Identification method, system, device and medium

Abstract

The present disclosure provides an identification method, which is applied to the field of financial technology or other fields, including: obtaining a first identification model sent by a master node; obtaining a second identification model sent by the master node, the second identification model is obtained by the master node training according to the first identification model and the first object to be identified; identifying the second object to be identified according to the first identification model and the second identification model respectively; judging whether the identification results of the two are the same, if so, outputting the identification result; otherwise, outputting the result identified by the second identification model, and deleting the first identification model. The present disclosure also provides an identification system, an electronic device, and a computer-readable storage medium.

Description

Technical Field

[0001] The present disclosure relates to the field of financial technology, and specifically to an identification method, system, device and medium. Background Art

[0002] In recent years, the whole society and the whole system have attached increasing importance to customer service, and the requirements for standardized and refined service management have been increasing. At present, banks generally adopt on-site and off-site service supervision methods to monitor the service quality of outlets. Among them, off-site supervision is achieved by reviewing the surveillance videos of outlets, which has a large workload, takes a long time, and is relatively inefficient. In addition, since reviewing the surveillance videos of outlets requires the storage of a large amount of video data, it is impossible to extract effective information from the image video, resulting in a certain waste of resources, and it is impossible to form effective pre-prevention and in-process control in some specific emergency scenarios. Summary of the invention

[0003] 1. Technical issues to be resolved

[0004] In response to the above problems, the present disclosure provides an identification method, system, device and medium for at least partially solving the technical problems of traditional identification methods such as large workload, long occupancy time and relatively low efficiency.

[0005] (II) Technical solution

[0006] On the one hand, the present disclosure provides an identification method, including: obtaining a first identification model sent by a master node; obtaining a second identification model sent by the master node, the second identification model being trained by the master node based on the first identification model and a first object to be identified; identifying the second object to be identified based on the first identification model and the second identification model respectively; determining whether the identification results of the two are the same, and if so, outputting the identification result; otherwise, outputting the result identified by the second identification model and deleting the first identification model.

[0007] Furthermore, obtaining the first recognition model sent by the master node includes: obtaining the first recognition model sent by the master node, obtaining the object to be identified, and determining whether the first recognition model can identify the object to be identified; if so, outputting the recognition result; otherwise, sending the object to be identified to the master node.

[0008] Furthermore, obtaining the second recognition model sent by the master node includes: if the first recognition model cannot recognize the object to be recognized, sending the object to be recognized to the master node; obtaining the second recognition model sent by the master node, the second recognition model is trained by the master node based on the first recognition model and the object to be recognized.

[0009] Furthermore, after deleting the first recognition model, the method further includes repeating the following steps: obtaining a third object to be identified, determining whether the second recognition model can identify the third object to be identified, and if so, outputting the recognition result, otherwise sending the object to be identified to the main node; obtaining a third recognition model sent by the main node, the third recognition model being trained by the main node based on the second recognition model and the third recognition object; identifying a fourth object to be identified based on the second recognition model and the third recognition model respectively; determining whether the recognition results of the two are the same, and if so, outputting the recognition result; otherwise, outputting the result recognized by the third recognition model, and deleting the second recognition model.

[0010] Furthermore, judging whether the recognition results of the two are the same includes: judging whether the two have recognized corresponding results, if neither of them has recognized corresponding results, sending the second object to be recognized to the main node; if both of them have recognized corresponding results, continuing to judge whether the recognition results of the two are the same, if so, outputting the recognition result; otherwise, outputting the result recognized by the second recognition model and deleting the first recognition model.

[0011] Furthermore, it also includes: automatically capturing violations and dangerous behaviors based on the output results, and issuing real-time warnings to the local area.

[0012] Further, identifying the second object to be identified according to the first identification model and the second identification model respectively includes: hot starting the first identification model and the second identification model to identify the second object to be identified respectively.

[0013] On the other hand, the present disclosure provides an identification method, including: sending a first identification model to each of at least one child node; obtaining a first object to be identified sent by one of the at least one child node, and identifying the first object to be identified to obtain a recognition result; training the first identification model according to the object to be identified and the recognition result to obtain a second identification model; and sending the second identification model to each of the at least one child node.

[0014] Furthermore, sending the first recognition model to each of at least one subnode includes: training the model based on initial data; verifying the recognition accuracy of the model; repeating the above steps until the accuracy reaches a preset value, obtaining the first recognition model, and sending the first recognition model to each of at least one subnode.

[0015] Further, obtaining the first object to be identified sent by one of the at least one child nodes includes: one of the at least one child nodes sends the first object to be identified to the central storage; the central storage screens out the first object to be identified; and obtaining the first object to be identified sent by the central storage.

