Unattended management method and device for smart stores based on scene adaptation

Abstract

The embodiments of the present application provide a method and device for unmanned management of smart stores based on scene adaptation, which collects store data through multiple cameras and sensors, and extracts customer behavior characteristics, product placement characteristics and environmental parameters. Innovatively construct scene recognition models and behavior analysis models to achieve intelligent recognition of scene states and real-time monitoring of abnormal behaviors. Dynamically match management strategies based on scene feature vectors, and formulate detailed management sub-strategies based on time periods, customer flow and environmental conditions. At the same time, analyze operational data through adaptive optimization models to achieve dynamic updating and optimization of management strategies. This method breaks through the limitations of traditional fixed management models and provides a comprehensive technical solution for unmanned operations of smart stores.

Description

Technical Field

[0001] The present application relates to the field of data processing, and specifically to a scene-adaptive unmanned management method and device for smart stores. Background Art

[0002] Traditional retail store management mainly relies on manual operation, which not only has high labor costs but also low management efficiency. With the development of unmanned retail, existing automated management systems often adopt fixed management strategies and lack the ability to adapt to different scenarios. In particular, in terms of customer flow fluctuations and emergency handling, the existing system is difficult to respond in a timely and effective manner.

[0003] At the same time, the existing store management system has obvious deficiencies in the identification and processing of abnormal behaviors. Traditional video surveillance systems are mainly used for passive recording, lack active identification and early warning mechanisms, and cannot effectively prevent and promptly handle abnormal events. The system also fails to fully utilize the fusion analysis of multi-source data, making it difficult to achieve comprehensive scene perception and intelligent decision-making.

[0004] In addition, the existing system is also relatively weak in operational data analysis and strategy optimization. The lack of an adaptive strategy optimization mechanism makes it impossible to dynamically adjust and optimize management strategies based on historical operational data, which affects the continuous improvement of management results. Solving these problems is of great significance to achieving efficient operation and safe management of smart stores. Summary of the invention

[0005] In response to the problems in the prior art, the present application provides a scenario-adaptive unmanned management method and device for smart stores, which can effectively improve the unmanned operation efficiency and reliability of smart stores.

[0006] In order to solve at least one of the above problems, the present application provides the following technical solutions:

[0007] In a first aspect, the present application provides a scenario-adaptive smart store unattended management method, comprising:

[0008] Collect video data and sensor data from multiple cameras in the store, perform target detection on the video data to extract customer behavior features and product placement features, perform feature extraction on the sensor data to obtain store environment parameters, input the customer behavior features, product placement features and store environment parameters into a pre-trained scene recognition model to obtain a store scene status label, perform semantic analysis on the store scene status label to obtain the main feature parameters of the current scene, construct a scene feature vector based on the main feature parameters, perform similarity calculation on the scene feature vector and the scene description vector in a preset scene management strategy library to select the management strategy with the highest similarity, match the time period information, customer flow information and environmental status information in the scene status label with the preset time period rule library, customer flow threshold rule library and environmental parameter rule library respectively to obtain corresponding time period management sub-strategies, customer flow management sub-strategies and environmental management sub-strategies, and combine the sub-strategies to form a complete scene management strategy;

[0009] Constructing a customer behavior time series sequence based on the customer behavior characteristics, inputting the customer behavior time series sequence into a pre-trained behavior analysis model to determine whether there is abnormal behavior, and recording the abnormal behavior type, occurrence time and location information into an abnormal event list when abnormal behavior is detected, triggering a corresponding security warning mechanism according to the danger level of the abnormal behavior in the abnormal event list, and calling a corresponding emergency response strategy from a preset emergency response solution library according to the store scene status label and the abnormal behavior type;

[0010] A statistical analysis of the store scene status label and the customer behavior characteristics in the time dimension is performed to obtain store operation data, and the store operation data is input into an adaptive optimization model to generate store management strategy optimization suggestions. The management strategies in the preset scene management strategy library are dynamically updated according to the management strategy optimization suggestions. At the same time, the store operation data, the abnormal event list and the management strategy optimization suggestions are sent to a cloud management platform for storage. The cloud management platform generates a store operation analysis report based on the received data.

[0011] Furthermore, the customer behavior characteristics, the product placement characteristics and the store environment parameters are input into a pre-trained scene recognition model to obtain a store scene status label, and the store scene status label is semantically analyzed to obtain the main characteristic parameters of the current scene, including:

[0012] The customer behavior features, the product placement features and the store environment parameters are feature-fused in a preset ratio to construct a multimodal input matrix, the multimodal input matrix is ​​normalized, the normalized feature data is input into a scene recognition model based on a convolutional neural network and a long short-term memory network, and the store scene status label is output through the scene recognition model;

[0013] The text information in the store scene status label is semantically segmented and word vector mapped through a natural language processing algorithm, and the spatiotemporal dimension features, personnel flow features and environmental status features in the scene label are extracted. The features are weighted and calculated based on preset feature item weights to obtain the main feature parameters of the current scene.

[0014] Furthermore, constructing a scene feature vector according to the main feature parameters, calculating the similarity between the scene feature vector and the scene description vector in the preset scene management strategy library, and selecting the management strategy with the highest similarity includes:

[0015] The main feature parameters are subjected to feature dimension reduction and quantization processing according to a preset feature vector construction rule to generate a scene feature vector of fixed dimension, the numerical features in the scene feature vector are standardized, and the category features in the scene feature vector are subjected to one-hot encoding conversion to generate a standardized scene feature vector;

[0016] The cosine similarity calculation method is used to calculate the similarity score between the standardized scene feature vector and each scene description vector in the preset scene management strategy library, and the similarity score is screened based on a preset similarity threshold, and the management strategy corresponding to the scene description vector with the highest similarity score that exceeds the similarity threshold is selected as the matching strategy for the current scene.

[0017] Furthermore, the time period information, passenger flow information and environmental status information in the scene state tag are matched with the preset time period rule base, passenger flow threshold rule base and environmental parameter rule base to obtain corresponding time period management sub-strategies, passenger flow management sub-strategies and environmental management sub-strategies, and the sub-strategies are combined to form a complete scene management strategy, including:

[0018] Parse the scene state label into time period information, passenger flow information and environment state information, divide the time period information according to a preset time granularity and compare it with the rule items in the time period rule base to obtain a time period management sub-strategy, perform density calculation on the passenger flow information and compare it with the rule items in the passenger flow threshold rule base to obtain a passenger flow management sub-strategy, perform parameter classification on the environment state information and compare it with the rule items in the environment parameter rule base to obtain an environment management sub-strategy;

[0019] Based on the preset policy priority, the time period management sub-strategy, the passenger flow management sub-strategy and the environment management sub-strategy are subjected to conflict detection and elimination processing, and the conflict-free sub-strategies are combined and optimized through a policy fusion algorithm to generate a complete scene management strategy. Access control instructions, video surveillance instructions and environment adjustment instructions are generated according to the scene management strategy.

[0020] Furthermore, constructing a customer behavior time series sequence based on the customer behavior characteristics, and inputting the customer behavior time series sequence into a pre-trained behavior analysis model to determine whether there is abnormal behavior, includes:

[0021] Sorting the customer behavior features according to timestamps and constructing a behavior feature sequence in segments, dividing the behavior feature sequence into time windows and performing feature completion processing to obtain a fixed-length behavior time series sequence, and normalizing and standardizing the posture features, motion trajectory features, and interaction features in the behavior time series sequence respectively;

[0022] The standardized behavior time series sequence is input into the abnormal behavior recognition model constructed based on the recurrent neural network. The time series correlation features and local spatial features in the behavior sequence are extracted through the abnormal behavior recognition model. The extracted features are subjected to abnormal detection and classification based on the preset behavior pattern feature library, and the abnormal behavior judgment result is output.

