A method of monitoring cash operations and related devices

Abstract

The application discloses a cash operation monitoring method and related device, which can be applied to the field of artificial intelligence or the field of finance. The terminal device receives a monitoring image of a target cash operation in real time, performs local image enhancement on a target local region of the monitoring image, inputs the enhanced monitoring image into a pre-configured risk identification model, obtains a risk identification result output by the risk identification model, judges whether the target cash operation conforms to a safe operation rule according to the risk identification result, and generates an alarm information if not. The scheme performs risk identification on the monitoring image of the cash operation through the risk identification model deployed on the terminal device, timely alarms the risk behavior, improves the real-time performance of the risk identification of the cash operation, improves the cash identification performance through the local image enhancement of the target local region, reduces the complexity of the local image enhancement, improves the identification efficiency, and improves the accuracy and robustness of the identification result of the risk identification model.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer vision, and in particular to a cash operation monitoring method and related device. BACKGROUND

[0002] A vault is configured with multiple cameras as a key security area in a financial scene. Taking a bank vault as an example, a bank security system performs all-weather monitoring on the vault environment through multiple monitors. Cash operation is a key regulatory object of the bank system. The bank system usually adopts a manual real-time observation of the monitoring screen to find out whether there is a cash operation risk. After finding out that there may be a risk, an alarm device of the bank security system is manually triggered.

[0003] The alarm mode of manually observing the monitoring screen has the problems of missed or false checking. The risk reason can only be investigated by using the historical monitoring video stream stored in the vault security system after a security event occurs, so that the real-time performance and accuracy of the risk alarm of the bank security system cannot be guaranteed. SUMMARY

[0004] In view of the above problems, the present application provides a cash operation monitoring method and related device to achieve the purpose of improving the real-time performance and accuracy of cash operation monitoring. The specific scheme is as follows:

[0005] The first aspect of the present application provides a cash operation monitoring method, comprising:

[0006] receiving a monitoring image of a target cash operation in real time, performing local image enhancement on a target local area of the monitoring image to obtain an enhanced monitoring image, the target local area being a specified area of the monitoring image in which cash is located when the target cash operation is performed;

[0007] inputting the enhanced monitoring image into a pre-configured risk identification model to obtain a risk identification result output by the risk identification model, the risk identification model comprising a cash identification model and a personnel identification model, and the risk identification result comprising a cash identification result output by the cash identification model and a personnel identification result output by the personnel identification model;

[0008] obtaining a safe operation rule of the target cash operation, the safe operation rule comprising at least a specified number of personnel in a cash area and a specified type of personnel;

[0009] judging whether the target cash operation conforms to the safe operation rule according to the risk identification result;

[0010] if the target cash operation does not conform to the safe operation rule, generating an alarm information, the alarm information comprising an alarm image, the risk identification result and the safe operation rule.

[0011] In a possible implementation, the target local area of the monitoring image is subjected to local image enhancement to obtain an enhanced monitoring image, including:

[0012] Based on the specified execution area of the target cash operation and the shooting angle of the monitoring image, the target local area is extracted from the monitoring image;

[0013] The target local area is subjected to contrast enhancement and edge sharpening to obtain the enhanced monitoring image.

[0014] In a possible implementation, the enhanced monitoring image is input into a pre-configured risk identification model to obtain a risk identification result output by the risk identification model, including:

[0015] After the enhanced monitoring image is input into a cash identification model to obtain a cash identification result, if the cash identification result indicates that there is a cash image, the enhanced monitoring image is input into a personnel identification model to obtain a personnel identification result.

[0016] In a possible implementation, the monitoring method of the cash operation further includes:

[0017] Obtaining feedback information of historical alarm information, the feedback information of the historical alarm information including whether the cash identification result is correct and whether the personnel identification result is correct, and the corrected cash identification result and the corrected personnel identification result

[0018] Based on the feedback information of the alarm information, an alarm image is labeled to obtain a retraining data set;

[0019] The retraining data set includes a cash identification retraining data set and a personnel identification retraining data set, the cash identification retraining data set including at least a first type of alarm image and a cash label, wherein the first type of alarm image is an alarm image with an incorrect cash identification result, and the cash label of the first type of alarm image is a corrected cash identification result, and the personnel identification retraining data set including at least a second type of alarm image and a personnel label, wherein the second type of alarm image is an alarm image with an incorrect personnel identification result, and the personnel label of the second type of alarm image is a corrected personnel identification result;

[0020] When a preset retraining opportunity is reached, the risk identification model is retrained based on the retraining data set to obtain retraining parameter information; or the retraining data set and a retraining instruction are sent to a server to enable the server to retrain the risk identification model based on the retraining data set and return retraining parameter information; the retraining parameter information including a hyperparameter group and a model parameter;

[0021] The risk identification model is optimized based on the retraining parameter information.

[0022] In a possible implementation, the retraining of the risk identification model based on the retraining data set comprises:

[0023] obtaining preconfigured hyperparameter search information, the hyperparameter search information comprising preset categories and corresponding preset value ranges;

[0024] if the number of training times does not reach a preset number threshold, configuring the risk identification model by using the preset hyperparameter group, retraining the risk identification model based on the retraining data set, and obtaining retraining parameter information after a preset training completion condition is reached;

[0025] wherein the preset hyperparameter group comprises preset values of hyperparameters of multiple preset categories obtained by random selection in the hyperparameter search information;

[0026] if the number of training times reaches the number threshold, obtaining a target hyperparameter group according to a hyperparameter search result, configuring the risk identification model by using the target hyperparameter group, training the risk identification model based on the retraining data set, and obtaining retraining parameter information after the training completion condition is reached; wherein the target hyperparameter group comprises target values of hyperparameters of multiple target categories.

[0027] In a possible implementation, the monitoring method of the cash operation further comprises, before the enhanced monitoring image is input into the preconfigured risk identification model:

[0028] receiving initial configuration information of a cash identification model and a personnel identification model obtained by a server based on training of a machine learning model on a training data set, and deploying the cash identification model and the personnel identification model based on the initial configuration information, the initial configuration information comprising an initial hyperparameter group and initial model parameters.

[0029] The second aspect of the present application provides a monitoring device for a cash operation, comprising:

[0030] a local enhancement unit configured to receive, in real time, a monitoring image of a target cash operation, perform local image enhancement on a target local area of the monitoring image, and obtain an enhanced monitoring image, the target local area being a specified area of the monitoring image in which cash is present when the target cash operation is performed;

[0031] a risk identification unit configured to input the enhanced monitoring image into a preconfigured risk identification model, and obtain a risk identification result output by the risk identification model, the risk identification model comprising a cash identification model and a personnel identification model, and the risk identification result comprising a cash identification result output by the cash identification model and a personnel identification result output by the personnel identification model;

[0032] An operation rule determination unit is configured to obtain a safe operation rule of the target cash operation, wherein the safe operation rule at least includes A specified number of people and a specified type of people included in the cash area;

[0033] A risk determination unit is configured to determine whether the target cash operation conforms to the safe operation rule according to the risk identification result.

[0034] A risk warning unit is configured to generate a warning information if the target cash operation does not conform to the safe operation rule, wherein the warning information includes a warning image, the risk identification result and the safe operation rule.

[0035] The third aspect of the present application provides a computer program product, which includes computer readable instructions, when the computer readable instructions are executed on an electronic device, the electronic device implements the cash operation monitoring method of the first aspect or any implementation manner of the first aspect.