[0016] Furthermore, the first object to be identified is identified to obtain an identification result, including: locally training a model based on the first object to be identified; setting parameters for model operation to obtain an identification result; verifying the recognition accuracy of the model; repeating the above steps until the accuracy reaches a preset value to obtain a second recognition model.

[0017] Another aspect of the present disclosure provides an identification system, including: a main node, used to send a first recognition model and a second recognition model to each of at least one child node, the second recognition model is trained by the main node based on the first recognition model and a first object to be identified; at least one child node, used to identify the second object to be identified based on the first recognition model and the second recognition model respectively; determine whether the recognition results of the two are the same, and if so, output the recognition result; otherwise, output the result of the second recognition model recognition and delete the first recognition model.

[0018] Furthermore, the main node is located on the cloud platform, and at least one child node is an edge computing node.

[0019] Another aspect of the present disclosure provides an electronic device, including: a processor; and a memory storing a computer executable program, which, when executed by the processor, enables the processor to perform the aforementioned identification method.

[0020] Another aspect of the present disclosure provides a computer-readable storage medium having a computer program stored thereon, wherein the program implements the aforementioned identification method when executed by a processor.

[0021] (III) Beneficial effects

[0022] The disclosed embodiments provide an identification method, system, device and medium, which respectively identify objects to be identified through a first identification model and a second identification model, and continuously iterate and update to ensure that the data encountered by each node can have a positive impact on the accuracy of the model, thereby solving the problem in traditional methods that the same type of unidentified objects need to be corrected by traversing each node. Through the cloud-edge architecture, each node can give full play to the edge computing capability and use a centrally deployed model for quasi-real-time response, and not all video data will be transmitted to the cloud, thereby reducing the bandwidth requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 An exemplary system architecture diagram of an identification method according to an embodiment of the present disclosure is schematically shown;

[0024] Figure 2 A flowchart of an identification method according to an embodiment of the present disclosure is schematically shown;

[0025] Figure 3The diagram schematically shows the interaction process between the master node and the sub-node according to the embodiment of the present disclosure;

[0026] Figure 4 A flowchart of performing recognition according to a first recognition model according to an embodiment of the present disclosure is schematically shown;

[0027] Figure 5 A flowchart of an identification method according to another embodiment of the present disclosure is schematically shown;

[0028] Figure 6 The complete workflow diagram of the identification method according to the embodiment of the present disclosure is schematically shown;

[0029] Figure 7 The overall structure diagram of the cloud-edge architecture of the recognition system according to the embodiment of the present disclosure is schematically shown;

[0030] Figure 8 The block diagram of a computer system according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0031] In order to make the objectives, technical solutions and advantages of the present disclosure more clearly understood, the present disclosure is further described in detail below in combination with specific embodiments and with reference to the accompanying drawings.

[0032] The embodiments of the present disclosure provide an identification method, system, device and medium, which make it possible to intelligently capture and analyze occasional specific actions through monitoring through technologies such as big data, artificial intelligence algorithms and face recognition models, and provide technical support for solving the problems mentioned above.

[0033] Figure 1 The exemplary system architecture 100 of the identification method according to the embodiment of the present disclosure is schematically shown. It should be noted that: Figure 1 What is shown is merely an example of a system architecture to which the embodiments of the present disclosure can be applied, in order to help those skilled in the art understand the technical content of the present disclosure, but it does not mean that the embodiments of the present disclosure cannot be used in other devices, systems, environments or scenarios.

[0034] like Figure 1 As shown, the system architecture 100 according to this embodiment may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0035] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, 103, such as camera application, photo application, web browser application, search application, instant messaging tool, social platform software, etc. (only as examples).

[0036] The terminal devices 101 , 102 , and 103 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, and desktop computers.

[0037] The server 105 may be a server that provides various services, such as a background management server (only an example) that provides support for websites browsed by users using the terminal devices 101, 102, and 103. The background management server may analyze and process the received data such as user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal device.

[0038] It should be noted that the identification method provided in the embodiment of the present disclosure can generally be executed by the server 105. Accordingly, the system for identification provided in the embodiment of the present disclosure can generally be set in the server 105. The identification method provided in the embodiment of the present disclosure can also be executed by a server or server cluster that is different from the server 105 and can communicate with the terminal devices 101, 102, 103 and / or the server 105. Accordingly, the system for identification provided in the embodiment of the present disclosure can also be set in a server or server cluster that is different from the server 105 and can communicate with the terminal devices 101, 102, 103 and / or the server 105.

[0039] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0040] Figure 2 The flowchart of the identification method according to the embodiment of the present disclosure is schematically shown.

[0041] S11, obtaining the first recognition model sent by the master node.

