Method and device for improving precision of AI supervision abnormal target detection algorithm
By training general AI models and establishing measurement models and measurement base libraries, the problems of poor generalization capabilities of AI supervision algorithms in chain stores, high sample training costs and high customized processing costs are solved, and more efficient algorithm application and replication promotion are achieved.
Patent Information
- Application Number
- CN202411980920.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-31
- Publication Date
- 2025-05-13
AI Technical Summary
In large-scale applications, existing chain store AI regulatory algorithms have problems such as poor algorithm generalization capabilities, high sample training costs and high customized processing costs. They cannot meet the consistency and stability requirements of chain stores, and it is difficult to achieve efficient replication and promotion.
By obtaining data from multiple stores for neural networks and deep learning training, a basic general AI model is generated, and a general measurement model and a personalized measurement base library are established. The generalized measurement model is used for measurement filtering and false positive filtering, improving the generalization ability and accuracy of the algorithm.
It improves the generalization ability of AI supervision algorithms in different stores, reduces the false alarm rate, reduces the cost of sample training and customized processing, and achieves more efficient replication and promotion.
Smart Images
Figure CN119992445A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of abnormal target supervision technology, and in particular to a method and device for improving the accuracy of an AI-supervised abnormal target detection algorithm. Background Art
[0002] Large-scale chain store AI supervision refers to the use of artificial intelligence algorithms to analyze and process surveillance videos of chain stores to improve store safety and management efficiency. The current store AI supervision algorithm is mainly based on target detection technology, which uses training models to identify and analyze people, objects, and events in stores.
[0003] The current AI supervision algorithm for chain stores has some problems and limitations in large-scale applications: 1) Poor algorithm generalization ability: The current algorithm performs well in some stores, but various false alarms occur in other stores, which cannot meet the consistency and stability requirements of chain stores. 2) High sample training cost: In order to improve the accuracy of the algorithm, it is necessary to continuously increase samples to retrain the model, but this leads to an increase in project delivery costs and cannot meet the needs of large-scale chain store replication and promotion. 3) High customized processing cost: Customized processing for special stores requires a lot of manpower and time costs, and cannot achieve efficient replication and promotion.
[0004] Therefore, a method and device for improving the accuracy of an AI-supervised abnormal target detection algorithm are provided to solve the above problems. Summary of the invention
[0005] The main purpose of the present invention is to solve some problems and limitations in the existing technology of the current AI supervision algorithm for chain stores in large-scale applications: 1) Poor generalization ability of the algorithm: The current algorithm performs well in some stores, but various false alarms occur in other stores, which cannot meet the consistency and stability requirements of chain stores. 2) High sample training cost: In order to improve the accuracy of the algorithm, it is necessary to continuously increase samples to retrain the model, but this leads to an increase in project delivery costs and cannot meet the needs of large-scale chain store replication and promotion. 3) High customized processing cost: Customized processing for special stores requires a lot of manpower and time costs, and cannot achieve efficient replication and promotion.
[0006] A first aspect of the present invention provides a method for improving the accuracy of an AI-supervised abnormal target detection algorithm, the method comprising: Obtain data from multiple stores as sample data, call the neural network system and deep learning system to train the sample data, and generate an AI model; Obtain the AI model and establish a general measurement model based on the AI model; generate a general measurement base library and a personalized measurement base library based on sample data; Obtain the data of the stores to be supervised, use the AI model to detect the data of the stores to be supervised, and generate anomaly detection target images; According to the anomaly detection target image, store information in the store data to be supervised described in the anomaly detection target image is obtained, and whether it is a special store is determined according to the store information. If so, the personalized measurement base database is used as the measurement base database of the general measurement model; if not, the general measurement base database is used as the measurement base database of the general measurement model; The measurement is filtered through the general measurement model, and the filtering result information is reported.
[0007] Optionally, obtain the store data to be supervised, use the AI model to detect the store data to be supervised, and generate anomaly detection target images including: Obtain the store data to be supervised, and extract RSTP stream information or frame-sampling image information from the store data to be supervised; Use AI models to detect RSTP stream information or frame extraction image information, and detect complete image information; Determine whether the detected image information is an abnormal detection target image; if not, end the detection process.