[0036] The fourth aspect of the present application provides an electronic device, which includes at least one processor and a memory connected with the processor, wherein:

[0037] The memory is configured to store a computer program;

[0038] The processor is configured to execute the computer program, so that the electronic device can implement the cash operation monitoring method of the first aspect or any implementation manner of the first aspect.

[0039] The fifth aspect of the present application provides a computer storage medium, which carries one or more computer programs, when the one or more computer programs are executed by an electronic device, the electronic device can implement the cash operation monitoring method of the first aspect or any implementation manner of the first aspect.

[0040] By the technical scheme, the application provides a cash operation monitoring method and related device, the terminal device receives a monitoring image of a target cash operation in real time, performs local image enhancement on a target local region of the monitoring image to obtain an enhanced monitoring image, the target local region is a specified region of the cash in the monitoring image when the target cash operation is performed, the enhanced monitoring image is input into a pre-configured risk identification model to obtain a risk identification result output by the risk identification model, the risk identification model includes a cash identification model and a personnel identification model, and the risk identification result includes a cash identification result output by the cash identification model and a personnel identification result output by the personnel identification model. A safe operation rule of the target cash operation is acquired, and the safe operation rule at least includes a specified number of personnel in the cash region and a specified type of personnel. Whether the target cash operation conforms to the safe operation rule is judged according to the risk identification result. If the target cash operation does not conform to the safe operation rule, an alarm information is generated, and the alarm information includes an alarm image, the risk identification result and the safe operation rule. The risk identification model deployed on the terminal device can identify the risk of the monitoring image of the cash operation in real time, and timely alarm the risk behavior that does not conform to the safe operation rule, thereby improving the real-time performance of the risk identification of the cash operation. The local enhancement of the target local region can improve the cash identification performance, reduce the complexity of the local image enhancement, improve the identification efficiency, and improve the accuracy and robustness of the identification result of the risk identification model. BRIEF DESCRIPTION OF DRAWINGS

[0041] The above and other features, advantages, and aspects of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the drawings, the same or similar reference numerals are used to refer to the same or similar elements. It should be understood that the drawings are diagrammatic and schematic representations of elements, and not necessarily to scale.

[0042] Figure 1 An architecture diagram of a distributed cash operation monitoring system provided by the application;

[0043] Figure 2 A hardware structure schematic diagram of a terminal provided by the application;

[0044] Figure 3 A hardware structure schematic diagram of a server provided by the application;

[0045] Figure 4 A specific structure schematic diagram of a cash operation monitoring system provided by the application;

[0046] Figure 5 A specific implementation flowchart of a cash operation monitoring method provided by an embodiment of the application;

[0047] Figure 6A flowchart of a cash operation monitoring method provided by an embodiment of the present application is shown in FIG. 1.

[0048] Figure 7 A structural diagram of a cash operation monitoring device provided by an embodiment of the present application is shown in FIG. 2.

[0049] Figure 8 A structural diagram of an electronic device provided by an embodiment of the present application is shown in FIG. 3. DETAILED DESCRIPTION

[0050] The embodiments of the present application are described below with reference to the accompanying drawings. The terms used in the embodiment part of the present application are only used to explain the specific embodiments of the present application, and are not intended to limit the present application.

[0051] The embodiments of the present application are described below with reference to the accompanying drawings. It is known to those of ordinary skill in the art that, as technology develops and new scenarios appear, the technical solutions provided by the embodiments of the present application are also applicable to similar technical problems.

[0052] The terms "first", "second", and the like in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or sequence. It should be understood that the terms used in this way can be interchanged under appropriate circumstances, and this is only a distinguishing way used in the description of the embodiments of the present application to describe the objects with the same attributes. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, so that the processes, methods, systems, products or devices containing a series of units do not have to be limited to those units, but can include other units not clearly listed or inherent to these processes, methods, products or devices.

[0053] The present application can be applied in the technical field of bank vault risk monitoring, and in particular can be applied in the monitoring of cash operations in the vault.

[0054] The importance of cash management in the vault cannot be ignored, which is directly related to the asset safety, operation efficiency and risk prevention and control of banks or financial institutions. The operation of cash by staff is the key supervision object of cash management. Cash in the vault needs to be strictly managed to ensure the safety of bank assets. At present, through strict cash management system, it can effectively prevent cash from being stolen, damaged or lost, and ensure the safety and integrity of assets. The cash management system includes but is not limited to multiple security protection measures such as strict implementation of requirements such as periodic inventory and limiting the number of personnel contact and authority, and formulating cash operation specifications for each cash operation. For example, the management system of limiting the number of personnel contact and authority specifically includes: the behavior involving cash operation in the vault is required to have multiple people on site and the type of personnel is specified. It can be understood that the multiple people on site with specified personnel types can effectively ensure the fairness and accuracy of cash operation. In addition, multiple people on site can also improve the transparency and traceability of cash operation. Under the joint participation of multiple people, each link in the cash operation process can be clearly recorded and effectively monitored, which helps to find and solve problems in time and ensure that the cash operation conforms to the safety operation rules and norms.

[0055] The existing monitoring method is mainly through manual monitoring of the cash operation area in the vault. The main method is to manually check the monitoring screen or play back the monitoring video after the event, to subjectively judge by the naked eye whether the type and number of personnel in the cash operation area during the cash operation conform to the cash management system, and whether there are behaviors that do not conform to the cash operation specification, and manually trigger the bank security system to alarm the vault risk alarm device to achieve the purpose of risk alarm. Limited by the subjectivity and ability difference of manual observation, the existing bank security system has the technical defects of low real-time, low accuracy and high lag in the alarm of cash management risk monitoring.

[0056] In order to solve the above technical defects, the present scheme proposes to use machine vision technology to collect monitoring images in real time and use a pre-trained risk monitoring model to identify whether there is a risk in the video image. If there is, the bank security system will automatically trigger the alarm device for the vault risk. The risk monitoring model is based on a deep learning model and is optimized, thereby improving the real-time monitoring of cash operation in the cash management of the vault. Further, the present scheme can solve the problems of large manual workload and inspection lag, and can cover all vault environments and all time periods, reducing the omission and subjective error of manual inspection.

[0057] Through research, it is found that the existing deep learning model has some technical defects in the monitoring scene of cash operation in the cash management of the vault, such as:

[0058] 1. Data Dependency and Quality Issues: The effectiveness of deep learning models largely depends on the quantity and quality of the input data. If there is insufficient data on vault cash operations, inaccurate labels, or noise, the model's training performance will be affected, potentially leading to false positives or false negatives. Furthermore, due to the nature of vaults, they cannot provide a large amount of operational data, further hindering training.

[0059] 2. Model accuracy: Due to incomplete training data and the complexity of the vault scenario, the accuracy rate was low after the initial deployment.

[0060] 3. The bank's security system is deployed in the form of a central server. The deep learning model is deployed on the server in the bank's security system (such as a cloud server). The image-sensitive information involved in various operations in the vault cash management involves security and privacy protection issues during the transmission to the server, which may lead to the leakage or malicious use of the image-sensitive information.

[0061] To address this, this solution further optimizes the cash handling monitoring scheme by combining machine vision and edge computing technologies to propose a distributed cash handling monitoring system. In this system, edge-side terminal devices deploy machine learning models using trained model parameters. These models capture real-time monitoring images via cameras and use the pre-deployed machine learning models to identify risks associated with cash handling within the images. Specifically, by analyzing the types and numbers of people handling cash in the monitoring images in real time, it determines whether any violations have occurred. Distributed deployment of edge-side terminals reduces the computational burden on servers and prevents the leakage or malicious use of sensitive image information, thereby improving data security.