[0023] Furthermore, when abnormal behavior is detected, the abnormal behavior type, occurrence time and location information are recorded in the abnormal event list, and the corresponding security warning mechanism is triggered according to the danger level of the abnormal behavior in the abnormal event list, and the corresponding emergency handling strategy is called from the preset emergency handling solution library according to the store scene status label and the abnormal behavior type, including:

[0024] Determine the type of abnormal behavior detected by the preset behavior classification rules, obtain the spatial coordinate information of the abnormal behavior based on the target tracking algorithm, and write the abnormal event record composed of the abnormal behavior type, timestamp and spatial coordinate information into the abnormal event list, score the abnormal behavior in the abnormal event list according to the preset danger level assessment rules, and determine the warning level based on the danger level score;

[0025] According to the warning level, the preset security warning rule library is queried to generate a warning instruction, the store scene status label and the abnormal behavior type information are input into the emergency processing strategy selection model constructed based on the decision tree, and the corresponding processing strategy is selected from the preset emergency processing solution library through the emergency processing strategy selection model to generate an emergency processing instruction sequence.

[0026] Furthermore, the statistical analysis of the store scene status label and the customer behavior characteristics in the time dimension is performed to obtain store operation data, and the store operation data is input into the adaptive optimization model to generate store management strategy optimization suggestions, including:

[0027] Aggregate the store scene status labels and customer behavior characteristics in segments according to a preset time window, extract multi-dimensional features from the aggregated data to obtain customer flow change trends, customer residence time distribution, product interaction frequency, and environmental parameter change curves, and perform correlation analysis and trend prediction on the various indicators based on a preset data analysis model to generate store operation data;

[0028] The store operation data is input into a strategy optimization model constructed based on a reinforcement learning algorithm. The execution effect of the existing management strategy is evaluated and scored through the strategy optimization model. The reward value of the strategy execution is calculated based on the preset strategy evaluation index. The policy gradient method is used to tune the parameters of the management strategy to generate strategy optimization suggestions.

[0029] In a second aspect, the present application provides a scene-adaptive smart store unattended management device, comprising:

[0030] A scene determination module is used to collect video data and sensor data from multiple cameras in a store, perform target detection on the video data to extract customer behavior features and commodity placement features, perform feature extraction on the sensor data to obtain store environment parameters, input the customer behavior features, commodity placement features and store environment parameters into a pre-trained scene recognition model to obtain a store scene state label, perform semantic analysis on the store scene state label to obtain the main feature parameters of the current scene, construct a scene feature vector based on the main feature parameters, perform similarity calculation on the scene feature vector and the scene description vector in a preset scene management strategy library to select the management strategy with the highest similarity, match the time period information, passenger flow information and environmental status information in the scene state label with a preset time period rule library, passenger flow threshold rule library and environmental parameter rule library respectively to obtain corresponding time period management sub-strategies, passenger flow management sub-strategies and environmental management sub-strategies, and combine the sub-strategies to form a complete scene management strategy;

[0031] An emergency response module is used to construct a customer behavior time series sequence based on the customer behavior characteristics, input the customer behavior time series sequence into a pre-trained behavior analysis model to determine whether there is abnormal behavior, and when abnormal behavior is detected, record the abnormal behavior type, occurrence time and location information into an abnormal event list, trigger a corresponding security warning mechanism according to the danger level of the abnormal behavior in the abnormal event list, and call a corresponding emergency response strategy from a preset emergency response solution library according to the store scene status label and the abnormal behavior type;

[0032] The strategy update module is used to perform statistical analysis on the store scene status label and the customer behavior characteristics in the time dimension to obtain store operation data, input the store operation data into the adaptive optimization model to generate store management strategy optimization suggestions, dynamically update the management strategy in the preset scene management strategy library according to the management strategy optimization suggestions, and send the store operation data, the abnormal event list and the management strategy optimization suggestions to the cloud management platform for storage. The cloud management platform generates a store operation analysis report based on the received data.

[0033] In a third aspect, the present application provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the scene-adaptive-based unattended management method for a smart store are implemented.

[0034] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the scene-adaptive-based unattended management method for smart stores.

[0035] In a fifth aspect, the present application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the scene-adaptive smart store unattended management method.

[0036] It can be seen from the above technical solution that the present application provides a method and device for unattended management of smart stores based on scene adaptation, which collects store data through multiple cameras and sensors to extract customer behavior characteristics, product placement characteristics and environmental parameters. Innovatively construct scene recognition models and behavior analysis models to realize intelligent recognition of scene states and real-time monitoring of abnormal behaviors. Dynamically match management strategies based on scene feature vectors, and formulate detailed management sub-strategies based on time periods, customer flow and environmental conditions. At the same time, analyze operational data through adaptive optimization models to realize dynamic updating and optimization of management strategies. This method breaks through the limitations of traditional fixed management models and provides a comprehensive technical solution for unmanned operations of smart stores. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0038] Figure 1 This is one of the flow charts of the scenario-adaptive unattended management method for smart stores in the embodiment of the present application;

[0039] Figure 2 This is the second flow chart of the unattended management method of a smart store based on scene adaptation in an embodiment of the present application;

[0040] Figure 3 This is a third flow chart of the unattended management method of a smart store based on scene adaptation in an embodiment of the present application;

[0041] Figure 4 This is a fourth flow chart of the unattended management method of a smart store based on scene adaptation in an embodiment of the present application;

[0042] Figure 5 This is a fifth flow chart of the unattended management method of a smart store based on scene adaptation in an embodiment of the present application;

[0043] Figure 6 This is a sixth flow chart of the scene-adaptive smart store unattended management method in the embodiment of the present application;

[0044] Figure 7 This is the seventh flow chart of the unattended management method of a smart store based on scene adaptation in the embodiment of the present application;

[0045] Figure 8 This is a structural diagram of a scene-adaptive smart store unattended management device in an embodiment of the present application;

[0046] Fig. 9 It is a schematic diagram of the structure of an electronic device in an embodiment of the present application.

[0047] Reference numerals:

[0048] Electronic device 9600, central processing unit 9100, memory 9140, communication module 9110, input unit 9120, audio processor 9130, display 9160, power supply 9170, buffer memory 9141, application / function storage unit 9142, data storage unit 9143, driver program storage unit 9144, antenna 9111, speaker 9131, microphone 9132. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0050] The acquisition, storage, use, and processing of data in the technical solution of this application comply with the relevant provisions of national laws and regulations.

[0051] Taking into account the problems existing in the prior art, the present application provides a method and device for unmanned management of smart stores based on scene adaptation, which collects store data through multiple cameras and sensors to extract customer behavior characteristics, product placement characteristics and environmental parameters. Innovatively construct scene recognition models and behavior analysis models to achieve intelligent recognition of scene states and real-time monitoring of abnormal behaviors. Dynamically match management strategies based on scene feature vectors, and formulate detailed management sub-strategies based on time periods, customer flow and environmental conditions. At the same time, analyze operational data through adaptive optimization models to achieve dynamic updating and optimization of management strategies. This method breaks through the limitations of traditional fixed management models and provides a comprehensive technical solution for unmanned operations of smart stores.

[0052] In order to effectively improve the efficiency and reliability of unmanned operation of smart stores, this application provides an embodiment of a smart store unmanned management method based on scene adaptation, see Figure 1 The scenario-adaptive unattended management method for smart stores specifically includes the following contents:

[0053] Step S101: Collect video data and sensor data from multiple cameras in the store, perform target detection on the video data to extract customer behavior features and commodity placement features, perform feature extraction on the sensor data to obtain store environment parameters, input the customer behavior features, commodity placement features and store environment parameters into a pre-trained scene recognition model to obtain a store scene state label, perform semantic analysis on the store scene state label to obtain the main feature parameters of the current scene, construct a scene feature vector according to the main feature parameters, perform similarity calculation on the scene feature vector and the scene description vector in a preset scene management strategy library to select the management strategy with the highest similarity, match the time period information, passenger flow information and environmental status information in the scene state label with a preset time period rule library, passenger flow threshold rule library and environmental parameter rule library respectively to obtain corresponding time period management sub-strategies, passenger flow management sub-strategies and environmental management sub-strategies, and combine the sub-strategies to form a complete scene management strategy;

[0054] Optionally, this embodiment realizes adaptive recognition and management strategy generation of smart store scenarios by combining multi-source data collection and deep learning models. First, multiple high-definition cameras are arranged in the store to cover key areas such as the cash register, shelf area, and entrance, and environmental sensors such as temperature, humidity, light, and air quality are configured to build a multi-dimensional data collection network.