[0008] Optionally, if the detected image information is an abnormal detection target image, then "execute the process of obtaining the store information in the store data to be supervised described in the abnormal detection target image according to the abnormal detection target image."
[0009] Optionally, the universal metric model includes a feature extraction network and a similarity calculation network, and the metric filtering through the universal metric model includes: Extracting abnormal feature information in the abnormality detection target image through the feature extraction network; Calculate the similarity between the abnormal feature information and the data information in the measurement base database through the similarity calculation network; Determine whether the similarity exceeds the preset value; if so, generate abnormal filtering result information; if not, it is a normal store.
[0010] Optionally, it also includes: establishing a black and white list mechanism based on sample data, and filtering out false alarms reported by general stores through AI model analysis through the black and white list mechanism.
[0011] Optionally, the performing metric filtering by using a universal metric model and reporting filtering result information includes: Perform metric filtering through a common metric model to generate filtering result information; Generate final alarm information based on the filtering result information and report the final alarm information.
[0012] A second aspect of the present invention provides an AI-supervised abnormal target detection algorithm accuracy improvement device, comprising: The AI model generation module is used to obtain data from multiple stores as sample data, call the neural network system and deep learning system to train the sample data, and generate an AI model; The base database generation module is used to obtain the AI model and establish a general measurement model based on the AI model; generate a general measurement base database and a personalized measurement base database based on sample data; The anomaly detection target image generation module is used to obtain the data of the stores to be supervised, use the AI model to detect the data of the stores to be supervised, and generate anomaly detection target images; The first judgment module is used to obtain store information in the store data to be supervised described in the anomaly detection target image according to the anomaly detection target image, and judge whether it is a special store according to the store information. If so, the personalized measurement base library is used as the measurement base library of the general measurement model; if not, the general measurement base library is used as the measurement base library of the general measurement model; The result reporting module is used to perform measurement filtering through a common measurement model and report the filtering result information.
[0013] The abnormal detection target image generation module includes: an image extraction unit, an image detection unit and an image judgment unit; An image extraction unit is used to obtain the supervised store data and extract RSTP stream information or frame extraction image information from the supervised store data; The picture detection unit is used to detect RSTP stream information or frame extraction picture information using an AI model to detect the complete picture information; The image judgment unit is used to judge whether the detected image information is an abnormal detection target image; if not, the detection process ends.
[0014] The result reporting module includes: an abnormal feature extraction unit, a similarity calculation unit and a similarity judgment unit; An abnormal feature extraction unit, used to extract abnormal feature information in the abnormal detection target image through the feature extraction network; A similarity calculation unit, used for calculating the similarity between the abnormal feature information and the data information in the measurement base database through the similarity calculation network; The similarity judgment unit is used to judge whether the similarity exceeds a preset value; if so, abnormal filtering result information is generated; if not, it is a normal store.
[0015] A third aspect of the present invention provides an electronic device, the electronic device comprising a memory and at least one processor, the memory storing instructions; The at least one processor calls the instructions in the memory so that the electronic device executes each step of the method for improving the accuracy of the AI-supervised abnormal target detection algorithm as described above.
[0016] A fourth aspect of the present invention provides a computer-readable storage medium having instructions stored thereon, which, when executed by a processor, implement the various steps of the method for improving the accuracy of the AI-supervised abnormal target detection algorithm as described above.