[0062] Furthermore, the cash handling monitoring system trains and optimizes the parameters of the machine learning model via the server, and then distributes the trained parameters to the terminal devices deployed at the edge, enabling the terminal devices to deploy and optimize the machine learning model. This solution improves the recognition accuracy of the machine learning model by training and optimizing its parameters via the server.

[0063] This application provides a distributed cash operation monitoring system, see [link to relevant documentation]. Figure 1 , Figure 1 A schematic diagram of a system architecture is shown. The system may include a terminal 100 and a server 200. The server 200 may include one or more servers (…). Figure 1 (The example includes a server), and the server 200 can provide the method provided in the embodiments of this application to one or more terminals.

[0064] The terminal 100 can be installed with a cash operation monitoring application. The application and the webpage can provide an interface. The terminal 100 can receive parameters input by a user on the cash operation monitoring interface and send the parameters to the server 200. The server 200 can obtain a processing result based on the received parameters and return the processing result to the terminal 100.

[0065] It should be understood that, in some optional implementations, the terminal 100 can also complete the action of obtaining a processing result based on received parameters by itself without the cooperation of the server. The embodiments of the present application are not limited.

[0066] Next, the product form of the terminal 100 is described. Figure 1

[0067] The terminal 100 in the embodiments of the present application can be a mobile phone, a tablet computer, a wearable device, a vehicle-mounted device, an augmented reality (AR) / virtual reality (VR) device, a notebook computer, an ultra-mobile personal computer (UMPC), a netbook, a personal digital assistant (PDA), and the like. The embodiments of the present application are not limited in this regard.

[0068] Figure 2 An optional hardware structure schematic diagram of the terminal 100 is shown.

[0069] Referring to FIG. 1, Figure 2 As shown in FIG. 1, the terminal 100 can include a radio frequency unit 110, a memory 120, an input unit 130, a display unit 140, a camera 150 (optional), an audio circuit 160 (optional), a speaker 161 (optional), a microphone 162 (optional), a headphone jack 163 (optional), a processor 170, an external interface 180, a power supply 190, and the like. Those skilled in the art can understand that Figure 2 The terminal or the multifunctional device is only an example and does not constitute a limitation on the terminal or the multifunctional device, which can include more or fewer components than those shown, or combine some components, or different components.

[0070] ​The input unit 130 can be used to receive inputted digital or character information, and to generate key signal inputs related to user settings of the portable multifunctional device and control of functions. Specifically, the input unit 130 can include a touch screen 131 (optional) and / or other input devices 132. The touch screen 131 can collect touch operations of a user thereon or thereabout (such as operations of the user using a finger, a knuckle, a stylus, or any suitable object on or near the touch screen), and drive corresponding connected devices according to pre-set programs. The touch screen can detect touch actions of the user on the touch screen, convert the touch actions into touch signals and send the touch signals to the processor 170, and can receive commands from the processor 170 and execute the commands; the touch signals at least include touch point coordinate information. The touch screen 131 can provide an input interface and an output interface between the terminal 100 and the user. In addition, the touch screen can be implemented in various types such as resistive, capacitive, infrared, and surface acoustic wave. In addition to the touch screen 131, the input unit 130 can also include other input devices. Specifically, the other input devices 132 can include, but are not limited to, one or more of a physical keyboard, function keys (such as volume control keys, on-off keys, etc.), trackballs, mice, joysticks, etc.

[0071] The input device 132 can receive inputted data, etc.

[0072] The display unit 140 can be used to display information inputted by the user or provided to the user, various menus of the terminal 100, interactive interfaces, file display, and / or playing of any kind of multimedia files. In the embodiments of the present application, the display unit 140 can be used to display interfaces of monitoring of cash operations, processing results, etc.

[0073] The storage 120 can be used to store instructions and data. The storage 120 can mainly include a storage instruction area and a storage data area. The storage data area can store various data such as multimedia files, texts, etc.; the storage instruction area can store software units such as operating systems, applications, instructions required by at least one function, etc., or their subsets, expanded sets. It can also include a non-volatile random access memory; provide the processor 170 with software and applications that include management of hardware, software, and data resources in the computing processing device, support control. It is also used for storage of multimedia files, and storage of running programs and applications.

[0074] The processor 170 is the control center of the terminal 100, connects each part of the whole terminal 100 by various interfaces and lines, executes various functions of the terminal 100 and processes data by running or executing the instructions stored in the memory 120 and calling the data stored in the memory 120, thereby performing overall control on the terminal device. Optionally, the processor 170 can include one or more processing units; preferably, the processor 170 can integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user interface and application program, etc., and the modem processor mainly processes wireless communication. It can be understood that the above-mentioned modem processor can also not be integrated into the processor 170. In some embodiments, the processor, the memory, can be implemented on a single chip, and in some embodiments, they can also be implemented on separate chips respectively. The processor 170 can also be used to generate corresponding operation control signals to send to corresponding components of the computing processing device, read and process data in the software, especially read and process data and programs in the memory 120, so that each functional module therein executes corresponding functions, thereby controlling the corresponding components to act according to the requirements of the instructions.

[0075] The memory 120 can be used to store software codes related to the monitoring method of the cash operation, and the processor 170 can execute the steps of the monitoring method of the cash operation, and can also dispatch other units (such as the above-mentioned input unit 130 and the display unit 140) to realize corresponding functions.

[0076] The RF unit 110 (optional) can be used for transmitting and receiving signals in information or communication processes, for example, receiving downlink information from a base station and sending uplink data to the base station. Generally, the RF circuit includes, but is not limited to, an antenna, at least one amplifier, a transceiver, a coupler, a low noise amplifier (LNA), a duplexer, etc. In addition, the RF unit 110 can also communicate with network devices and other devices through wireless communication. The wireless communication can use any communication standard or protocol, including but not limited to Global System for Mobile Communication (GSM), General Packet Radio Service (GPRS), Code Division Multiple Access (CDMA), Wideband Code Division Multiple Access (WCDMA), Long Term Evolution (LTE), email, Short Messaging Service (SMS), etc.

[0077] In the embodiments of the present application, the RF unit 110 can send data to the server 200 and receive the processing result sent by the server 200.

[0078] It should be understood that the RF unit 110 is optional, which can be replaced by other communication interfaces, for example, a network interface.

[0079] The terminal 100 also includes a power supply 190 (such as a battery) for supplying power to each component. Preferably, the power supply can be logically connected to the processor 170 through a power management system, so as to realize the functions of power management, such as charge management, discharge management and power consumption management, through the power management system.

[0080] The terminal 100 also includes an external interface 180, which can be a standard Micro USB interface or a multi-pin connector, and can be used for connecting the terminal 100 with other devices for communication, or for connecting a charger to charge the terminal 100.

[0081] Although not shown, the terminal 100 can also include a flash, a wireless fidelity (WiFi) module, a Bluetooth module, different function sensors, etc., which will not be described here. Some or all of the methods described below can be applied in the terminal 100 as shown. Figure 2

[0082] Next, the following describes​Figure 1 Product form of the server 200;

[0083] Figure 3 A structural diagram of the server 200 is provided, as shown in the figure, the server 200 includes a bus 201, a processor 202, a communication interface 203 and a memory 204. The processor 202, the memory 204 and the communication interface 203 communicate through the bus 201. Figure 3

[0084] The bus 201 can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, Figure 3 In the figure, only one thick line is used, but it does not mean that there is only one bus or one type of bus.

