Judging whether the similarity value is greater than a preset threshold;

[0157] If so, marking it as content of interest and outputting the detection result at the current moment;

[0158] Detecting the input content at the next moment.

[0159] In one embodiment, before the step of detecting the content to be detected through a preset time series and model to obtain a detection result, it includes:

[0160] Optimizing the model to obtain an optimized model; wherein, the optimization methods include but are not limited to compressing the number of parameters, simplifying the structure, and adjusting the training strategy;

[0161] Deploying the optimized model on the edge device.

[0162] In one embodiment, in the step of detecting the content to be detected through a preset time series and model to obtain a detection result, it includes:

[0163] Detecting the content to be detected simultaneously through multiple preset processors.

[0164] In one embodiment, in the step of detecting the content to be detected simultaneously through multiple preset processors, it includes:

[0165] Identifying the overall information of the content to be detected; wherein, the overall information includes content of interest and content not of interest.

[0166] In one embodiment, if so, in the step of comparing the input content at the current moment with the input content at the previous moment through a preset similarity comparison algorithm to obtain a similarity value, it includes:

[0167] Extracting the intermediate representation features of the overall information and graying the intermediate representation features;

[0168] Generate codes based on intermediate representation features;

[0169] Perform an exclusive OR operation on each bit of the codes to quantify the overall information at the current moment and the overall information at the previous moment, and subtract them to obtain a similarity value.

[0170] In one embodiment, in the step of extracting the intermediate representation features of the overall information and graying the intermediate representation features, it includes:

[0171] Extract the low-frequency components of the overall information to obtain a set of low-frequency components.

[0172] In one embodiment, after the step of extracting the low-frequency components of the overall information to obtain a set of low-frequency components, it includes:

[0173] Calculate the mean value of the low-frequency components;

[0174] Compare each element of the set of low-frequency components with the mean value respectively and generate binary codes.

[0175] In one embodiment, in the step of determining whether the similarity value is greater than a preset threshold, it includes:

[0176] If not, mark it as uninteresting content, and calculate the similarity value between the input content at the current moment and the input content at the next moment through a similarity comparison algorithm;

[0177] Determine whether the similarity value is greater than the threshold;

[0178] If so, no detection is performed;

[0179] If not, detect again through the model.

[0180] In summary, the present application provides a content detection method, device and computer device based on artificial intelligence. The present invention uses an optimized model to detect sequential input content on an edge device, which can reduce unnecessary overhead while making full use of the computing power of the edge device, thereby improving the real-time requirements of detection and ensuring the high accuracy of the overall input content detection. Based on the detection results of the input content at the previous moment, and using a similarity comparison algorithm to discriminate the input content before and after in time sequence, so as to avoid detecting approximate content, which can not only improve the real-time detection performance, but also improve the detection accuracy. The present invention can detect interesting information in real time according to the time sequence relationship of the input content, reduce the computational overhead and storage access overhead of the edge device for identifying interesting content by comparing the similarity of the overall information between the previous moment and the next moment, can replace the traditional object detection algorithm with a similarity comparison algorithm, and can dynamically switch the detection algorithm to avoid the large computational overhead caused by a single algorithm, and further meet the real-time requirements of different application scenarios.

[0181] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium provided in this application and used in the embodiments can include non-volatile and / or volatile memories. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0182] It should be noted that in this article, the terms "include", "comprise", or any other variant thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that includes a series of elements includes not only those elements but also other elements not expressly listed, or elements that are inherent to such process, apparatus, article, or method. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, apparatus, article, or method that includes the element.

[0183] The above are only the preferred embodiments of this application, and do not limit the patent scope of this application accordingly. Any equivalent structural or equivalent process transformation made by using the specification and drawings of this application, or directly or indirectly applied in other related technical fields, is equally included in the patent protection scope of this application.

Claims

1. A content detection method based on artificial intelligence, characterized in that: The following steps are involved: Receive input content at the current moment through a preset edge device to obtain content to be detected; wherein the content to be detected includes but is not limited to video, picture, text and audio; Detect the content to be detected by using a preset time series and model to obtain a detection result; Determining whether the detection result is content of interest; If so, the input content at the current moment is compared with the input content at the previous moment by a preset similarity comparison algorithm to obtain a similarity value; Determine whether the similarity value is greater than a preset threshold; If yes, mark it as interesting content and output the detection result at the current moment; Detect the input content at the next moment.

2. The content detection method based on artificial intelligence according to claim 1, characterized in that: Before the step of detecting the content to be detected by using a preset time series and model to obtain a detection result, the step includes: Optimizing the model to obtain an optimized model; wherein the optimization method includes but is not limited to compressing parameter quantity, simplifying structure and adjusting training strategy; The optimization model is deployed on the edge device.

3. The content detection method based on artificial intelligence according to claim 1, characterized in that: The step of detecting the content to be detected by using a preset time series and model to obtain a detection result includes: The content to be detected is detected simultaneously by multiple preset processors.

4. The content detection method based on artificial intelligence according to claim 3 is characterized in that: The step of simultaneously detecting the content to be detected by using a plurality of preset processors includes: Identify the overall information of the content to be detected; wherein the overall information includes interesting content and uninteresting content.

5. The content detection method based on artificial intelligence according to claim 4 is characterized in that: If so, the step of comparing the input content at the current moment with the input content at the previous moment by using a preset similarity comparison algorithm to obtain a similarity value includes: Extracting intermediate representation features of the overall information and graying the intermediate representation features; Generate a code based on the intermediate representation feature; An XOR operation is performed bit by bit on the code to quantize the overall information at the current moment and the overall information at the previous moment, and the similarity value is obtained by subtracting them.

6. The artificial intelligence-based content detection method according to claim 5, characterized in that: The step of extracting the intermediate representation features of the overall information and graying the intermediate representation features includes: The low-frequency components of the overall information are extracted to obtain a low-frequency component set.

7. The artificial intelligence-based content detection method according to claim 6, characterized in that: After the step of extracting the low-frequency components of the overall information to obtain a low-frequency component set, the method further comprises: Calculating the mean of the low-frequency components; Each element of the low frequency component set is compared with the mean value and a binary code is generated.

8. The content detection method based on artificial intelligence according to claim 1, characterized in that: The step of determining whether the similarity value is greater than a preset threshold value includes: If not, mark it as uninteresting content, and calculate the similarity value between the input content at the current moment and the input content at the next moment by using the similarity comparison algorithm; Determine whether the similarity value is greater than the threshold; If yes, no testing is done; If not, the test is performed again through the model.

9. A content detection device based on artificial intelligence, characterized in that: include: A receiving module, used to receive input content at the current moment through a preset edge device; The first detection module detects the content to be detected through a preset time series and model to obtain a detection result; A first judging module, used to judge whether the detection result is a content of interest; a comparison module, for, if yes, comparing the input content at the current moment with the input content at the previous moment by using a preset similarity comparison algorithm to obtain a similarity value; The second module is used to determine whether the similarity value is greater than a preset threshold; A marking module, used for marking the content as interesting content if yes, and outputting the detection result at the current moment; The second detection module is used to detect the input content at the next moment.

10. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps in the artificial intelligence-based content detection method described in any one of claims 1 to 8 are implemented.

11. A computer storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the artificial intelligence-based content detection method described in any one of claims 1 to 8 are implemented.

Images