[0055] This embodiment uses an improved YOLOv5 target detection algorithm in the video data processing link, and extracts the customer's posture features, movement trajectory and interactive behavior through a pre-trained model. For example, skeleton key point detection is used to identify whether the customer is staying in front of the shelf and whether he is taking goods, and trajectory analysis is used to determine the customer's movement path and length of stay. At the same time, the product display area is monitored in real time to identify the placement, quantity and arrangement regularity of the products.

[0056] In terms of environmental data processing, this embodiment performs time series correlation analysis and outlier processing on various sensor data. The changing trend of environmental parameters is calculated through a sliding time window, and a multi-dimensional feature vector reflecting the environmental status is generated in combination with the spatial layout characteristics of the store.

[0057] This embodiment innovatively designs a scene recognition model based on multimodal feature fusion. The model adopts a dual-stream network structure to process visual features and environmental features respectively, and realizes adaptive weight allocation of features through the attention mechanism. The output layer of the model generates scene state labels through a multi-label classifier, which contains semantic information in multiple dimensions such as time period attributes, passenger flow density, and environmental conditions.

[0058] In the scene feature analysis phase, this embodiment uses the BERT pre-trained model to perform semantic analysis on the scene state label, extract the core feature words and calculate the word vector representation. The features are reduced and quantified by hierarchical clustering to construct a scene feature vector of fixed dimension. This processing method ensures the consistency and comparability of feature representation.

[0059] This embodiment adopts an improved cosine similarity calculation method in the management strategy matching process. By introducing feature weight coefficients, the contribution of important features in similarity calculation is highlighted. After finding the most similar scene description vector, the system will further combine time period rules, passenger flow thresholds and environmental parameter rules to make refined strategy adjustments.

[0060] In the strategy combination phase, this embodiment designs a priority-based conflict elimination mechanism. For example, when the passenger flow management strategy recommends lowering the air conditioning temperature, and the environmental management strategy recommends raising the temperature, the system will select a more suitable strategy based on the current scene characteristics. In this way, the consistency and executability of the final generated scene management strategy are ensured.

[0061] Through this multi-level scenario perception and strategy generation solution, this embodiment significantly improves the accuracy and adaptability of smart store management. Especially in special scenarios such as dense customer flow during holidays and sudden weather changes, the solution can adjust management strategies in a timely manner to maintain the good operation status of the store. At the same time, the explainability and traceability of the strategy generation process provide reliable data support for store management decisions.

[0062] Step S102: construct a customer behavior time series sequence based on the customer behavior characteristics, input the customer behavior time series sequence into a pre-trained behavior analysis model to determine whether there is abnormal behavior, and when abnormal behavior is detected, record the abnormal behavior type, occurrence time and location information into an abnormal event list, trigger a corresponding security warning mechanism according to the danger level of the abnormal behavior in the abnormal event list, and call a corresponding emergency response strategy from a preset emergency response solution library according to the store scene status label and the abnormal behavior type;

[0063] Optionally, this embodiment designs a behavior recognition solution based on time series feature analysis for the problem of abnormal behavior detection in smart stores. First, the customer behavior features extracted from the video data, including information such as skeleton key points, motion trajectories, and interactive actions, are constructed in the order of timestamps to construct a behavior feature sequence. To ensure the continuity and integrity of the feature sequence, the sliding window method is used to segment the original feature sequence, and the window size is dynamically adjusted according to the duration of the typical abnormal behavior.

[0064] This embodiment adopts a two-layer LSTM network structure in the design of the behavior analysis model. The first layer of LSTM is responsible for extracting local temporal features and capturing short-term change patterns of customer behavior; the second layer of LSTM focuses on long-term dependencies and identifies persistent abnormal behavior patterns. For example, for theft, the model will focus on the combination of temporal features such as abnormal length of stay of customers in the shelf area, frequent actions of repeatedly checking the surroundings, and quick departure.

[0065] In the preprocessing stage of behavior features, this embodiment uses an adaptive normalization method to process different types of feature data. For posture features, individual differences are eliminated through relative coordinate transformation; for motion trajectories, path features are extracted using a curve fitting method; for interactive behaviors, a behavior-item association matrix is ​​constructed to describe the interaction pattern. This multi-dimensional feature representation ensures the model's robust recognition of abnormal behaviors.

[0066] When the behavior analysis model detects abnormal behavior, this embodiment will immediately start the event recording mechanism. The location information of the abnormal behavior subject is updated in real time through the target tracking algorithm, and the standardized spatial coordinates are generated in combination with the spatial layout of the store. At the same time, based on the predefined behavior classification rules, the abnormal behavior is divided into different types, such as product theft, fighting, product damage, etc., and detailed timestamp information is recorded.

[0067] This embodiment innovatively designs a dynamic risk assessment mechanism. By comprehensively considering factors such as the type, duration, number of people involved, and potential hazards of abnormal behavior, the risk score is calculated and the warning level is determined. For example, a fight involving multiple people will trigger a higher level of warning, while a short-term suspicious resident behavior may only require special attention.

[0068] In the selection of emergency response strategies, this embodiment adopts a decision tree model based on knowledge graph. By correlating the current store scene status (such as customer flow density, time period characteristics) with the abnormal behavior type, the most suitable processing strategy is selected from the preset emergency response solution library. For example, if abnormal behavior occurs during a period of high customer flow, the system will give priority to the processing method that does not affect the normal customer shopping experience.

[0069] This embodiment significantly improves the security management capabilities of smart stores through real-time abnormal behavior analysis and rapid response mechanisms. Especially during crowded periods such as holidays, the solution can promptly detect and handle various abnormal events and effectively prevent safety hazards. At the same time, the adaptive characteristics of the system ensure that appropriate processing strategies can be selected in different scenarios, which not only ensures store safety but also maintains a normal shopping environment.

[0070] Step S103: Perform statistical analysis on the store scene status label and the customer behavior characteristics in the time dimension to obtain store operation data, input the store operation data into an adaptive optimization model to generate store management strategy optimization suggestions, dynamically update the management strategies in the preset scene management strategy library according to the management strategy optimization suggestions, and simultaneously send the store operation data, the abnormal event list and the management strategy optimization suggestions to a cloud management platform for storage, and the cloud management platform generates a store operation analysis report based on the received data.

[0071] From the above description, it can be seen that the unattended management method of smart stores based on scene adaptation provided by the embodiment of the present application can collect store data through multiple cameras and sensors, and extract customer behavior characteristics, product placement characteristics and environmental parameters. Innovatively construct scene recognition models and behavior analysis models to achieve intelligent recognition of scene states and real-time monitoring of abnormal behaviors. Dynamically match management strategies based on scene feature vectors, and formulate detailed management sub-strategies based on time periods, customer flow and environmental conditions. At the same time, analyze operational data through adaptive optimization models to achieve dynamic updating and optimization of management strategies. This method breaks through the limitations of traditional fixed management models and provides a comprehensive technical solution for the unmanned operation of smart stores.

[0072] Optionally, this embodiment designs a data-driven store operation optimization solution. First, perform time series aggregation analysis on the store scene status tags, including statistical indicators such as customer flow distribution, dwell time distribution, shopping path heat map, etc. in different time periods. At the same time, aggregate customer behavior characteristics according to time windows to extract key operation indicators such as shopping behavior patterns, product interaction frequency, and regional preferences.

[0073] This embodiment adopts a multi-dimensional feature fusion method in the operational data analysis stage. Through the association rule mining algorithm, the potential relationship between customer flow changes, customer behavior characteristics and product sales is discovered. For example, the correlation between the length of time customers stay in different areas during a specific period of time and the actual purchase conversion rate is analyzed to identify the key factors that affect purchase decisions.

[0074] In the design of the adaptive optimization model, this embodiment innovatively adopts a reinforcement learning framework. The store operation environment is modeled as a Markov decision process, where the state space includes multiple dimensions such as customer flow characteristics, environmental parameters, and product display, and the action space corresponds to the adjustment options of various management strategies. By defining a reasonable reward function, the model is guided to learn the optimal strategy adjustment plan.