[0085] The processor 202 can be any one or more of a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor (MP), or a digital signal processor (DSP).

[0086] The memory 204 can include a volatile memory, such as a random access memory (RAM). The memory 204 can also include a non-volatile memory, such as a read-only memory (ROM), a flash memory, a mechanical hard drive (HDD) or a solid state drive (SSD).

[0087] Among them, the memory 204 can be used to store software code related to the monitoring method of the cash operation, the processor 202 can execute the steps of the chip monitoring method of the cash operation, and can also dispatch other units to realize the corresponding function.

[0088] ​It should be understood that the terminal 100 and the server 200 described above can be centralized or distributed devices, and the processors (for example, the processor 170 and the processor 202) in the terminal 100 and the server 200 can be hardware circuits (for example, an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a general-purpose processor, a digital signal processing (DSP), a microprocessor, a microcontroller, or the like) or a combination of the hardware circuits. For example, the processor can be a hardware system with an instruction execution function, such as a CPU, a DSP, or the like, or a hardware system without an instruction execution function, such as an ASIC, an FPGA, or the like, or a combination of the hardware system without an instruction execution function and the hardware system with an instruction execution function.

[0089] Figure 4 A specific structure diagram of a distributed cash operation monitoring system provided by an embodiment of the present application is provided. The monitoring system includes a camera, a model detection module, an algorithm reasoning module, an edge management module, and a model optimization module. The model detection module, the algorithm reasoning module, and the edge management module are deployed on a terminal device at an edge side, the terminal device is connected to a camera pre-deployed in a cash operation area in a vault, and the model optimization module is deployed on a server.

[0090] The camera is configured to capture a pre-configured cash operation area in real time and transmit a formed monitoring stream to the model detection module in the edge computing box.

[0091] The model detection module is configured to call a risk identification model deployed in the edge computing box to detect cash operation images in the vault in real time, obtain a risk identification result, and send the risk identification result to the algorithm reasoning module. The risk identification model includes a cash identification model and a personnel identification model, and the risk identification result includes a cash identification result output by the cash identification model and a personnel identification result output by the personnel identification model.

[0092] The algorithm reasoning module is configured to receive the risk identification result and analyze whether to send an alarm to the edge management module according to the risk identification result.

[0093] The edge management module is used to configure the camera's data stream and detection area during deployment. It also displays alarm information and images, and, based on image labeling rules, marks whether alarm images are valid and whether image detection results are correct, and corrects these errors. Furthermore, upon reaching the upload timeframe, the edge management module sends labeled image information to the model optimization module. This labeled image information includes alarm images and labels, with the labels indicating whether the image detection results are correct.

[0094] The model optimization module is used for model training and optimization. The model optimization process includes receiving labeled image information from the edge management module, performing model hyperparameter search and training, selecting the best-performing model hyperparameters and sending them to the model detection module for model hyperparameter update.

[0095] This application provides a distributed cash operation monitoring system. Through various functional modules deployed on edge-side terminal devices, it achieves real-time monitoring of cash operations and identifies potential risks in cash operation images, i.e., behaviors that do not comply with cash management regulations, providing timely alerts for risky behaviors. Functional modules deployed on the server side perform model hyperparameter search, training, and optimization, and distribute the model hyperparameter set and model parameters to each terminal device. Thus, it improves the real-time performance of risk identification in cash operations within the vault using a risk identification model, enhances the accuracy of risk identification by training and optimizing model parameters using the server's high computing power, and improves data security by leveraging the edge computing capabilities of each terminal device.

[0096] Furthermore, the model detection module enhances the recognizability of smaller items such as banknotes through data augmentation methods such as local contrast enhancement and image edge sharpening, thereby improving the accuracy and robustness of the risk identification model's recognition results.

[0097] Furthermore, the edge management module obtains sample data for optimizing the risk identification model by marking whether the alarm image is a valid image and whether the image detection result is correct. The model optimization module automatically optimizes the risk identification model based on the marked image information, completing the continuous iterative update of the model and improving the accuracy of the risk identification results of the risk identification model.

[0098] Specifically, in the monitoring scenario of cash operation, the model detection module of the edge computing box decodes, pre-processes, and model detects the received real-time video stream, detects the cash operation scene in the vault in real time, obtains the banknote item information and the personnel category information, and finally sends the inference result to the algorithm inference module. The algorithm inference module receives the result of model detection, and analyzes and judges whether the specified cash operation personnel in the cash operation area appears less than the specified number of cash according to the result. If it is less than the specified personnel, an alarm information is sent to the edge management module. The edge management module can display the alarm information and the corresponding image in real time, and send these alarm information to the relevant personnel through the network channel to remind. At the same time, the staff can also mark whether these alarm images are valid images, and whether the image detection result is correct and correct. The module periodically sends the above marked images to the model optimization module. The actual alarm data collected periodically constitutes a data set imported into the model optimization module, and the model hyperparameter search training is performed. The best performing model is periodically selected and sent to the model detection module for model replacement, and the detection model is continuously optimized autonomously.

[0099] Figure 5 The specific implementation flowchart of the cash operation monitoring method provided by the embodiment of the application is shown in Figure 5 The method specifically includes the following steps.

[0100] S501, collect original data.

[0101] In this embodiment, the original data includes vault scene data and public resource data, the vault scene data includes a plurality of vault cash operation region background images, a plurality of cash images in the vault cash operation scene, and a plurality of vault staff images, and the public resource data includes personnel data, cash data and negative sample data.

[0102] S502, data processing is performed on the original data to obtain a training data set.

[0103] In this embodiment, the process of data processing on the original data includes:

[0104] 1. Preprocessing the data, wherein the data preprocessing includes de-duplication, standardization, and screening, and the like.

[0105] 2. Data labeling is performed on the preprocessed data. Wherein, the personnel type and personnel information in each image in the data are labeled, wherein the personnel information includes the number of personnel and the type of personnel, and the cash in each image in the data is labeled, and the cash labeling includes whether there is cash in the image.

[0106] 3. Perform a data augmentation operation on the labeled data to obtain a training data set, the training data set including a cash recognition training data set and a personnel recognition training data set, wherein the cash recognition training data set includes sample images and cash labels, and the personnel recognition training data set includes sample images and personnel information labels. The data augmentation includes preset enhancement operations such as image splicing, scaling, color transformation, and cropping.

[0107] S503. Perform image enhancement on the images in the training data set.

[0108] In this embodiment, the image enhancement method includes contrast enhancement and image edge sharpening.

[0109] Specifically, the contrast enhancement method includes:

[0110] The monitoring image is adaptively segmented into multiple local regions, and the contrast of the pixels inside each local region is recalculated. The enhancement parameter is determined according to the contrast of the local region, and the local region with lower contrast is enhanced more strongly. Each local region is subjected to contrast enhancement processing according to the enhancement parameter, and is merged into a complete image.

[0111] The image edge sharpening method includes:

[0112] The Laplace operator is applied to identify the edge map of the monitoring image and continuously adjust the parameters to achieve the best effect, and finally the edge map of the image is directly superimposed on the monitoring image.