[0042] For a diagram of the interaction between the master node and the sub-node, see Figure 3, the recognition method generally includes a main node EN and multiple sub-nodes N1-31, N2-32, N3-33, N4-34 communicating with it. The main node 30 can be a multi-core computing host for model establishment and training; the sub-node 31 can also be a host, and can also include a data acquisition device for performing model operations in different scenarios, outputting operation results, and feeding back corresponding data to the main node. In step S11, the main node EN sends the trained first recognition model to each sub-node N1, N2, N3, N4. Of course, the sub-nodes here can be any number up to at least one.

[0043] S12, obtaining a second recognition model sent by the master node, where the second recognition model is trained by the master node based on the first recognition model and the first object to be recognized.

[0044] After the child node obtains the first recognition model sent by the main node, it can also obtain the second recognition model sent by the main node. The second recognition model is obtained by the main node after model training and algorithm tuning based on the first recognition model. The recognition ability of the second recognition model is usually stronger than that of the first recognition model.

[0045] S13, identifying a second object to be identified according to the first identification model and the second identification model respectively.

[0046] The child node runs the first recognition model and the second recognition model simultaneously, so as to recognize and calculate the same scene at the same time and obtain recognition results respectively.

[0047] S14, determining whether the recognition results of the two are the same, if so, outputting the recognition result; otherwise, outputting the recognition result of the second recognition model and deleting the first recognition model.

[0048] If the first recognition model and the second recognition model have the same recognition results for the same data source, the same recognition results will be output; if the recognition results are different, the first recognition model will be deleted and the second recognition model will work. At this time, the recognition ability of the second recognition model is stronger than that of the first recognition model, so the result obtained when re-recognizing will be more accurate and reliable. At the same time, the second recognition model is also sent to all other child nodes at the same time. That is, through this dual-model operation method, it can be guaranteed that the data encountered by each child node can have a positive impact on the accuracy of the model, and it also solves the problem that the same type of unrecognized data needs to traverse each child node to achieve correction.

[0049] In order to more clearly understand the identification method of the present disclosure, the method is described below through a specific embodiment, taking the scenario in which the head office of a bank supervises the services of each branch outlet as an example.

[0050] Traditional bank head offices generally use on-site and off-site service supervision methods to monitor the service quality of branch outlets. Off-site supervision is achieved by reviewing branch surveillance videos, which is labor-intensive, time-consuming and relatively inefficient. In addition, since reviewing branch surveillance videos requires the storage of a large amount of video data, it is impossible to extract effective information from the image video, resulting in a certain amount of resource waste. In addition, it is impossible to form effective pre-emptive prevention and in-process control in some specific emergency scenarios.

[0051] In the identification method disclosed in the present invention, the system platform of the bank head office is the main node, and the network video recorder (NVR) of each branch outlet is the child node. The bank head office uses AI learning and intelligent analysis methods to set up an outlet monitoring model, uses modern monitoring technology and builds a system platform to remotely monitor and analyze the daily business conditions of the outlets, and conducts non-contact supervision of the service management and customer service work of the bank branches. The bank head office can monitor whether the branch outlets are fully operational. For example, by using the classification algorithm in machine learning to classify the image data of the gates of the outlets across the country when they are opened and closed, a threshold can be obtained. When the threshold is not met, it belongs to the gate not open state. Regularly inspect the switch status of the rolling gate of our bank outlets to determine whether the gate is not opened or half-opened during working hours, or the gate is closed before the end of get off work. The bank head office can also monitor the image standards of the branch outlets. For example, by using the classification algorithm in machine learning to classify the image data of employee uniforms, it can complete the inspection of whether the personnel in the morning meeting time period and the morning meeting area are wearing work clothes. The bank head office can also monitor the dedicated services of branch outlets. For example, by using the classification algorithm in machine learning to identify the image data of employees making phone calls, it can complete the alarm and record the situation of making phone calls or playing with mobile phones in the scanning relevant areas.

[0052] Based on the above application scenario, the main node of the bank head office first needs to deploy the first recognition model to the sub-nodes of each branch outlet, for example, sub-node N1 is located in the urban area of ​​Guangzhou, sub-node N2 is located in the suburbs of Guangzhou, and so on. The first recognition model is specifically a model for judging the opening and closing state of the gate. The recognition result of the first recognition model is that the gate is open or closed. The surveillance cameras of each branch outlet judge whether the branch outlet is fully open by monitoring the light in front of the gate. For example, at 9 o'clock in the morning, the surveillance camera of sub-node N1 in Guangzhou cannot judge whether the gate of the branch is open or an incorrect judgment is made based on the monitoring data, that is, the first recognition model fails to identify the first object to be identified. At this time, the monitoring data within this period of time needs to be sent to the main node of the bank head office. The main node performs iterative model training and accuracy verification operations based on the monitoring data and other data of related scenarios, and sets the parameters of the model operation on the PC client. After the training is completed, the second recognition model is deployed online. At this time, all child nodes, including N1, N2, etc., have received the second recognition model. The second recognition model has solved the problem that child node N1 cannot judge or incorrectly judges the opening and closing status of the gate. Since the geographical location of child node N2 is close to N1, the sunlight in front of its gate is similar to that of child node N1. At this time, the second recognition model running by child node N2 already has the function of accurately judging the opening and closing status of the gate under this condition, and there is no need to experience the problem encountered by child node N1 again. Except for child nodes N1 and N2, other child nodes can also accurately judge the opening and closing status of the gate under this condition, that is, this method solves the problem of the same type of unrecognized, which requires traversing each node to achieve correction; and the data encountered by each child node can have a positive impact on the accuracy of the model.