[0075] This embodiment takes into account multiple business objectives during the generation of strategy optimization suggestions. For example, when it is found that the customer residence time in a certain area is significantly lower than expected, the model will analyze the correlation between this phenomenon and factors such as product display, lighting conditions, and aisle width, and generate corresponding optimization suggestions. At the same time, the model will evaluate the implementation costs and expected benefits of different optimization solutions to ensure the feasibility of the suggestions.

[0076] In view of the dynamic update mechanism of the management strategy library, this embodiment designs an iterative optimization process based on effect feedback. By tracking and recording the changes in operating indicators after the implementation of the strategy, the optimization effect is evaluated, and the parameter settings in the strategy library are adjusted accordingly. For example, if the passenger flow management strategy for a certain period of time is not effective in practice, the system will automatically adjust the relevant threshold parameters or replace the strategy content.

[0077] In the process of uploading data to the cloud, this embodiment adopts an incremental synchronization mechanism to ensure the real-time and integrity of the data. Operational data is uploaded in batches according to predefined time granularity, and abnormal event information is pushed in real time. To improve data transmission efficiency, the system will compress and encode the original data and implement differentiated storage strategies.

[0078] In this embodiment, an intelligent data interpretation model is designed in the cloud analysis report generation link. The model can automatically identify abnormal fluctuations in key business indicators, analyze the causes of fluctuations, and generate corresponding analysis conclusions. For example, by comparing and analyzing the operating data of different periods, it is found that the decline in sales conversion rate may be related to the newly implemented product display strategy, and specific optimization suggestions are given.

[0079] This embodiment significantly improves the scientificity and efficiency of store operation management by building a complete closed loop of data analysis and strategy optimization. In particular, in key scenarios such as business strategy adjustment and abnormal event handling, the solution can provide data-supported decision-making suggestions to help store managers better grasp business trends and adjust operation strategies in a timely manner. At the same time, the automatic generation mechanism of cloud analysis reports greatly reduces the labor cost of data analysis and improves the timeliness of management decisions.

[0080] In one embodiment of the scenario-adaptive smart store unattended management method of the present application, see Figure 2 , and can also include the following:

[0081] Step S201: The customer behavior features, the product placement features and the store environment parameters are integrated according to a preset ratio to construct a multimodal input matrix, the multimodal input matrix is ​​normalized, the normalized feature data is input into a scene recognition model constructed based on a convolutional neural network and a long short-term memory network, and the store scene status label is output through the scene recognition model;

[0082] Step S202: Perform semantic segmentation and word vector mapping on the text information in the store scene status label through a natural language processing algorithm, extract the spatiotemporal dimension features, personnel flow features and environmental status features in the scene label, and perform weighted calculation on each feature based on preset feature item weights to obtain the main feature parameters of the current scene.

[0083] Optionally, this embodiment designs a multimodal feature fusion scenario analysis solution for the problem of scene feature extraction and state recognition of smart stores. First, customer behavior features (including human posture, motion trajectory, behavior type, etc.), product placement features (including product location, placement density, category distribution, etc.) and store environment parameters (including lighting, temperature, air quality, etc.) are feature vectorized. In the feature fusion process, the attention mechanism is used to dynamically adjust the weight ratio of different features to ensure that the fused feature matrix can fully reflect the information of each dimension.

[0084] This embodiment adopts a hierarchical normalization strategy in the feature preprocessing stage. For discrete features, one-hot encoding is used to convert them into numerical representation; for continuous features, the Z-score normalization method is used to eliminate dimensional differences; for time series features, the sliding window method is used to maintain time series correlation. This multi-level normalization process ensures the comparability of different types of features and the training stability of the model.

[0085] The scene recognition model adopts an innovative two-stream network structure, in which the convolutional neural network is responsible for extracting spatial features and the long short-term memory network processes temporal features. In the convolutional layer design, multi-scale convolution kernels are used to capture spatial patterns of different granularities, such as local product display features and overall spatial layout features. The LSTM network selectively retains and updates temporal information through a gating mechanism, effectively capturing the dynamic changes in scene status.

[0086] This embodiment adopts a hierarchical loss function design in the model training stage. The main loss function focuses on the accuracy of scene classification, while the auxiliary loss function urges the model to learn the correlation between features. For example, when a high passenger flow scene is identified, the model will also focus on related features such as the density distribution of people and the congestion of the commodity area, thereby providing a more comprehensive description of the scene state.

[0087] In the semantic processing of scene state labels, this embodiment designs a semantic analysis framework based on the attention mechanism. First, the scene description text is encoded through the BERT model to obtain context-related word vector representations. Then, the syntactic analysis tool is used to extract the core semantic components and identify key information such as time, space, personnel, and environment. For example, for a scene description such as "the cashier area is crowded with customers during the morning rush hour", it is possible to accurately extract dimensional information such as time (morning rush hour), space (cashier area), and state (cashier area) (cashier area).

[0088] In the feature weighted calculation phase, this embodiment adopts an adaptive weight allocation mechanism. By analyzing the contribution of different features in historical data to the scene state judgment, the feature weight is dynamically adjusted. For example, during a promotional event, the weight of the personnel flow feature will be appropriately increased to better reflect the dynamic changes of the scene.

[0089] This embodiment achieves accurate recognition and feature extraction of store scene states through the combination of multimodal feature fusion and deep learning models. This solution can effectively handle complex and changeable store scenes and provide reliable data support for subsequent scene management and decision optimization. Especially in the case of large fluctuations in customer flow and frequent scene changes, this solution can still maintain a stable recognition effect and effectively support the intelligent operation and management of stores.

[0090] In one embodiment of the scenario-adaptive smart store unattended management method of the present application, see Figure 3 , and can also include the following:

[0091] Step S301: performing feature dimension reduction and quantization processing on the main feature parameters according to a preset feature vector construction rule to generate a scene feature vector of fixed dimension, performing standardization processing on the numerical features in the scene feature vector, and performing one-hot encoding conversion on the category features in the scene feature vector to generate a standardized scene feature vector;

[0092] Step S302: Use the cosine similarity calculation method to calculate the similarity score between the standardized scene feature vector and each scene description vector in the preset scene management strategy library, filter the similarity score based on a preset similarity threshold, and select the management strategy corresponding to the scene description vector with the highest similarity score that exceeds the similarity threshold as the matching strategy for the current scene.

[0093] Optionally, this embodiment designs a scene matching solution based on feature vectorization for the problem of standardized processing of store scene features and matching of management strategies. First, the main feature parameters are reduced in dimension, and the most representative feature combination is extracted using the principal component analysis (PCA) method. In the feature selection process, the key dimensions that have a significant impact on the judgment of the scene state are retained, such as personnel density, activity intensity, environmental parameters, etc.

[0094] This embodiment adopts a hierarchical processing strategy in the feature vector construction stage. For the time dimension feature, the time period is divided into fixed intervals and one-hot encoding is performed; for the space dimension feature, the store area is divided into several units using a gridding method, and the feature state of each unit is recorded; for the environmental state feature, the parameters of different dimensions are mapped to a unified interval through min-max normalization. This multi-dimensional feature processing method ensures the integrity and comparability of scene features.

[0095] In the process of standardization, this embodiment adopts differentiated processing strategies for different types of features. For continuous numerical features (such as temperature, humidity, illumination, etc.), the Z-score standardization method is used to eliminate the dimension effect; for discrete category features (such as scene type, activity status, etc.), they are converted into numerical representations through one-hot encoding. This processing method not only maintains the original information of the features, but also ensures the accuracy of vector calculation.

[0096] This embodiment innovatively adopts an improved cosine similarity calculation method in the feature matching stage. By introducing feature weight factors, different importance coefficients are assigned to features of different dimensions. For example, in a crowded scene, the weights of the features of personnel density and activity intensity will be increased accordingly to more accurately match the corresponding management strategy.