[0017] In the technical solution of the present invention, by collecting data from multiple stores and conducting neural network and deep learning training, a basically general AI model can be trained to improve the accuracy of the algorithm. By establishing a general measurement model, a general measurement base library and a black and white list mechanism, false alarms of reported alarm images can be filtered, and the algorithm can have better generalization capabilities in different stores and reduce the false alarm rate. For special stores, by establishing a store-level measurement base library, more accurate alarm filtering can be performed efficiently for special scenarios, further improving the accuracy of the algorithm. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] Figure 1 A flowchart of a method for improving the accuracy of an AI-supervised abnormal target detection algorithm provided in an embodiment of the present invention; Figure 2 A schematic diagram of the structure of a device for improving the accuracy of an AI-supervised abnormal target detection algorithm provided by an embodiment of the present invention; Figure 3 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention; Figure 4 A flowchart for generating anomaly detection target images according to the present invention. DETAILED DESCRIPTION
[0019] The embodiment of the present invention provides a method for improving the accuracy of an AI-supervised abnormal target detection algorithm, including obtaining data from multiple stores as sample data, calling a neural network system and a deep learning system to train the sample data to generate an AI model; obtaining the AI model, and establishing a general measurement model according to the AI model; generating a general measurement base library and a personalized measurement base library according to the sample data; obtaining the data of the stores to be supervised, using the AI model to detect the data of the stores to be supervised, and generating an abnormal detection target image; according to the abnormal detection target image, obtaining the store information in the data of the stores to be supervised described in the abnormal detection target image, and judging whether it is a special store according to the store information, and if so, using the personalized measurement base library as the measurement base library of the general measurement model; if not, using the general measurement base library as the general measurement base library The measurement base library of the measurement model; the measurement filtering is performed through the general measurement model, and the filtering result information is reported. The present invention solves some problems and limitations of the current chain store AI supervision algorithm in large-scale applications in the prior art: 1) Poor generalization ability of the algorithm: The current algorithm performs well in some stores, but various false alarms occur in other stores, and it cannot meet the consistency and stability requirements of chain stores; 2) High sample training cost: In order to improve the accuracy of the algorithm, it is necessary to continuously increase the samples for retraining the model, but this leads to an increase in the project delivery cost and cannot meet the needs of large-scale chain store replication and promotion; 3) High customized processing cost: Customized processing for special stores requires a lot of manpower and time costs, and cannot achieve efficient replication and promotion.
[0020] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0021] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , the first embodiment of the method for improving the accuracy of the AI supervision abnormal target detection algorithm in the embodiment of the present invention includes: Obtain data from multiple stores as sample data, call the neural network system and deep learning system to train the sample data, and generate an AI model; Obtain an AI model, and establish a universal measurement model based on the AI model; the universal measurement model includes feature extraction and similarity calculation; generate a universal measurement base library and a personalized measurement base library based on sample data; Obtain the data of the stores to be supervised, use the AI model to detect the data of the stores to be supervised, and generate anomaly detection target images; Specifically, they include: Obtain the store data to be supervised, and extract RSTP stream information or frame-sampling image information from the store data to be supervised; Use AI models to detect RSTP stream information or frame extraction image information, and detect complete image information; Determine whether the detected image information is an abnormal detection target image; if not, end the detection process; if the detected image information is an abnormal detection target image, enter the "procedure of obtaining store information in the store data to be supervised described in the abnormal detection target image according to the abnormal detection target image"; According to the anomaly detection target image, store information in the store data to be supervised described in the anomaly detection target image is obtained, and whether it is a special store is determined according to the store information. If so, the personalized measurement base database is used as the measurement base database of the general measurement model; if not, the general measurement base database is used as the measurement base database of the general measurement model; Perform metric filtering through a common metric model and report the filtering result information; Specifically, it includes: extracting abnormal feature information in the abnormal detection target image through the feature extraction network; Calculate the similarity between the abnormal feature information and the data information in the measurement base database through the similarity calculation network; Determine whether the similarity exceeds the preset value; if so, generate abnormal filtering result information; if not, it is a normal store.
[0022] See also Figure 1 , a second embodiment of the method for improving the accuracy of an AI-supervised abnormal target detection algorithm in an embodiment of the present invention includes: Obtain data from multiple stores as sample data, call the neural network system and deep learning system to train the sample data, and generate an AI model; Obtain an AI model, and establish a universal measurement model based on the AI model; the universal measurement model includes feature extraction and similarity calculation; generate a universal measurement base library and a personalized measurement base library based on sample data; Obtain the data of the stores to be supervised, use the AI model to detect the data of the stores to be supervised, and generate anomaly detection target images; Specifically, they include: Obtain the store data to be supervised, and extract RSTP stream information or frame-sampling image information from the store data to be supervised; Use AI models to detect RSTP stream information or frame extraction image information, and detect complete image information; Determine whether the detected image information is an abnormal detection target image; if not, end the detection process; if the detected image information is an abnormal detection target image, enter the "procedure of obtaining store information in the store data to be supervised described in the abnormal detection target image according to the abnormal detection target image"; According to the anomaly detection target image, store information in the store data to be supervised described in the anomaly detection target image is obtained, and whether it is a special store is determined according to the store information. If so, the personalized measurement base database is used as the measurement base database of the general measurement model; if not, the general measurement base database is used as the measurement base database of the general measurement model; Perform metric filtering through a common metric model and report the filtering result information; Specifically, it includes: extracting abnormal feature information in the abnormal detection target image through the feature extraction network; Calculate the similarity between the abnormal feature information and the data information in the measurement base database through the similarity calculation network; Determine whether the similarity exceeds the preset value; if so, generate abnormal filtering result information; if not, it is a normal store.