[0113] S504. Train a machine learning model based on the training data set through a pre-built model automatic training platform to obtain a cash recognition model and a personnel recognition model, and deploy the cash recognition model and the personnel recognition model to an edge computing box.

[0114] In this embodiment, the cash recognition model is used to output a cash recognition result, and the cash recognition result is used to indicate whether there is cash in the image. The personnel recognition model outputs a personnel recognition result, and the personnel recognition result includes the type and number of personnel in the image.

[0115] In this embodiment, the cash recognition model is trained based on the cash recognition training data set until the training completion condition is reached, the parameters of the cash recognition model are obtained, and the cash recognition model is deployed in the edge computing box based on the hyperparameter group and the parameters. The personnel recognition model is trained based on the personnel recognition training data set until the training completion condition is reached, the hyperparameters of the personnel recognition model are obtained, and the personnel recognition model is deployed in the edge computing box based on the hyperparameters. The values of each type of hyperparameter in the hyperparameter group of the cash recognition model and the personnel recognition model are obtained from the initial hyperparameters randomly selected from the preconfigured hyperparameter types and values,

[0116] In this embodiment, the edge computing box is configured in the terminal device at the edge side.

[0117] It should be noted that S501-S504 are executed by the server, the server can connect multiple bank systems, collect as much original data as possible, rely on powerful computing power to improve the quality of the original data, obtain training data with large data volume and high quality, and improve the accuracy of the risk identification model.

[0118] S505, real-time receiving monitoring image of target cash operation, performing local image enhancement on the target local region of the monitoring image.

[0119] In this embodiment, the monitoring image is collected and transmitted in real time by the camera configured in the cash operation area.

[0120] In this embodiment, the target local region is the specified region of the cash in the monitoring image when performing the target cash operation. Optionally, based on the specified execution region of the target cash operation and the shooting angle of the monitoring image, the target local region is predicted and extracted from the monitoring image.

[0121] For example, when the target cash operation is a dispensing operation, the target local region is the dispensing box region, and the dispensing box region is the region of the dispensing box in the monitoring image. It should be noted that the spatial positional relationship between the fixed dispensing box and the camera is fixed, so the dispensing box region can be determined by the shooting angle of the camera and other information. The recognition of the moving dispensing box can be realized based on image recognition. Since the dispensing box is relatively large and the features are not complex, the local image region to be enhanced can be determined based on the easily determined dispensing box region, without performing local image enhancement on each region of the monitoring image.

[0122] In this embodiment, the local image enhancement includes contrast enhancement and edge sharpening.

[0123] It should be noted that due to the reasons such as the far distance between the camera and the cash operation area, and the many subtle features of the cash as the target detection object, the difficulty of cash recognition is increased. Therefore, the monitoring image is subjected to local image enhancement of the target local region before the cash detection model is input.

[0124] The local contrast enhancement strategy selectively enhances the contrast of the banknote part, thereby reducing the interference of background information to recognition. The image edge sharpening enhances the edge part of the image, thereby making the edge of the banknote more clear and prominent, and greatly improving the recognizability of the banknote. Thus, the local contrast enhancement and image edge sharpening are used to strengthen the subtle features of the possible region, highlight the details of the cash, and improve the accuracy of cash detection. Through target local positioning, the complexity of local image enhancement is reduced, and the image processing speed is improved.

[0125] S506, input the enhanced monitoring image to the cash identification model and the personnel identification model in the edge computing box to obtain a cash identification result output by the cash identification model and a personnel identification result output by the personnel identification model.

[0126] In this embodiment, after inputting the enhanced monitoring image, i.e., the enhanced monitoring image, to the cash identification model in the edge computing box to obtain a cash identification result, if the cash identification result indicates that there is a cash image, the enhanced monitoring image is input to the personnel identification model in the edge computing box for personnel identification.

[0127] This step improves the model detection efficiency and reduces the consumption of model computing resources by configuring the identification order of the cash identification model and the personnel identification model, and starting the personnel identification model after the cash identification result indicates that the monitoring image region is a cash region.

[0128] S507, according to the cash identification result and the personnel identification result, determine whether the cash operation conforms to the pre-configured cash operation specification.

[0129] In this embodiment, the cash operation specification includes a corresponding safety operation rule of each cash operation, and the safety operation rule at least includes a specified number of personnel and a specified type of personnel in the cash region.

[0130] In this embodiment, the corresponding safety operation rule of the target cash operation is searched from the cash operation specification as a target safety operation rule, and whether the cash operation conforms to the target safety operation rule is determined according to the cash identification result and the personnel identification result. The target cash operation is the current cash operation, which can be determined by other monitoring systems or business systems.

[0131] S508, if not, an alarm information of the cash operation is sent.

[0132] In this embodiment, the alarm information of the cash operation includes an alarm image, a cash identification result, a personnel identification result, and a target safety operation rule, and the target safety operation rule is a safety operation rule that does not conform to the cash identification result and the personnel identification result. For example, in the cash dispensing scene, if the model detection result indicates that the number of personnel in the cash dispensing region is less than 2, the monitoring image is taken as the alarm image, and the alarm information of the cash dispensing is generated.

[0133] In this embodiment, the specific method of sending the alarm information of the cash operation includes: sending the alarm information to a pre-configured safety supervision client, including a short message client, an email client, and an application program client, etc., so that the supervisor can receive and display the alarm information in time through the safety supervision client to investigate the cash operation risk.

[0134] S509, acquire feedback information of the historical alarm information.

[0135] In this embodiment, the feedback information of the historical alarm information is acquired according to a preset period, and the feedback information of the historical alarm information includes whether the cash recognition result and the personnel recognition result are correct, and the corrected cash recognition result and the corrected personnel recognition result.

[0136] The alarm information generated in actual application is collected periodically, and the staff can mark whether the alarm image is a valid image and whether the image detection result needs to be corrected. Then the above marked image is periodically sent to the model optimization part for optimization iteration of the model, and the accuracy and robustness of the algorithm are continuously enhanced.

[0137] S510, label each alarm image based on the feedback information of the historical alarm information to obtain a retraining data set.

[0138] In this embodiment, the retraining data set includes a cash recognition retraining data set and a personnel recognition retraining data set, the cash recognition retraining data set at least includes a first type of alarm image and a cash label, wherein the first type of alarm image is an alarm image with an incorrect cash recognition result, and the cash label of the first type of alarm image is a corrected cash recognition result, and the personnel recognition retraining data set at least includes a second type of alarm image and a personnel label, wherein the second type of alarm image is an alarm image with an incorrect personnel recognition result, and the personnel label of the second type of alarm image is a corrected personnel recognition result.

[0139] It should be noted that S505-S510 are executed by a terminal device at the edge, and the terminal device automatically identifies personnel information and cash regions in the monitoring image through a risk identification model deployed at the edge, avoiding the risk of image sensitive information leakage, and improving the real-time performance of risk monitoring and reducing the resource consumption of the server.

[0140] S511, when the retraining opportunity is reached, a pre-built model automatic training platform is used to retrain the risk identification model based on the retraining data set, so as to optimize the risk identification model.

[0141] In this embodiment, the retraining opportunity includes that the performance evaluation result of at least one risk detection model does not meet the model performance condition, or a preset retraining interval length is reached.

[0142] In this embodiment, the cash recognition model is retrained based on the cash recognition retraining data set until the retraining completion condition is reached, the optimized parameters of the cash recognition model are obtained, and the cash recognition model is updated in the edge computing box based on the optimized parameters. The cash recognition model is retrained based on the personnel recognition retraining data set until the retraining completion condition is reached, the optimized parameters of the personnel recognition model are obtained, and the personnel recognition model is updated in the edge computing box based on the optimized parameters.