[0097] In order to optimize the similarity calculation, this embodiment designs an adaptive threshold adjustment mechanism. The system will dynamically adjust the similarity threshold according to the historical matching results, while ensuring the matching accuracy and improving the coverage of policy recommendations. For example, when it is detected that the matching success rate of a certain type of scene is low, the system will appropriately lower the similarity threshold of this type of scene and add corresponding policy verification steps.

[0098] This embodiment uses a multi-round screening mechanism in the strategy matching process. First, a preliminary screening is performed based on the similarity score, and then a secondary verification is performed in combination with the temporal change trend of the scene characteristics. For example, when it is detected that the passenger flow continues to increase, the system will give priority to predictive management strategies and prepare for scene control in advance.

[0099] In practical applications, this embodiment can effectively handle various scenarios in store operations. For example, when a queue congestion is detected in the cashier area, the system will quickly match a suitable customer flow diversion strategy; when it is found that the length of time people stay in a certain commodity area is abnormal, the corresponding commodity display optimization plan will be recommended in a timely manner.

[0100] This embodiment achieves accurate matching between scene features and management strategies through feature vectorization processing and similarity matching calculation. This solution can effectively respond to various scene changes in store operations and provide timely and accurate strategy recommendations for managers. Especially during special periods such as holidays, this solution can quickly adjust management strategies according to changes in scene features to improve store operation efficiency and service quality.

[0101] In one embodiment of the scenario-adaptive smart store unattended management method of the present application, see Figure 4 , and can also include the following:

[0102] Step S401: parsing the scene state label into time period information, passenger flow information and environment state information, dividing the time period information according to a preset time granularity and comparing it with the rule items in the time period rule base to obtain a time period management sub-strategy, performing density calculation on the passenger flow information and comparing it with the rule items in the passenger flow threshold rule base to obtain a passenger flow management sub-strategy, performing parameter classification on the environment state information and comparing it with the rule items in the environment parameter rule base to obtain an environment management sub-strategy;

[0103] Step S402: Based on the preset policy priority, the time period management sub-strategy, the passenger flow management sub-strategy and the environment management sub-strategy are subjected to conflict detection and elimination processing, and the conflict-free sub-strategies are combined and optimized through a policy fusion algorithm to generate a complete scene management strategy, and access control instructions, video surveillance instructions and environment adjustment instructions are generated according to the scene management strategy.

[0104] Optionally, this embodiment designs a management solution based on multi-dimensional information analysis and strategy fusion for the problem of generating scene management strategies for smart stores. First, semantic decomposition is performed on scene status labels to extract key information dimensions. Time period information includes specific time points, duration, etc.; customer flow information includes the number of people, distribution density, etc.; environmental status information includes parameters such as temperature, humidity, and illumination.

[0105] This embodiment adopts a multi-level time granularity division strategy in time period management. A day is divided into morning, mid-day, evening peak periods and off-peak periods, and differentiated management rules are formulated for each period based on historical data characteristics. For example, during the morning peak period, the system will automatically adjust the number of access control openings based on the flow of people; during the night period, the lighting and air conditioning settings are adjusted accordingly to save energy.

[0106] In the process of generating passenger flow management strategies, this embodiment designs a dynamic density calculation model. Through the regional grid method, the real-time density of people in each functional area is calculated and compared with the preset safety threshold. When the density of people in a certain area is close to the warning threshold, the system will give priority to starting the crowd diversion strategy, such as adjusting the access control channel, opening the backup channel, etc.

[0107] This embodiment implements a parameter linkage control mechanism in environmental management. By establishing a correlation model between environmental parameters, multi-parameter coordinated adjustment is achieved. For example, when an increase in indoor carbon dioxide concentration is detected, the system will adjust the fresh air volume and air conditioning operating parameters at the same time to maintain environmental comfort while ensuring air quality.

[0108] To address the problem of policy conflicts, this embodiment innovatively uses the analytic hierarchy process to establish a policy priority system. First, the basic weights are determined based on factors such as safety, comfort, and energy efficiency, and then the priorities of each policy are dynamically adjusted based on real-time scenario characteristics. For example, during large-scale promotional activities, the priority of the safety diversion policy will be significantly increased.

[0109] This embodiment adopts a progressive combination optimization method in the policy fusion process. First, the implementation of the basic security policy is ensured, then the comfort optimization policy is superimposed on this basis, and finally the energy efficiency policy is considered. This progressive policy fusion method not only ensures the realization of management goals, but also avoids mutual interference between policies.

[0110] In the control command generation link, this embodiment designs an intelligent command conversion mechanism. The system can automatically convert abstract management strategies into specific device control commands. For example, when it is necessary to implement a crowd diversion strategy, the system will generate a series of collaborative control commands including access control opening mode adjustment, surveillance camera angle adjustment, LED display information update, etc.

[0111] This embodiment achieves refined management of store scenes through multi-dimensional information analysis and strategy fusion optimization. The solution can adjust management strategies and generate corresponding control instructions in a timely manner according to the dynamic changes of scene characteristics. Especially in abnormal situations such as drastic fluctuations in customer flow or sudden changes in environmental parameters, the system can respond quickly and implement effective management measures to ensure the safety of store operations and service quality.

[0112] This embodiment shows good adaptability and reliability in practical applications. Through comprehensive analysis of time characteristics, customer flow characteristics and environmental characteristics, the system can accurately grasp the trend of scene changes, make management plans in advance, and effectively improve the intelligent management level and operational efficiency of the store.

[0113] In one embodiment of the scenario-adaptive smart store unattended management method of the present application, see Figure 5 , and can also include the following:

[0114] Step S501: sorting the customer behavior features according to timestamps and constructing a behavior feature sequence in segments, dividing the behavior feature sequence into time windows and performing feature completion processing to obtain a fixed-length behavior time series sequence, and normalizing and standardizing the posture features, motion trajectory features, and interaction features in the behavior time series sequence respectively;

[0115] Step S502: Input the standardized behavior time series sequence into the abnormal behavior recognition model constructed based on the recurrent neural network, extract the time series correlation features and local spatial features in the behavior sequence through the abnormal behavior recognition model, perform anomaly detection and classification on the extracted features based on the preset behavior pattern feature library, and output the abnormal behavior judgment result.

[0116] Optionally, this embodiment designs a deep learning-based abnormal behavior recognition solution for the problem of customer behavior analysis in smart stores. First, the collected customer behavior features are processed in time series, and the discrete behavior feature points are reorganized in timestamp order to construct a continuous behavior feature sequence. These features include the customer's walking speed, dwell time, posture changes, and interaction with the product.

[0117] This embodiment uses sliding window technology in the time window division link, and the window size is dynamically adjusted according to the typical duration of different types of abnormal behaviors. For example, for the identification of theft behavior, a longer time window is used to capture the complete behavior chain; for sudden abnormal behavior, a shorter time window is used to improve the real-time recognition. When the sequence length is insufficient, the linear interpolation method is used to complete the features to ensure the integrity of the input sequence.

[0118] In terms of feature standardization, this embodiment adopts differentiated processing strategies for different types of behavioral features. For posture features, the skeleton point coordinate normalization method is used to eliminate the influence of human body size differences; for motion trajectory features, the trajectory is mapped to a unified coordinate space through trajectory resampling and coordinate transformation; for interaction features, the numerical conversion is performed by combining one-hot encoding and normalization.

[0119] This embodiment innovatively uses a bidirectional long short-term memory network (Bi-LSTM) structure when constructing an abnormal behavior recognition model. The input layer receives the standardized behavior time sequence, extracts time sequence features through multi-layer Bi-LSTM units, and captures the long-term and short-term dependencies in the behavior sequence. For example, the model can learn typical time sequence patterns in normal shopping behaviors, such as the behavior chain of browsing-selecting-trying-purchasing.

[0120] For spatial feature extraction, this embodiment introduces an attention mechanism into the network. By calculating the attention weights of different time steps and spatial positions, the model can automatically focus on the key time points and spatial areas where abnormal behaviors occur. For example, when a customer shows suspicious wandering behavior in the shelf area, the model will focus on the behavioral characteristics of this area.