[0053] At this time, the child node N2 still runs the first recognition model and the second recognition model at the same time. When the recognition results of the first recognition model and the second recognition model are different, the first recognition model is deleted and the second recognition model continues to recognize and judge. By continuously iterating and updating the recognition model, the recognition accuracy of each child node and the ability to cope with different scenarios can be significantly improved.

[0054] Of course, after identifying the output results, if they do not comply with service management regulations, for example, if the branch outlets fail to open the gates on time or close the gates ahead of time, it is also necessary to issue real-time warnings for service problems and service risks at the branch outlets’ sub-nodes, conduct quasi-real-time processing, reduce the duration of the problems, and realize intelligent, automated, and information-based service monitoring.

[0055] Based on the above embodiment, obtaining the first recognition model sent by the main node includes: obtaining the first recognition model sent by the main node, obtaining the object to be identified, and determining whether the first recognition model can identify the object to be identified. If so, outputting the recognition result, otherwise sending the object to be identified to the main node.

[0056] The child node first uses the first recognition model for recognition, see Figure 4 The object to be identified may be, for example, video data or image data captured by a camera. If the identification result can be obtained according to the first identification model, the result is directly output for real-time processing; if the identification result cannot be obtained, the data needs to be sent to the main node for further processing.

[0057] On the basis of the above embodiment, obtaining the second recognition model sent by the master node includes: if the first recognition model cannot identify the object to be identified, sending the object to be identified to the master node; obtaining the second recognition model sent by the master node, the second recognition model is trained by the master node based on the first recognition model and the object to be identified.

[0058] The master node performs model training and algorithm optimization based on the data sent by the child node, and verifies the accuracy. The parameters of the model operation are set on the PC client. The model training in this stage is carried out until the accuracy verification meets the requirements. The second recognition model is obtained, and the master node sends the second recognition model to the child node. The child node runs the first recognition model and the second recognition model at the same time.

[0059] On the basis of the above embodiment, after deleting the first recognition model, it also includes repeating the following steps: obtaining a third object to be identified, determining whether the second recognition model can identify the third object to be identified, and if so, outputting the recognition result, otherwise sending the object to be identified to the main node; obtaining the third recognition model sent by the main node, the third recognition model is trained by the main node according to the second recognition model and the third recognition object; identifying the fourth object to be identified according to the second recognition model and the third recognition model respectively; determining whether the recognition results of the two are the same, and if so, outputting the recognition result; otherwise, outputting the result recognized by the third recognition model, and deleting the second recognition model.

[0060] When the child node runs the first recognition model and the second recognition model at the same time, if the recognition results of the first recognition model and the second recognition model are different, since the second recognition model is trained on the basis of the first recognition model, the accuracy and reliability of the second recognition model will be higher, then the first recognition model needs to be deleted and the second recognition model needs to be run alone. If the second recognition model fails to recognize the object to be recognized again, the data of the object to be recognized will continue to be sent to the main node, and the main node will then perform model training and algorithm tuning based on the data of the object to be recognized, and perform accuracy verification, set the parameters of the model operation on the PC client, and the model training at this stage will be carried out until the accuracy verification meets the requirements, and the third recognition model will be obtained. The main node sends the third recognition model to the child node, and the child node runs the second recognition model and the third recognition model at the same time. By continuously iteratively updating the recognition model, the recognition accuracy of each child node and the ability to cope with different scenarios can be significantly improved.

[0061] Based on the above embodiment, judging whether the recognition results of the two are the same includes: judging whether the two have recognized corresponding results, if neither of them has recognized corresponding results, sending the second object to be recognized to the main node; if both of them have recognized corresponding results, continuing to judge whether the recognition results of the two are the same, if so, outputting the recognition result; otherwise, outputting the result recognized by the second recognition model and deleting the first recognition model.

[0062] When the first recognition model and the second recognition model are running at the same time, different results may appear for the same data to be recognized. If both fail to recognize, a new problem may appear and it needs to be sent to the master node for model training. After the training is completed, the master node sends a new model again. If the recognition results of the two are the same, the same recognition results are output and the two models continue to run at the same time; if the recognition results of the two are different, the first recognition model is deleted and the second recognition model continues to run.