[0143] In this embodiment, the automatic training platform has a hyperparameter search training function and a received data autonomous training function. The specific method of retraining the machine learning model based on the retraining data set through the pre-built model automatic training platform includes:

[0144] The preconfigured hyperparameter search information is obtained, and the hyperparameter search information includes a preset type and a corresponding preset value range.

[0145] After receiving the training start instruction, if the training frequency does not reach the preset frequency threshold, the risk identification model is configured using the preset hyperparameter group, and the risk identification model is trained based on the retraining data set. After the training completion condition is reached, the trained risk identification model and its parameters are obtained.

[0146] The preset hyperparameter group includes preset values of hyperparameters of multiple preset types. The preset hyperparameter group is obtained by randomly selecting values in the hyperparameter search information.

[0147] If the training frequency reaches the frequency threshold, the target hyperparameter group is obtained according to the hyperparameter search result, the risk identification model is configured using the target hyperparameter group, and the risk identification model is trained based on the retraining data set. After the training completion condition is reached, the trained risk identification model and its parameters are obtained.

[0148] The target hyperparameter group includes target values of hyperparameters of multiple target types.

[0149] It should be noted that hyperparameters are parameters that need to be manually set before model training, and they have a crucial impact on the final performance of the model. These parameters include but are not limited to learning rate, regularization strength, network level structure, training batch size, training frequency, etc. By adjusting these hyperparameters through hyperparameter search, the accuracy and generalization ability of the model can be effectively improved.

[0150] The goal of hyperparameter search is to find the optimal hyperparameter group to minimize the error of the model or maximize the performance of the model. Hyperparameter search analyzes the degree of influence of different types of hyperparameters on the model results and analyzes whether the value of the hyperparameter has a positive or negative correlation with the model. The range of the specified hyperparameters is then selected during configuration, and then training is performed to ultimately obtain the best model.

[0151] It should be noted that S511 can be implemented by the server, and after the terminal device reaches the retraining opportunity, the retraining instruction and the retraining data set are sent to the server, the server re-trains the risk identification model configured in the terminal device based on at least the retraining data set of the terminal device, and the updated parameters and the to-be-updated hyperparameter group are issued, and the terminal device updates the parameters and the hyperparameters to realize the optimization of the risk identification model.

[0152] From the above technical solutions, it can be seen that the cash operation monitoring method provided by the embodiments of the present application can realize real-time risk identification of the cash operation monitoring image through the risk identification model deployed in the terminal device, and timely alarm for risk behavior, thereby improving the real-time performance and accuracy of the risk identification of the cash operation. The server searches for the model hyperparameters, trains and optimizes the model parameters, and issues the model hyperparameter group and the model parameters to each terminal device. The high computing power of the server is used to train and optimize the model parameters, improve the identification accuracy of the risk identification model, and use the edge computing capability of each terminal device to improve the data security.

[0153] Further, through the local contrast enhancement and image edge sharpening of the target local area, the cash recognition is improved, the complexity of local image enhancement is reduced, the recognition speed is improved, and the accuracy and robustness of the identification result of the risk identification model are improved.

[0154] Further, the terminal device marks whether the alarm image is a valid image and whether the image detection result is correct, obtains a retraining data set for optimizing the local risk identification model, so that the server re-trains the risk identification model corresponding to the terminal device based on the retraining data set, improves the matching degree of the risk identification model and the corresponding cash vault environment of the terminal device, and improves the accuracy of the risk identification model.

[0155] The present scheme realizes real-time monitoring and real-time alarm through real-time risk identification of the monitoring image, can significantly reduce the risk control risk and labor cost of manual supervision in this scene, and realizes real-time alarm to reduce time loss and solve the problem of hysteresis. The model automatic training platform is used to periodically train the real data collected in actual application, filter out the best-performing model, and perform self-updating iteration of the model, thereby further improving the accuracy of risk identification.

[0156] It should be noted that the above is only an optional specific implementation process of the cash operation monitoring method provided in the embodiment, and the present application can also be implemented by other specific implementations. For example, the hyperparameter search method includes but is not limited to a variety of optional search methods such as grid search, random search, and Bayesian optimization search. For another example, another implementation method of S505 is to receive the monitoring image in real time, input the monitoring image into the cash identification model and the personnel identification model in the edge computing box at the same time, and obtain the cash identification result output by the cash identification model and the personnel identification result output by the personnel identification model. For another example, S511 can be implemented by a server or automatically implemented by a terminal device. In another optional embodiment, the terminal device locally implements retraining of the risk identification model and completes the risk identification model optimization by itself. The local implementation of targeted model optimization makes the risk identification model have high matching degree with the current vault monitoring scene.

[0157] In summary, Figure 6 A cash operation monitoring method provided in the embodiment of the present application, as shown in Figure 6 The method comprises the following steps:

[0158] S601, a monitoring image of a target cash operation is received in real time, a local image enhancement is performed on a target local region of the monitoring image, and an enhanced monitoring image is obtained.

[0159] In the embodiment, the monitoring image is collected and transmitted in real time by a camera arranged in a cash operation region.

[0160] In the embodiment, the target local region is a specified region of the cash in the monitoring image when the target cash operation is performed. Optionally, the target local region is predicted based on a specified execution region of the target cash operation and a shooting angle of the monitoring image, and the target local region is extracted from the monitoring image.

[0161] S602, the enhanced monitoring image is input into a pre-configured risk identification model, and a risk identification result output by the risk identification model is obtained.

[0162] In the embodiment, the risk identification model includes a cash identification model and a personnel identification model, and the risk identification result includes a cash identification result output by the cash identification model and a personnel identification result output by the personnel identification model.

[0163] In the embodiment, after the enhanced monitoring image is input into the cash identification model in the edge computing box to obtain the cash identification result, if the cash identification result indicates that there is a cash image, the enhanced monitoring image is input into the personnel identification model in the edge computing box for personnel identification.

[0164] S603, a safe operation rule of the target cash operation is obtained.

[0165] In this embodiment, the safety operation rule at least includes the specified personnel, the number and the specified personnel type in the cash area.

[0166] S604, judging whether the target cash operation conforms to the safety operation rule according to the risk identification result.

[0167] S605, generating an alarm information if the target cash operation does not conform to the safety operation rule.

[0168] In this embodiment, the alarm information includes an alarm image, the risk identification result and the safety operation rule. The alarm image is a monitoring image corresponding to the target cash operation not conforming to the safety operation rule.

[0169] As can be seen from the above technical solution, the present application provides a kind of cash operation monitoring method, the risk identification model of deployment in terminal equipment is in real time to the risk identification of cash operation monitoring image, and the risk behavior of not conforming to safety operation rule is timely alarmed, improve the real-time of cash operation risk identification, wherein, through the local image enhancement of target local area, the cash identifiable is targetedly improved, the complexity of local image enhancement is reduced, the identification speed is improved, the accuracy and robustness of the identification result of risk identification model are improved.

[0170] It should be noted that the present application provides a kind of cash operation monitoring method and related device can be used in artificial intelligence field or financial field. The above is only an example, and does not limit the application field of the present application to a kind of cash operation monitoring method and related device.