[0121] This embodiment designs a hierarchical anomaly detection mechanism in the feature matching link. First, the anomaly score of the behavior sequence is calculated based on the extracted time series features, and then the anomaly score is matched with the standard pattern in the preset behavior pattern feature library to achieve accurate classification of abnormal behaviors. For example, the system can distinguish different types of abnormal behaviors such as fighting, commodity theft, and vandalism.

[0122] In practical applications, this embodiment uses an incremental learning strategy to continuously optimize model performance. By collecting and annotating new abnormal behavior samples, regularly updating model parameters and behavior pattern feature libraries, the model's ability to identify new abnormal behaviors is improved.

[0123] This embodiment uses deep learning technology to achieve real-time recognition of abnormal behaviors in store scenarios. The solution can accurately capture subtle anomalies in behavior sequences and provide timely warning information. Especially in shopping mall security management, the system can help security personnel discover potential safety hazards in advance and effectively prevent the occurrence of various emergencies.

[0124] This embodiment plays an important role in the daily operation of stores. It not only improves the intelligent level of security management, but also provides a scientific decision-making basis for shopping mall managers, effectively ensuring the safety of the shopping environment and the shopping experience of customers.

[0125] In one embodiment of the scenario-adaptive smart store unattended management method of the present application, see Figure 6 , and can also include the following:

[0126] Step S601: Determine the type of abnormal behavior detected by the preset behavior classification rule, obtain the spatial coordinate information of the abnormal behavior based on the target tracking algorithm, and write the abnormal event record composed of the abnormal behavior type, timestamp and spatial coordinate information into the abnormal event list, score the abnormal behavior in the abnormal event list according to the preset danger level assessment rule, and determine the warning level based on the danger level score;

[0127] Step S602: Generate a warning instruction by querying the preset security warning rule library according to the warning level, input the store scene status label and the abnormal behavior type information into the emergency processing strategy selection model constructed based on the decision tree, select the corresponding processing strategy from the preset emergency processing solution library through the emergency processing strategy selection model, and generate an emergency processing instruction sequence.

[0128] Optionally, this embodiment designs a set of abnormal event response and processing solutions for the security management of smart stores. First, the detected abnormal behaviors are identified based on the behavior classification rules, including but not limited to fighting, theft, destruction of facilities and other abnormal types. The spatial location information of the abnormal behavior subject is obtained in real time through the multi-target tracking algorithm to ensure the accurate positioning of abnormal events.

[0129] This embodiment adopts a structured storage method in the abnormal event recording link. Each record contains key information such as abnormal behavior type, occurrence time, spatial coordinates, and number of people involved. In order to improve the availability of data, the spatial coordinates are represented by a relative coordinate system to facilitate subsequent regional correlation analysis and trajectory tracking.

[0130] For the risk level assessment, this embodiment designs a multi-dimensional scoring system. The scoring indicators include the risk of the behavior, the scope of impact, the duration, the number of people affected, etc. For example, abnormal behaviors involving violent conflicts will receive a higher risk score, while abnormal behaviors occurring in crowded areas will receive a higher score in the impact scope dimension.

[0131] This embodiment adopts a dynamic threshold mechanism in the division of warning levels. Based on historical data statistics and expert experience, score intervals for different warning levels are set. When the comprehensive score of a certain type of abnormal behavior exceeds the corresponding threshold, a warning of the corresponding level is triggered. For example, a minor abnormal behavior may trigger a yellow warning, while a behavior that seriously threatens personal safety may trigger a red warning.

[0132] In the phase of emergency response strategy selection, this embodiment innovatively introduces an intelligent decision-making model based on a decision tree. The construction of the decision tree takes into account two dimensions: scene status (such as crowd density and time characteristics) and abnormal behavior characteristics, and achieves accurate matching of strategies through the combination of multiple layers of decision nodes. For example, when there is a risk of crowding and trampling during a period of dense crowd flow, the model will give priority to evacuation and diversion strategies.

[0133] This embodiment realizes the dynamic combination of strategies in the process of generating emergency handling instructions. Based on the current scene status and abnormal behavior characteristics, the most suitable basic strategy is selected from the emergency handling solution library, and then the strategy is adjusted and optimized according to the actual situation. For example, when commodity theft is detected, the system will generate a series of collaborative processing instructions including target tracking, exit control, and police force deployment.

[0134] In response to complex emergency scenarios, this embodiment designs a hierarchical processing mechanism. For low-risk abnormal events, the system can automatically execute the preset processing flow; for high-risk events, the early warning system notifies manual intervention to ensure the accuracy and timeliness of the processing measures.

[0135] This embodiment takes into account the scene linkage factor in the design of the warning rules. When an abnormal event occurs in a certain area, the system will automatically evaluate the potential impact on the surrounding areas and take corresponding preventive measures. For example, when an abnormality occurs in a key area, the monitoring of adjacent areas will be automatically strengthened.

[0136] This embodiment significantly improves the security management efficiency of the store through the intelligent abnormal event processing process. The solution can quickly formulate and implement appropriate emergency response strategies based on the specific characteristics and scenario conditions of the abnormal event, effectively reducing the incidence and impact of safety accidents.

[0137] This embodiment shows excellent adaptability and reliability in practical applications. Through intelligent analysis and rapid response to abnormal events, it provides all-round security for stores, effectively improving the safety management level of shopping malls and customers' shopping experience.

[0138] In one embodiment of the scenario-adaptive smart store unattended management method of the present application, see Figure 7 , and can also include the following:

[0139] Step S701: Aggregate the store scene status label and the customer behavior characteristics in segments according to a preset time window, extract multi-dimensional features from the aggregated data to obtain customer flow change trends, customer residence time distribution, commodity interaction frequency, and environmental parameter change curves, and perform correlation analysis and trend prediction on the various indicators based on a preset data analysis model to generate store operation data;

[0140] Step S702: Input the store operation data into the strategy optimization model constructed based on the reinforcement learning algorithm, evaluate and score the execution effect of the existing management strategy through the strategy optimization model, calculate the reward value of the strategy execution based on the preset strategy evaluation index, and use the policy gradient method to tune the parameters of the management strategy to generate strategy optimization suggestions.

[0141] Optionally, this embodiment designs a data-driven management strategy optimization solution for the smart store operation optimization problem. First, the collected scene status and customer behavior data are segmented into time windows, and the dynamic aggregation of data is achieved through the sliding window method to ensure the continuity and real-time nature of the analysis.

[0142] This embodiment adopts a multi-dimensional analysis method in the feature extraction stage. The analysis of customer flow change trends includes intraday fluctuations, periodic changes and special time distribution; customer residence time analysis focuses on the stay characteristics and spatial distribution patterns in different areas; commodity interaction frequency statistics include multiple levels of interaction behaviors such as touching, picking up and purchasing commodities; and environmental parameter analysis covers the change characteristics of multiple dimensions such as temperature, humidity, and lighting.

[0143] For the construction of data analysis models, this embodiment innovatively adopts a method combining time series analysis and spatial statistics. Through time series decomposition technology, the changes of various indicators are decomposed into trend items, seasonal items and random items, and the law of data changes is deeply explored. For example, by analyzing the relationship between customer residence time and product layout, the key factors affecting the shopping experience are found.

[0144] This embodiment introduces multi-level association rule mining in the correlation analysis phase. By calculating the correlation coefficients and mutual information between different indicators, the explicit and implicit associations between indicators are identified. For example, the impact of environmental parameter changes on customer stay time is analyzed, or the relationship between product display methods and interaction frequency is explored.

[0145] For trend prediction, this embodiment designs a prediction model based on multiple factors. The model input includes multiple dimensions such as historical data, time characteristics, and scene status. The nonlinear relationship between features is extracted through a deep learning network to achieve accurate prediction of future operating indicators. For example, the change of passenger flow in a specific period of time is predicted to provide a basis for personnel deployment.

[0146] This embodiment uses a deep reinforcement learning framework in the construction of the strategy optimization model. The store operation environment is abstracted into a state space, including multiple dimensions such as customer flow, product display, and staffing; the management strategy is mapped into an action space, including decision options such as product layout adjustment, promotion strategy formulation, and human resource deployment.