[0063] Based on the above embodiment, it also includes: automatically capturing illegal and dangerous behaviors according to the output results, and issuing real-time warnings to the local area.

[0064] Based on the output results, if the service is judged to be non-compliant or there are hidden dangers, an early warning will be issued to the sub-node, and real-time processing will continue to shorten the duration of the problem.

[0065] On the basis of the above embodiment, identifying the second object to be identified according to the first identification model and the second identification model respectively includes: hot starting the first identification model and the second identification model to identify the second object to be identified respectively.

[0066] The first recognition model and the second recognition model are deployed online at the same time, that is, the new and old models are used in parallel for the same scene, which avoids monitoring gaps and enhances the effective application of image data resources.

[0067] Figure 5 The flowchart of the identification method according to another embodiment of the present disclosure is schematically shown.

[0068] S21: Send a first identification pattern to each of the at least one child node.

[0069] The master node can be a multi-core computing host for model establishment and training; the sub-node can also be a host, and can also include a data acquisition device for performing model operations in different scenarios, outputting operation results, and feeding back corresponding data to the master node. In step S21, the master node EN sends the trained first recognition model to each sub-node N1, N2, N3, N4, of course, the sub-nodes here can be any number up to at least one.

[0070] S22, obtaining a first object to be identified sent by a child node among at least one child node, and identifying the first object to be identified to obtain an identification result.

[0071] If the first recognition model cannot identify the object to be identified, the object to be identified is sent to the master node; the master node performs model training and algorithm tuning based on the data sent by the child node, obtains the recognition result, and verifies the accuracy.

[0072] S23, training the first recognition model according to the object to be recognized and the recognition result to obtain a second recognition model.

[0073] The parameters of the model operation are set on the PC client. The model training in this stage is carried out until the accuracy verification meets the requirements to obtain the second recognition model.

[0074] S24: Send a second recognition model to each of the at least one child node.

[0075] The master node sends the second recognition model to each child node, and the child node runs the first recognition model and the second recognition model simultaneously.

[0076] Based on the above embodiment, sending a first recognition model to each of at least one subnode includes: training the model based on initial data; verifying the recognition accuracy of the model; repeating the above steps until the accuracy reaches a preset value, obtaining a first recognition model, and sending the first recognition model to each of at least one subnode.

[0077] The master node performs model training and accuracy verification operations based on the industry's general algorithms for the initial data of the relevant scenarios, and sets the parameters for model operations on the PC client. The model is trained at this stage until the accuracy verification meets the requirements, and the first recognition model is obtained, which is then sent to each child node.

[0078] Based on the above embodiment, obtaining the first object to be identified sent by one of the at least one child nodes includes: one of the at least one child nodes sends the first object to be identified to the central storage; the central storage screens out the first object to be identified; and obtaining the first object to be identified sent by the central storage.

[0079] The subnode and the main node also include a central storage between the transmission data. The subnode is the edge computing node, and the main node is the cloud. If all videos are to interact with the cloud server in real time, there will be high requirements for bandwidth. The cloud-edge architecture currently used can enable each node to play the edge computing capabilities, use the centrally deployed model for quasi-real-time response, and will not transmit all video data to the cloud, reducing the bandwidth requirements. The data that needs to be processed by the main node is received, stored and forwarded through the central storage, reducing the bandwidth requirements.

[0080] On the basis of the above embodiment, the first object to be identified is identified to obtain the identification result, including: locally training the model according to the first object to be identified; setting the parameters of the model operation to obtain the identification result; verifying the recognition accuracy of the model; repeating the above steps until the accuracy reaches a preset value to obtain a second recognition model.

[0081] The master node performs model training and algorithm optimization based on the data sent by the child node, and verifies the accuracy. The parameters of the model operation are set on the PC client. The model training in this stage is carried out until the accuracy verification meets the requirements. The second recognition model is obtained, and the master node sends the second recognition model to the child node. The child node runs the first recognition model and the second recognition model at the same time.

[0082] Another embodiment of the present disclosure provides a recognition system, including: a main node, used to send a first recognition model and a second recognition model to each of at least one child node, the second recognition model is trained by the main node according to the first recognition model and a first object to be recognized; at least one child node, used to recognize the second object to be recognized according to the first recognition model and the second recognition model respectively; determine whether the recognition results of the two are the same, and if so, output the recognition result; otherwise, output the result recognized by the second recognition model and delete the first recognition model.