[0023] It also includes: establishing a black and white list mechanism based on sample data, and filtering out false alarms reported by general stores through AI model analysis through the black and white list mechanism.
[0024] The above describes the method for improving the accuracy of the AI-supervised abnormal target detection algorithm in the embodiment of the present invention. The following describes the device for improving the accuracy of the AI-supervised abnormal target detection algorithm in the embodiment of the present invention. Figure 2 , the device for improving the accuracy of the abnormal target detection algorithm supervised by AI in the embodiment of the present invention includes: The AI model generation module 201 is used to obtain data from multiple stores as sample data, call the neural network system and the deep learning system to train the sample data, and generate an AI model; The base database generation module 202 is used to obtain an AI model, establish a general measurement model according to the AI model, and generate a general measurement base database and a personalized measurement base database according to sample data; The abnormality detection target image generation module 203 is used to obtain the store data to be supervised, use the AI model to detect the store data to be supervised, and generate an abnormality detection target image; The first judgment module 204 is used to obtain store information in the store data to be supervised described in the anomaly detection target image according to the anomaly detection target image, and judge whether it is a special store according to the store information. If so, the personalized measurement base database is used as the measurement base database of the general measurement model; if not, the general measurement base database is used as the measurement base database of the general measurement model; The result reporting module 205 is used to perform metric filtering through the general metric model and report the filtering result information.
[0025] The abnormal detection target image generation module includes: an image extraction unit, an image detection unit and an image judgment unit; An image extraction unit is used to obtain the supervised store data and extract RSTP stream information or frame extraction image information from the supervised store data; The picture detection unit is used to detect RSTP stream information or frame extraction picture information using an AI model to detect the complete picture information; The image judgment unit is used to judge whether the detected image information is an abnormal detection target image; if not, the detection process ends.
[0026] The result reporting module includes: an abnormal feature extraction unit, a similarity calculation unit and a similarity judgment unit; An abnormal feature extraction unit, used to extract abnormal feature information in the abnormal detection target image through the feature extraction network; A similarity calculation unit, used for calculating the similarity between the abnormal feature information and the data information in the measurement base database through the similarity calculation network; The similarity judgment unit is used to judge whether the similarity exceeds a preset value; if so, abnormal filtering result information is generated; if not, it is a normal store.
[0027] above Figure 2 The device for improving the accuracy of abnormal target detection algorithm under AI supervision in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The electronic device in the embodiment of the present invention is described in detail from the perspective of hardware processing.
[0028] Figure 37 is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. The electronic device 700 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 710 (for example, one or more processors) and a memory 720, and one or more storage media 730 (for example, one or more mass storage devices) storing application programs 733 or data 732. Among them, the memory 720 and the storage medium 730 can be temporary storage or permanent storage. The program stored in the storage medium 730 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the electronic device 700. Furthermore, the processor 710 may be configured to communicate with the storage medium 730 to execute a series of instruction operations in the storage medium 730 on the electronic device 700.
[0029] The electronic device 700 may also include one or more power supplies 740, one or more wired or wireless network interfaces 750, one or more input and output interfaces 750, and / or one or more operating systems 731, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. It will be appreciated by those skilled in the art that Figure 3 The structure of the electronic device shown does not constitute a limitation on the electronic device, and may include more or less components than shown in the figure, or combine some components, or arrange the components differently.
[0030] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the method for improving the accuracy of an AI-supervised abnormal target detection algorithm.