[0171] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0172] The above introduces a kind of cash operation monitoring method provided by the present application, and the following will introduce the device for executing the cash operation monitoring method described above.

[0173] Please refer to Figure 7 , Figure 7 The structure diagram of a kind of cash operation monitoring device provided by the present application embodiment. As Figure 7 shown, the cash operation monitoring device 700 includes:

[0174] The local enhancement unit 701 is configured to receive a monitoring image of a target cash operation in real time, perform local image enhancement on a target local region of the monitoring image to obtain an enhanced monitoring image, and the target local region is a specified region of the cash in the monitoring image when the target cash operation is performed.

[0175] The risk identification unit 702 is configured to input the enhanced monitoring image into a pre-configured risk identification model to obtain a risk identification result output by the risk identification model, the risk identification model includes a cash identification model and a personnel identification model, and the risk identification result includes a cash identification result output by the cash identification model and a personnel identification result output by the personnel identification model.

[0176] The operation rule determination unit 703 is configured to obtain a safe operation rule of the target cash operation, and the safe operation rule at least includes a specified number of personnel and a specified type of personnel in a cash region.

[0177] The risk judgment unit 704 is configured to judge whether the target cash operation conforms to the safe operation rule according to the risk identification result.

[0178] The risk alarm unit 705 is configured to generate an alarm information if the target cash operation does not conform to the safe operation rule, and the alarm information includes an alarm image, the risk identification result and the safe operation rule.

[0179] In a possible implementation, when the local enhancement unit is configured to perform local image enhancement on a target local region of the monitoring image to obtain an enhanced monitoring image, the local enhancement unit is specifically configured to:

[0180] extract the target local region from the monitoring image based on a specified execution region of the target cash operation and a shooting angle of the monitoring image;

[0181] perform contrast enhancement and edge sharpening on the target local region to obtain the enhanced monitoring image.

[0182] In a possible implementation, when the risk identification unit is configured to input the enhanced monitoring image into a pre-configured risk identification model to obtain a risk identification result output by the risk identification model, the risk identification unit is specifically configured to:

[0183] after the cash identification result indicates that there is a cash image, input the enhanced monitoring image into a personnel identification model to obtain a personnel identification result.

[0184] In a possible implementation, the monitoring device of the cash operation further includes a retraining unit, configured to:

[0185] obtain feedback information of historical alarm information, the feedback information of the historical alarm information including whether the cash recognition result is correct and whether the personnel recognition result is correct, and the corrected cash recognition result and the corrected personnel recognition result

[0186] label the alarm image based on the feedback information of the alarm information to obtain a retraining data set;

[0187] The retraining data set includes a cash recognition retraining data set and a personnel recognition retraining data set. The cash recognition retraining data set at least includes a first type of alarm image and cash labeling. The first type of alarm image is an alarm image with an incorrect cash recognition result. The cash labeling of the first type of alarm image is a corrected cash recognition result. The personnel recognition retraining data set at least includes a second type of alarm image and personnel labeling. The second type of alarm image is an alarm image with an incorrect personnel recognition result. The personnel labeling of the second type of alarm image is a corrected personnel recognition result.

[0188] When a preset retraining opportunity is reached, the risk recognition model is retrained based on the retraining data set to obtain retraining parameter information. Alternatively, the retraining data set and a retraining instruction are sent to a server to enable the server to retrain the risk recognition model based on the retraining data set and return retraining parameter information. The retraining parameter information includes a hyperparameter group and model parameters.

[0189] The risk recognition model is optimized based on the retraining parameter information.

[0190] In a possible implementation, when the retraining unit is used to retrain the risk recognition model based on the retraining data set, the retraining unit is specifically configured to:

[0191] obtain preconfigured hyperparameter search information, the hyperparameter search information including a preset category and a corresponding preset value range;

[0192] If the number of training times does not reach a preset number threshold, the risk recognition model is configured using the preset hyperparameter group, and the risk recognition model is retrained based on the retraining data set. After a preset training completion condition is reached, retraining parameter information is obtained.

[0193] The preset hyperparameter group includes preset values of hyperparameters of multiple preset categories obtained by random selection in the hyperparameter search information.

[0194] If the number of training times reaches the number threshold, a target hyperparameter group is obtained according to a hyperparameter search result, the risk identification model is configured using the target hyperparameter group, and the risk identification model is trained based on the retraining data set, and after the training completion condition is reached, retraining parameter information is obtained; wherein the target hyperparameter group includes target values of hyperparameters of multiple target categories.

[0195] In a possible implementation, the monitoring device of the cash operation further includes a model deployment unit, configured to, before inputting the enhanced monitoring image into the pre-configured risk identification model:

[0196] The receiving server trains a machine learning model based on a training data set to obtain initial configuration information of a cash identification model and a personnel identification model, and deploys the cash identification model and the personnel identification model based on the initial configuration information, wherein the initial configuration information includes an initial hyperparameter group and initial model parameters.

[0197] The embodiments of the present application also provide an electronic device. Referring to Figure 8 The embodiments of the present application also provide an electronic device. Referring to Figure 8 The electronic device shown is only an example, and should not impose any limitation on the functions and use range of the embodiments of the present application.

[0198] As Figure 8 shown, the electronic device can include a processing device (such as a central processing unit, a graphics processing unit, etc.) 801, which can perform various appropriate actions and processes according to programs stored in a read-only memory (ROM) 802 or loaded from a storage device 808 into a random access memory (RAM) 803. In the state that the electronic device is powered on, the RAM 803 also stores various programs and data required for the operation of the electronic device. The processing device 801, the ROM 802, and the RAM 803 are connected to each other through a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0199] Generally, the following devices can be connected to the I / O interface 805: input devices 806 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 807 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 808 including, for example, a memory card, a hard disk, etc.; and communication devices 809. The communication devices 809 can allow the electronic device to communicate with other devices wirelessly or by wire to exchange data. Although Figure 6Electronic devices having various apparatuses are shown, but it is understood that all of the illustrated apparatuses are not required, and that any one or combination of apparatuses can be implemented.

[0200] The embodiment of the present application further provides a computer program product comprising computer readable instructions, which, when executed on an electronic device, cause the electronic device to implement any one of the cash operation monitoring methods provided by the embodiments of the present application.

[0201] The embodiment of the present application further provides a computer readable storage medium, which carries one or more computer programs, and when the one or more computer programs are executed by an electronic device, the electronic device can implement any one of the cash operation monitoring methods provided by the embodiments of the present application.

[0202] In addition, it should be noted that the apparatus embodiments described above are merely illustrative, and the units described as separate components can or can not be physically separated, and the components displayed as units can or can not be physical units, that is, they can be located in one place, or distributed on multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiment. In addition, the connection relationship between the modules in the apparatus embodiment provided by the present application indicates that there is a communication connection between them, which can be implemented as one or more communication buses or signal lines.

[0203] Through the description of the above embodiments, those skilled in the art can clearly understand that the present application can be implemented by means of software and necessary general hardware, and of course, it can also be implemented by special hardware including special integrated circuits, special CPUs, special memories, special components, etc. Generally, functions completed by computer programs can be easily implemented by corresponding hardware, and the specific hardware structure for implementing the same function can also be various, such as analog circuits, digital circuits or special circuits. However, for the present application, software program implementation is a better embodiment. Based on this understanding, the technical solutions of the present application can be embodied in the form of a software product, which is stored in a readable storage medium, such as a computer's floppy disk, U disk, mobile hard disk, ROM, RAM, magnetic disk or optical disk, etc., including a plurality of instructions to make a computer device (which can be a personal computer, training device, or network device, etc.) execute the methods described in various embodiments of the present application.