[0083] The master node EN sends the trained first recognition model to each child node N1, N2, N3, and N4. Of course, the number of child nodes here can be any number other than at least one. After the child node obtains the first recognition model sent by the master node, it can also obtain the second recognition model sent by the master node. The second recognition model is obtained by the master node after model training and algorithm tuning based on the first recognition model. The recognition ability of the second recognition model is usually stronger than that of the first recognition model. The child node runs the first recognition model and the second recognition model at the same time, so that they can recognize and calculate the same scene at the same time and obtain recognition results respectively. If the first recognition model and the second recognition model have the same recognition results for the same data source, the same recognition result is output; if the recognition results of the two are different, the first recognition model is deleted and the second recognition model is enabled to work. At this time, the recognition ability of the second recognition model is stronger than that of the first recognition model, so the result obtained when recognizing again will be more accurate and reliable. At the same time, the second recognition model is also sent to all other child nodes. That is, through this dual-model operation mode, it can be ensured that the data encountered by each child node can have a positive impact on the accuracy of the model, and it also solves the situation where the same type of unrecognized data needs to traverse each child node to be corrected.

[0084] Based on the above embodiment, the main node is located on the cloud platform, and at least one child node is an edge computing node.

[0085] The sub-nodes are edge computing nodes, and the main node is the cloud. The "cloud-edge architecture" reduces the bandwidth requirements between branch network data and central storage, saves storage resources, and improves data efficiency. Edge computing improves the quasi-real-time processing capabilities of branch network problems and reduces the duration of problems. Cloud computing improves the accuracy of the overall monitoring model and strengthens the effective application of image data resources. The system platform forms a centralized and unified management of the iterative update and online deployment of the model.

[0086] Figure 6 As a specific embodiment of the identification method of the present disclosure, a workflow diagram of the identification method designed by the present disclosure is schematically shown.

[0087] Figure 7 The overall structure diagram of the cloud-edge architecture of the identification system disclosed in the present invention is schematically shown.

[0088] Smart Brain System Platform: As a model generation platform, it performs algorithm training on video data and includes the following modules:

[0089] 1. PC: Client, which can configure relevant parameters during model training.

[0090] 2. "Smart Brain" all-in-one machine: a model training and algorithm tuning platform that uses different algorithms to build relevant models for different scenarios.

[0091] Comprehensive security platform: It can complete the specific application of the model and the storage of video data, including the following modules:

[0092] 1. PC: The client manages image data and models, including the transmission of image data to the cloud and the deployment of models.

[0093] 2. Server: It is responsible for the operation of the model and other background programs. It is connected to the network with the network cameras and related detection equipment to collect data for analysis. It also sets the system quality inspection screening strategy according to the preset quality inspection items to find out the recordings that do not meet the quality inspection specifications and realize the automatic screening of quality inspection.

[0094] 3. Network camera: Captures the actions in the scene and transmits the captured image data to the server, including bucket cameras, hemispherical cameras, ball cameras, and gun cameras.

[0095] 4. NVR: Network video recorder, which manages video data. Video playback can be performed through NVR to screen unrecognized video data and optimize the model.

[0096] 5. Central storage: relevant storage for some videos that need to be recorded.

[0097] The specific workflow of the recognition system is described as follows:

[0098] Step 601: Generate the first generation model, i.e. S11, and obtain the first recognition model sent by the master node. Through the Huinao system platform, the initial data of the relevant scene is used on the "Huinao all-in-one machine" to perform model training and accuracy verification based on the industry general algorithm, and the parameters of the model operation are set on the PC client. The model training at this stage is carried out until the accuracy verification meets the requirements.

[0099] Step 602: Deployment of the first-generation model. The integrated security platform implements the deployment of the first-generation model of the corresponding scenario and deploys each model in the corresponding device of each edge computing node.

[0100] Step 603: After the model is deployed online, each edge computing node collects video data through the camera.

[0101] Step 604: There are two results for the scenarios to which the model is applicable, namely, corresponding recognition and non-recognition.

[0102] Step 605: Branch 1, identifying the corresponding result: this part of the video data is not sent up, but is directly processed in quasi-real time at each edge computing node.

[0103] Step 606: Branch 2, the corresponding result is not identified: This part of the video data is stored in the NVR. According to certain rules, the data is regularly sent to the integrated security platform and transmitted to the central storage. The image data is screened again on the security platform through the PC to obtain valid unidentified data, which is sent to the Smart Brain system platform.

[0104] Step 607: Model iterative deployment. Through the Smart Brain system platform, iterative model training and accuracy verification operations for related scenarios are performed on the "Smart Brain All-in-One Machine". Parameters for model operations are set on the PC client. After the training is completed, the model is deployed online according to step 2, which is S12. The second recognition model sent by the master node is obtained. The second recognition model is trained by the master node based on the first recognition model and the first object to be recognized.

[0105] Step 608: hot start, the integrated security platform deploys the new and old version models online at the same time, that is, the new and old models are deployed in parallel for the same scene.