[0147] This embodiment designs a comprehensive evaluation system for the strategy evaluation indicators. The evaluation indicators include multiple dimensions such as the effect of increasing customer flow, changes in customer stay time, growth in the frequency of product interactions, and control of operating costs. By designing a reasonable reward function, the improvement degree of each indicator is converted into a reward value for strategy execution.

[0148] This embodiment adopts a progressive tuning method in the policy optimization process. The policy network parameters are continuously optimized through the policy gradient algorithm to achieve the optimal decision-making rules learned from historical experience. For example, the system can automatically adjust the product display and promotion strategy according to the characteristics of customer flow in different time periods.

[0149] This embodiment significantly improves store operation efficiency through an intelligent strategy optimization solution. The solution can accurately capture the changing patterns in operation data and continuously optimize management strategies through reinforcement learning technology, effectively improving operating efficiency.

[0150] This embodiment shows adaptability and scalability in practical applications. Through data-driven decision optimization, it provides scientific decision support for store managers, effectively improving operational efficiency and customer satisfaction. At the same time, the system's adaptive learning ability ensures that the optimization strategy can continue to evolve with environmental changes and maintain long-term effectiveness.

[0151] In order to effectively improve the efficiency and reliability of unmanned operation of smart stores, the present application provides an embodiment of a scene-adaptive smart store unmanned management device for realizing all or part of the contents of the scene-adaptive smart store unmanned management method, see Figure 8 The scene-adaptive smart store unattended management device specifically includes the following contents:

[0152] The scene determination module 10 is used to collect video data and sensor data from multiple cameras in the store, perform target detection on the video data to extract customer behavior characteristics and commodity placement characteristics, perform feature extraction on the sensor data to obtain store environment parameters, input the customer behavior characteristics, the commodity placement characteristics and the store environment parameters into a pre-trained scene recognition model to obtain a store scene state label, perform semantic analysis on the store scene state label to obtain the main feature parameters of the current scene, construct a scene feature vector according to the main feature parameters, perform similarity calculation on the scene feature vector and the scene description vector in a preset scene management strategy library to select the management strategy with the highest similarity, match the time period information, passenger flow information and environmental status information in the scene state label with a preset time period rule library, passenger flow threshold rule library and environmental parameter rule library respectively to obtain corresponding time period management sub-strategies, passenger flow management sub-strategies and environmental management sub-strategies, and combine the sub-strategies to form a complete scene management strategy;

[0153] The emergency handling module 20 is used to construct a customer behavior time series sequence based on the customer behavior characteristics, input the customer behavior time series sequence into a pre-trained behavior analysis model to determine whether there is abnormal behavior, and when abnormal behavior is detected, record the abnormal behavior type, occurrence time and location information into an abnormal event list, trigger a corresponding security warning mechanism according to the danger level of the abnormal behavior in the abnormal event list, and call the corresponding emergency handling strategy from a preset emergency handling solution library according to the store scene status label and the abnormal behavior type;

[0154] The strategy update module 30 is used to perform statistical analysis on the store scene status label and the customer behavior characteristics in the time dimension to obtain store operation data, input the store operation data into the adaptive optimization model to generate store management strategy optimization suggestions, dynamically update the management strategy in the preset scene management strategy library according to the management strategy optimization suggestions, and at the same time send the store operation data, the abnormal event list and the management strategy optimization suggestions to the cloud management platform for storage, and the cloud management platform generates a store operation analysis report based on the received data.

[0155] From the above description, it can be seen that the unmanned management device for smart stores based on scene adaptation provided by the embodiment of the present application can collect store data through multiple cameras and sensors, and extract customer behavior characteristics, product placement characteristics and environmental parameters. Innovatively construct scene recognition models and behavior analysis models to achieve intelligent recognition of scene states and real-time monitoring of abnormal behaviors. Dynamically match management strategies based on scene feature vectors, and formulate detailed management sub-strategies based on time periods, customer flow and environmental conditions. At the same time, analyze operational data through adaptive optimization models to achieve dynamic updating and optimization of management strategies. This method breaks through the limitations of traditional fixed management models and provides a comprehensive technical solution for the unmanned operation of smart stores.

[0156] From the hardware level, in order to effectively improve the unmanned operation efficiency and reliability of smart stores, the present application provides an embodiment of an electronic device for implementing all or part of the contents of the scenario-adaptive smart store unmanned management method, and the electronic device specifically includes the following contents:

[0157] Processor, memory, communication interface and bus; wherein the processor, memory and communication interface communicate with each other through the bus; the communication interface is used to realize information transmission between the scene-adaptive smart store unattended management device and the core business system, user terminal and related database and other related devices; the logic controller can be a desktop computer, a tablet computer and a mobile terminal, etc., but the present embodiment is not limited thereto. In the present embodiment, the logic controller can be implemented with reference to the embodiment of the scene-adaptive smart store unattended management method and the scene-adaptive smart store unattended management device in the embodiment, and the contents are incorporated herein, and the repeated parts are not repeated.

[0158] It is understandable that the user terminal may include a smart phone, a tablet electronic device, a network set-top box, a portable computer, a desktop computer, a personal digital assistant (PDA), a vehicle-mounted device, a smart wearable device, etc. Among them, the smart wearable device may include smart glasses, a smart watch, a smart bracelet, etc.

[0159] In practical applications, part of the scenario-adaptive smart store unattended management method can be executed on the electronic device side as described above, or all operations can be completed in the client device. The specific selection can be based on the processing capability of the client device and the limitations of the user's usage scenario. This application does not limit this. If all operations are completed in the client device, the client device may also include a processor.

[0160] The client device may have a communication module (i.e., a communication unit) that can communicate with a remote server to achieve data transmission with the server. The server may include a server on the task scheduling center side, and other implementation scenarios may also include a server on an intermediate platform, such as a server on a third-party server platform that has a communication link with the task scheduling center server. The server may include a single computer device, or a server cluster consisting of multiple servers, or a server structure of a distributed device.

[0161] Fig. 9 FIG. 9 is a schematic block diagram of the system structure of the electronic device 9600 according to an embodiment of the present application. Fig. 9 As shown, the electronic device 9600 may include a central processor 9100 and a memory 9140; the memory 9140 is coupled to the central processor 9100. It is worth noting that Fig. 9 is exemplary; other types of structures may also be used to supplement or replace this structure to implement telecommunication functions or other functions.

[0162] In one embodiment, the function of the scenario-adaptive smart store unattended management method can be integrated into the central processor 9100. The central processor 9100 can be configured to perform the following control:

[0163] Step S101: Collect video data and sensor data from multiple cameras in the store, perform target detection on the video data to extract customer behavior features and commodity placement features, perform feature extraction on the sensor data to obtain store environment parameters, input the customer behavior features, commodity placement features and store environment parameters into a pre-trained scene recognition model to obtain a store scene state label, perform semantic analysis on the store scene state label to obtain the main feature parameters of the current scene, construct a scene feature vector according to the main feature parameters, perform similarity calculation on the scene feature vector and the scene description vector in a preset scene management strategy library to select the management strategy with the highest similarity, match the time period information, passenger flow information and environmental status information in the scene state label with a preset time period rule library, passenger flow threshold rule library and environmental parameter rule library respectively to obtain corresponding time period management sub-strategies, passenger flow management sub-strategies and environmental management sub-strategies, and combine the sub-strategies to form a complete scene management strategy;

[0164] Step S102: construct a customer behavior time series sequence based on the customer behavior characteristics, input the customer behavior time series sequence into a pre-trained behavior analysis model to determine whether there is abnormal behavior, and when abnormal behavior is detected, record the abnormal behavior type, occurrence time and location information into an abnormal event list, trigger a corresponding security warning mechanism according to the danger level of the abnormal behavior in the abnormal event list, and call a corresponding emergency response strategy from a preset emergency response solution library according to the store scene status label and the abnormal behavior type;

[0165] Step S103: Perform statistical analysis on the store scene status label and the customer behavior characteristics in the time dimension to obtain store operation data, input the store operation data into an adaptive optimization model to generate store management strategy optimization suggestions, dynamically update the management strategies in the preset scene management strategy library according to the management strategy optimization suggestions, and simultaneously send the store operation data, the abnormal event list and the management strategy optimization suggestions to a cloud management platform for storage, and the cloud management platform generates a store operation analysis report based on the received data.