[0204] In the above embodiments, all or part can be implemented by software, hardware, firmware or any combination thereof. When implemented by software, it can be implemented in the form of a computer program product in whole or in part.

[0205] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transmitted from one website, computer, training device or data center to another website, computer, training device or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can store or a data storage device such as a training device, a data center, etc. integrated with one or more available media sets. The available media can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium (for example, a solid state disk (SSD)), etc.

Claims

1. A method of monitoring cash operations, characterized by Applied to a terminal, comprising: Real-time receiving a monitoring image of a target cash operation, performing local image enhancement on a target local area of the monitoring image to obtain an enhanced monitoring image, the target local area being a specified area of the monitoring image in which cash is located when the target cash operation is performed; Inputting the enhanced monitoring image into a pre-configured risk identification model to obtain a risk identification result output by the risk identification model, the risk identification model comprising a cash identification model and a personnel identification model, and the risk identification result comprising a cash identification result output by the cash identification model and a personnel identification result output by the personnel identification model; Obtaining a safe operation rule of the target cash operation, the safe operation rule comprising at least a specified number of personnel and a specified type of personnel in the cash area; Judging whether the target cash operation conforms to the safe operation rule according to the risk identification result; If the target cash operation does not conform to the safe operation rule, generating an alarm information, the alarm information comprising an alarm image, the risk identification result, and the safe operation rule; Obtaining feedback information of historical alarm information, the feedback information comprising whether the cash identification result is correct and whether the personnel identification result is correct, and a corrected cash identification result and a corrected personnel identification result; Labeling the alarm image based on the feedback information of the alarm information to obtain a retraining data set; The retraining data set comprises a cash identification retraining data set and a personnel identification retraining data set, the cash identification retraining data set comprising at least a first type of alarm image and a cash label, the first type of alarm image being an alarm image with an incorrect cash identification result, and the cash label of the first type of alarm image being a corrected cash identification result, and the personnel identification retraining data set comprising at least a second type of alarm image and a personnel label, the second type of alarm image being an alarm image with an incorrect personnel identification result, and the personnel label of the second type of alarm image being a corrected personnel identification result; When a preset retraining opportunity is reached, retraining the risk identification model based on the retraining data set to obtain retraining parameter information, or sending the retraining data set and a retraining instruction to a server to enable the server to retrain the risk identification model based on the retraining data set and return retraining parameter information, the retraining parameter information comprising a hyperparameter group and a model parameter; Optimizing the risk identification model based on the retraining parameter information.

2. The cash-operated monitoring method according to claim 1, characterized by, Performing local image enhancement on a target local area of the monitoring image to obtain an enhanced monitoring image, comprising: Extracting the target local area from the monitoring image based on a specified execution area of the target cash operation and a shooting angle of the monitoring image; Performing contrast enhancement and edge sharpening on the target local area to obtain the enhanced monitoring image.

3. The cash-operated monitoring method according to claim 1, characterized by, The inputting the enhanced monitoring image into a pre-configured risk identification model to obtain a risk identification result output by the risk identification model, comprising: After the enhanced monitoring image is input into the cash identification model to obtain a cash identification result, if the cash identification result indicates that there is a cash image, the enhanced monitoring image is input into a personnel identification model to obtain a personnel identification result.

4. The cash-operated monitoring method according to claim 1, characterized by, The retraining of the risk identification model based on the retraining data set comprises: obtaining preconfigured hyperparameter search information, the hyperparameter search information comprising a preset category and a corresponding preset value range; if the number of training times does not reach a preset number threshold, configuring the risk identification model using a preset hyperparameter group and retraining the risk identification model based on the retraining data set, and obtaining retraining parameter information after a preset training completion condition is reached; wherein the preset hyperparameter group comprises preset values of hyperparameters of multiple preset categories obtained by random selection in the hyperparameter search information; if the number of training times reaches the number threshold, obtaining a target hyperparameter group according to a hyperparameter search result, configuring the risk identification model using the target hyperparameter group, and training the risk identification model based on the retraining data set, and obtaining retraining parameter information after the training completion condition is reached; wherein the target hyperparameter group comprises target values of hyperparameters of multiple target categories.

5. The cash-operated monitoring method according to claim 1, characterized by, The cash operation monitoring method further comprises, before the enhanced monitoring image is input into the preconfigured risk identification model: receiving initial configuration information of a cash identification model and a personnel identification model obtained by a server based on training of a machine learning model using a training data set, and deploying the cash identification model and the personnel identification model based on the initial configuration information, the initial configuration information comprising an initial hyperparameter group and initial model parameters.

6. A cash handling device, characterized in that comprises: a local enhancement unit configured to receive a monitoring image of a target cash operation in real time, perform local image enhancement on a target local region of the monitoring image to obtain an enhanced monitoring image, and the target local region is a specified region of cash in the monitoring image when the target cash operation is performed; a risk identification unit configured to input the enhanced monitoring image into a preconfigured risk identification model to obtain a risk identification result output by the risk identification model, the risk identification model comprising a cash identification model and a personnel identification model, and the risk identification result comprising a cash identification result output by the cash identification model and a personnel identification result output by the personnel identification model; an operation rule determination unit configured to obtain a safe operation rule of the target cash operation, the safe operation rule comprising at least a specified number of personnel in a cash region and a specified type of personnel; a risk determination unit configured to determine whether the target cash operation complies with the safe operation rule according to the risk identification result; a risk warning unit configured to generate warning information if the target cash operation does not comply with the safe operation rule, the warning information comprising a warning image, the risk identification result, and the safe operation rule; a retraining unit configured to: obtain feedback information of historical alarm information, the feedback information of the historical alarm information including whether the cash recognition result is correct and whether the personnel recognition result is correct, and the corrected cash recognition result and the corrected personnel recognition result, label the alarm image based on the feedback information of the alarm information to obtain a retraining data set; The retraining data set includes a cash recognition retraining data set and a personnel recognition retraining data set. The cash recognition retraining data set includes at least a first type of alarm image and a cash label. The first type of alarm image is an alarm image with an incorrect cash recognition result. The cash label of the first type of alarm image is a corrected cash recognition result. The personnel recognition retraining data set includes at least a second type of alarm image and a personnel label. The second type of alarm image is an alarm image with an incorrect personnel recognition result. The personnel label of the second type of alarm image is a corrected personnel recognition result. When a preset retraining opportunity is reached, the risk recognition model is retrained based on the retraining data set to obtain retraining parameter information; or the retraining data set and a retraining instruction are sent to a server to enable the server to retrain the risk recognition model based on the retraining data set and return retraining parameter information. The retraining parameter information includes hyperparameters and model parameters. The risk recognition model is optimized based on the retraining parameter information.

7. A computer program product, characterised in that, The computer readable instructions, when executed on an electronic device, enable the electronic device to implement the cash operation monitoring method of any one of claims 1 to 5.

8. An electronic device, comprising: The memory is configured to store computer programs. The processor is configured to execute the computer programs to enable the electronic device to implement the cash operation monitoring method of any one of claims 1 to 5. The storage medium carries one or more computer programs, which, when executed by an electronic device, enable the electronic device to implement the cash operation monitoring method of any one of claims 1 to 5.

9. A computer storage medium, characterized in that ​

Images