[0106] Step 609: After the new model is running, the two models work in parallel, that is, S13, and the second object to be identified is identified according to the first recognition model and the second recognition model. If both models fail to identify it, there may be a new scene that cannot be identified, and the data is sent to the master node for training of the new model.

[0107] Step 610: There are three results for the scenario applicable to the model, namely, the same recognition results (recognized, not recognized) and different recognition results (one recognized and the other not recognized), which is S14, and it is determined whether the recognition results of the two are the same. If so, the recognition result is output; otherwise, the result recognized by the second recognition model is output and the first recognition model is deleted.

[0108] Step 611: The same results are identified and all are identified: this part of the video data is not sent up, but is directly processed in quasi-real time at each edge computing node.

[0109] Step 612: Different recognition results: Such results shall be processed in quasi-real time based on the recognition of the new model. At the same time, such data shall be sent to the integrated security platform for verification according to the rules, and those with low accuracy shall be taken offline.

[0110] In the present disclosure, since the model of the same scene can be deployed on multiple edge computing nodes, and the unrecognized part of the data collected by any edge node is sent to the cloud integrated security platform for screening, it is processed centrally and uniformly by the smart brain, and the model accuracy is improved by the hot start method of dual model parallelism, that is, it is guaranteed that the data encountered by each edge node can have a positive impact on the accuracy of the model. The system solves the situation where the same type of unrecognized data needs to be traversed through each node to be corrected. In addition, in the past processing mode, it is generally performed manually to evaluate the scene problems by retaining the video reading, and there is basically no quasi-real-time capability for in-process processing, or because all videos need to interact with the cloud server in real time, there are high requirements for bandwidth. The cloud-edge architecture currently used can enable each node to exert the edge computing capabilities, use the centrally deployed model for quasi-real-time response, and will not transmit all video data to the cloud, reducing the bandwidth requirements.

[0111] Figure 8 A block diagram of an electronic device according to another embodiment of the present disclosure is schematically shown.

[0112] like Figure 8 As shown, the electronic device 800 includes a processor 810 and a computer-readable storage medium 820. The electronic device 800 can execute the method according to the embodiment of the present disclosure.

[0113] Specifically, the processor 810 may include, for example, a general-purpose microprocessor, an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 810 may also include an onboard memory for cache purposes. The processor 810 may be a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0114] Computer readable storage medium 820, for example, can be any medium capable of containing, storing, conveying, propagating or transmitting instructions. For example, readable storage media can include, but are not limited to, electrical, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices, devices or propagation media. Specific examples of readable storage media include: magnetic storage devices, such as magnetic tapes or hard disks (HDD); optical storage devices, such as compact disks (CD-ROMs); memory, such as random access memory (RAM) or flash memory; and / or wired / wireless communication links.

[0115] The computer-readable storage medium 820 may include a computer program 821 , which may include code / computer-executable instructions, which, when executed by the processor 810 , enable the processor 810 to perform the method flow according to the embodiment of the present disclosure and any variation thereof.

[0116] The computer program 821 may be configured to have, for example, a computer program code including a computer program module. For example, in an example embodiment, the code in the computer program 821 may include one or more program modules, for example, including 821A, module 821B, ... It should be noted that the division method and number of modules are not fixed, and those skilled in the art may use appropriate program modules or program module combinations according to actual conditions, and when these program module combinations are executed by the processor 810, the processor 810 may execute the method flow according to the embodiment of the present disclosure and any variation thereof.

[0117] The present disclosure also provides a computer-readable storage medium, which may be included in the device / system / system described in the above embodiments, or may exist independently without being assembled into the device / system / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present disclosure is implemented.

[0118] Although the present disclosure has been shown and described with reference to specific exemplary embodiments of the present disclosure, it should be understood by those skilled in the art that various changes in form and details may be made to the present disclosure without departing from the spirit and scope of the present disclosure as defined by the appended claims and their equivalents. Therefore, the scope of the present disclosure should not be limited to the above-mentioned embodiments, but should be determined not only by the appended claims, but also by the equivalents of the appended claims.

[0119] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present disclosure. It should be understood that the above description is only a specific embodiment of the present disclosure and is not intended to limit the present disclosure. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present disclosure should be included in the protection scope of the present disclosure.