[0031] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device, or unit can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.
[0032] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.
[0033] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for improving the accuracy of an AI-supervised abnormal target detection algorithm, characterized in that: The method for improving the accuracy of the AI-supervised abnormal target detection algorithm includes: Obtain data from multiple stores as sample data, call the neural network system and deep learning system to train the sample data, and generate an AI model; Obtain the AI model and establish a general measurement model based on the AI model; generate a general measurement base library and a personalized measurement base library based on sample data; Obtain the data of the stores to be supervised, use the AI model to detect the data of the stores to be supervised, and generate anomaly detection target images; According to the anomaly detection target image, store information in the store data to be supervised described in the anomaly detection target image is obtained, and whether it is a special store is determined according to the store information. If so, the personalized measurement base database is used as the measurement base database of the general measurement model; if not, the general measurement base database is used as the measurement base database of the general measurement model; The measurement is filtered through the general measurement model, and the filtering result information is reported.
2. The method for improving the accuracy of the AI-supervised abnormal target detection algorithm according to claim 1 is characterized in that: Obtain the store data to be supervised, use the AI model to detect the store data to be supervised, and generate abnormal detection target images including: Obtain the store data to be supervised, and extract RSTP stream information or frame-sampling image information from the store data to be supervised; Use AI models to detect RSTP stream information or frame extraction image information, and detect complete image information; Determine whether the detected image information is an abnormal detection target image; if not, end the detection process.
3. The method for improving the accuracy of the AI-supervised abnormal target detection algorithm according to claim 2 is characterized in that: The step of obtaining store information in the store data to be supervised in the abnormal detection target image according to the abnormal detection target image includes: If the detected image information is an abnormal detection target image, the process of obtaining the store information in the store data to be supervised in the abnormal detection target image according to the abnormal detection target image is executed.
4. The method for improving the accuracy of the AI-supervised abnormal target detection algorithm according to claim 1 is characterized in that: The universal metric model includes a feature extraction network and a similarity calculation network, and the metric filtering is performed by the universal metric model, including: Extracting abnormal feature information in the abnormality detection target image through the feature extraction network; Calculate the similarity between the abnormal feature information and the data information in the measurement base database through the similarity calculation network; Determine whether the similarity exceeds the preset value; if so, generate abnormal filtering result information; if not, it is a normal store.
5. The method for improving the accuracy of the AI-supervised abnormal target detection algorithm according to claim 1 is characterized in that: Also includes: A black and white list mechanism is established based on sample data, and false alarms reported by general stores through AI model analysis are filtered out through the black and white list mechanism.
6. The method for improving the accuracy of the AI-supervised abnormal target detection algorithm according to claim 1 is characterized in that: The metric filtering is performed by using a universal metric model, and the filtering result information is reported, including: Perform metric filtering through a common metric model to generate filtering result information; Generate final alarm information based on the filtering result information and report the final alarm information.
7. An AI-supervised abnormal target detection algorithm accuracy improvement device, characterized in that: include: The AI model generation module is used to obtain data from multiple stores as sample data, call the neural network system and deep learning system to train the sample data, and generate an AI model; The base database generation module is used to obtain the AI model and establish a general measurement model based on the AI model; generate a general measurement base database and a personalized measurement base database based on sample data; The anomaly detection target image generation module is used to obtain the data of the stores to be supervised, use the AI model to detect the data of the stores to be supervised, and generate anomaly detection target images; The first judgment module is used to obtain store information in the store data to be supervised described in the anomaly detection target image according to the anomaly detection target image, and judge whether it is a special store according to the store information. If so, the personalized measurement base library is used as the measurement base library of the general measurement model; If not, the universal measurement base is used as the measurement base of the universal measurement model; The result reporting module is used to perform measurement filtering through a common measurement model and report the filtering result information.
8. An electronic device, comprising a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory so that the electronic device executes each step of the method for improving the accuracy of the AI-supervised abnormal target detection algorithm as described in any one of claims 1-6.
9. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instruction is executed by the processor, each step of the method for improving the accuracy of the AI-supervised abnormal target detection algorithm as claimed in any one of claims 1 to 6 is implemented.