Claims

1. A recognition method, characterized in that: include: Obtain a first recognition model sent by the master node, obtain an object to be identified, determine whether the first recognition model can identify the object to be identified, and if so, output a recognition result; otherwise, send the object to be identified to the master node; The object to be identified is video data or picture data; Acquire a second recognition model sent by the master node, where the second recognition model is trained by the master node according to the first recognition model and a first object to be recognized; Identify a second object to be identified according to the first identification model and the second identification model respectively; Determine whether the two recognize corresponding results. If neither recognizes corresponding results, send the second object to be recognized to the master node. If both recognize corresponding results, continue to determine whether the recognition results of the two are the same. If so, output the recognition result. Otherwise, output the result recognized by the second recognition model and delete the first recognition model. After deleting the first recognition model, the step further includes repeating the following steps: Obtain a third object to be identified, determine whether the second recognition model can identify the third object to be identified, and if so, output a recognition result; otherwise, send the object to be identified to the master node; Acquire a third recognition model sent by the master node, where the third recognition model is trained by the master node according to the second recognition model and the third recognition object; Identify a fourth object to be identified according to the second identification model and the third identification model respectively; Determine whether the recognition results of the two are the same. If so, output the recognition result; otherwise, output the result recognized by the third recognition model and delete the second recognition model.

2. The identification method according to claim 1, characterized in that: Also includes: According to the output results, illegal and dangerous behaviors are automatically captured, and real-time warnings are given to the local area.

3. The identification method according to claim 1, characterized in that: Identifying the second to-be-identified object according to the first recognition model and the second recognition model respectively includes: The first recognition model and the second recognition model are hot started to respectively recognize the second object to be recognized.

4. A recognition method, characterized in that: include: Sending a first recognition model to each of the at least one child node; Acquire a first object to be identified sent by one of the at least one child nodes, and identify the first object to be identified to obtain an identification result; the object to be identified is video data or picture data; Training the first recognition model according to the object to be recognized and the recognition result to obtain a second recognition model; Sending a second recognition model to each of the at least one child node; The sub-node identifies the second object to be identified according to the first identification model and the second identification model respectively; Determine whether the two recognize corresponding results. If neither recognizes corresponding results, send the second object to be recognized to the master node. If both recognize corresponding results, continue to determine whether the recognition results of the two are the same. If so, output the recognition result. Otherwise, output the result recognized by the second recognition model and delete the first recognition model. After deleting the first recognition model, the step further includes repeating the following steps: Obtain a third object to be identified, determine whether the second recognition model can identify the third object to be identified, and if so, output a recognition result; otherwise, send the object to be identified to the master node; Acquire a third recognition model sent by the master node, where the third recognition model is trained by the master node according to the second recognition model and the third recognition object; Identify a fourth object to be identified according to the second identification model and the third identification model respectively; Determine whether the recognition results of the two are the same. If so, output the recognition result; otherwise, output the result recognized by the third recognition model and delete the second recognition model.

5. The identification method according to claim 4, characterized in that: The sending the first recognition model to each of the at least one child node comprises: Train the model based on the initial data; Verifying the recognition accuracy of the model; Repeat the above steps until the accuracy reaches a preset value, obtain the first recognition model, and send the first recognition model to each of the at least one child node.

6. The identification method according to claim 5, characterized in that: The acquiring of the first object to be identified sent by one of the at least one subnodes comprises: One of the at least one subnode sends a first object to be identified to a central storage; The central storage screens out the first object to be identified; Acquire the first object to be identified sent by the central storage.

7. The identification method according to claim 6, characterized in that: The first recognition model is trained according to the object to be recognized and the recognition result to obtain the second recognition model, including: Locally performing model training according to the first object to be identified; Setting the parameters of the model operation to obtain the recognition result; Verifying the recognition accuracy of the model; Repeat the above steps until the accuracy reaches a preset value to obtain the second recognition model.

8. A recognition system, characterized in that: include: The master node is used to send a first recognition model and a second recognition model to each of the at least one child node, where the second recognition model is trained by the master node according to the first recognition model and a first object to be recognized; The object to be identified is video data or picture data; At least one child node is used to identify the second object to be identified according to the first recognition model and the second recognition model respectively; determine whether the two recognize corresponding results, and if neither recognizes the corresponding result, send the second object to be identified to the main node; if both recognize corresponding results, continue to determine whether the recognition results of the two are the same, and if so, output the recognition result; otherwise, output the result recognized by the second recognition model and delete the first recognition model; after deleting the first recognition model, it also includes repeating the following steps: obtaining a third object to be identified, determining whether the second recognition model can identify the third object to be identified, and if so, outputting the recognition result, otherwise sending the object to be identified to the main node; obtaining a third recognition model sent by the main node, and the third recognition model is trained by the main node according to the second recognition model and the third recognition object; Identify the fourth object to be identified according to the second identification model and the third identification model respectively; determine whether the identification results of the two are the same, and if so, output the identification result; otherwise, output the result identified by the third identification model and delete the second identification model.

9. The identification system according to claim 8, characterized in that: The master node is located on a cloud platform, and the at least one child node is an edge computing node.

10. An electronic device comprising: processor; A memory storing a computer executable program, which, when executed by the processor, enables the processor to perform the identification method according to any one of claims 1 to 3.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the recognition method as described in any one of claims 1 to 3 is implemented.

Images