[0166] From the above description, it can be seen that the electronic device provided in the embodiment of the present application collects store data through multiple cameras and sensors to extract customer behavior characteristics, product placement characteristics and environmental parameters. Innovatively construct scene recognition models and behavior analysis models to achieve intelligent recognition of scene states and real-time monitoring of abnormal behaviors. Dynamically match management strategies based on scene feature vectors, and formulate detailed management sub-strategies based on time periods, passenger flow and environmental conditions. At the same time, analyze operational data through adaptive optimization models to achieve dynamic updating and optimization of management strategies. This method breaks through the limitations of traditional fixed management models and provides a comprehensive technical solution for unmanned operations of smart stores.

[0167] In another embodiment, the scene-adaptive unmanned management device for a smart store can be configured separately from the central processor 9100. For example, the scene-adaptive unmanned management device for a smart store can be configured as a chip connected to the central processor 9100, and the function of the scene-adaptive unmanned management method for a smart store can be realized through the control of the central processor.

[0168] like Fig. 9 As shown, the electronic device 9600 may also include: a communication module 9110, an input unit 9120, an audio processor 9130, a display 9160, and a power supply 9170. It is worth noting that the electronic device 9600 does not necessarily have to include Fig. 9 In addition, the electronic device 9600 may also include Fig. 9 For components not shown, reference may be made to the prior art.

[0169] like Fig. 9 As shown, the central processing unit 9100 is sometimes also referred to as a controller or an operation control, and may include a microprocessor or other processor device and / or logic device. The central processing unit 9100 receives input and controls the operation of various components of the electronic device 9600.

[0170] The memory 9140 may be, for example, one or more of a cache, a flash memory, a hard drive, a removable medium, a volatile memory, a non-volatile memory or other suitable devices. The above-mentioned information related to the failure may be stored, and a program for executing the relevant information may also be stored. The CPU 9100 may execute the program stored in the memory 9140 to implement information storage or processing, etc.

[0171] The input unit 9120 provides input to the central processing unit 9100. The input unit 9120 is, for example, a key or a touch input device. The power supply 9170 is used to provide power to the electronic device 9600. The display 9160 is used to display display objects such as images and texts. The display may be, for example, an LCD display, but is not limited thereto.

[0172] The memory 9140 may be a solid-state memory, such as a read-only memory (ROM), a random access memory (RAM), a SIM card, etc. It may also be a memory that saves information even when the power is off, can be selectively erased, and is provided with more data, examples of which are sometimes referred to as EPROMs, etc. The memory 9140 may also be some other type of device. The memory 9140 includes a buffer memory 9141 (sometimes referred to as a buffer). The memory 9140 may include an application / function storage unit 9142, which is used to store application programs and function programs or processes for executing the operation of the electronic device 9600 through the central processor 9100.

[0173] The memory 9140 may also include a data storage unit 9143 for storing data, such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. The driver storage unit 9144 of the memory 9140 may include various drivers for communication functions of the electronic device and / or for executing other functions of the electronic device (such as messaging applications, address book applications, etc.).

[0174] The communication module 9110 is a transmitter / receiver that sends and receives signals via the antenna 9111. The communication module 9110 (transmitter / receiver) is coupled to the central processor 9100 to provide input signals and receive output signals, which may be the same as the case of a conventional mobile communication terminal.

[0175] Based on different communication technologies, multiple communication modules 9110 may be provided in the same electronic device, such as a cellular network module, a Bluetooth module and / or a wireless LAN module. The communication module 9110 (transmitter / receiver) is also coupled to a speaker 9131 and a microphone 9132 via an audio processor 9130 to provide an audio output via the speaker 9131 and receive an audio input from the microphone 9132, thereby realizing a common telecommunication function. The audio processor 9130 may include any suitable buffer, decoder, amplifier, etc. In addition, the audio processor 9130 is also coupled to the central processor 9100, so that recording can be performed on the local machine through the microphone 9132, and the sound stored on the local machine can be played through the speaker 9131.

[0176] The embodiments of the present application also provide a computer-readable storage medium capable of implementing all the steps of the scenario-adaptive unattended management method for a smart store in the above embodiment, where the execution subject is a server or a client. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, all the steps of the scenario-adaptive unattended management method for a smart store in the above embodiment are implemented. For example, when the processor executes the computer program, the following steps are implemented:

[0177] Step S101: Collect video data and sensor data from multiple cameras in the store, perform target detection on the video data to extract customer behavior features and commodity placement features, perform feature extraction on the sensor data to obtain store environment parameters, input the customer behavior features, commodity placement features and store environment parameters into a pre-trained scene recognition model to obtain a store scene state label, perform semantic analysis on the store scene state label to obtain the main feature parameters of the current scene, construct a scene feature vector according to the main feature parameters, perform similarity calculation on the scene feature vector and the scene description vector in a preset scene management strategy library to select the management strategy with the highest similarity, match the time period information, passenger flow information and environmental status information in the scene state label with a preset time period rule library, passenger flow threshold rule library and environmental parameter rule library respectively to obtain corresponding time period management sub-strategies, passenger flow management sub-strategies and environmental management sub-strategies, and combine the sub-strategies to form a complete scene management strategy;

[0178] Step S102: construct a customer behavior time series sequence based on the customer behavior characteristics, input the customer behavior time series sequence into a pre-trained behavior analysis model to determine whether there is abnormal behavior, and when abnormal behavior is detected, record the abnormal behavior type, occurrence time and location information into an abnormal event list, trigger a corresponding security warning mechanism according to the danger level of the abnormal behavior in the abnormal event list, and call a corresponding emergency response strategy from a preset emergency response solution library according to the store scene status label and the abnormal behavior type;

[0179] Step S103: Perform statistical analysis on the store scene status label and the customer behavior characteristics in the time dimension to obtain store operation data, input the store operation data into an adaptive optimization model to generate store management strategy optimization suggestions, dynamically update the management strategies in the preset scene management strategy library according to the management strategy optimization suggestions, and simultaneously send the store operation data, the abnormal event list and the management strategy optimization suggestions to a cloud management platform for storage, and the cloud management platform generates a store operation analysis report based on the received data.

[0180] From the above description, it can be seen that the computer-readable storage medium provided in the embodiment of the present application collects store data through multiple cameras and sensors to extract customer behavior characteristics, product placement characteristics and environmental parameters. Innovatively construct scene recognition models and behavior analysis models to achieve intelligent recognition of scene states and real-time monitoring of abnormal behaviors. Dynamically match management strategies based on scene feature vectors, and formulate detailed management sub-strategies based on time periods, customer flow and environmental conditions. At the same time, analyze operational data through adaptive optimization models to achieve dynamic updating and optimization of management strategies. This method breaks through the limitations of traditional fixed management models and provides a comprehensive technical solution for unmanned operations of smart stores.

[0181] The embodiments of the present application also provide a computer program product capable of implementing all the steps of the scenario-adaptive smart store unattended management method in the above embodiments, where the execution subject is a server or a client. When the computer program / instruction is executed by a processor, the steps of the scenario-adaptive smart store unattended management method are implemented. For example, the computer program / instruction implements the following steps:

[0182] Step S101: Collect video data and sensor data from multiple cameras in the store, perform target detection on the video data to extract customer behavior features and commodity placement features, perform feature extraction on the sensor data to obtain store environment parameters, input the customer behavior features, commodity placement features and store environment parameters into a pre-trained scene recognition model to obtain a store scene state label, perform semantic analysis on the store scene state label to obtain the main feature parameters of the current scene, construct a scene feature vector according to the main feature parameters, perform similarity calculation on the scene feature vector and the scene description vector in a preset scene management strategy library to select the management strategy with the highest similarity, match the time period information, passenger flow information and environmental status information in the scene state label with a preset time period rule library, passenger flow threshold rule library and environmental parameter rule library respectively to obtain corresponding time period management sub-strategies, passenger flow management sub-strategies and environmental management sub-strategies, and combine the sub-strategies to form a complete scene management strategy;

[0183] Step S102: construct a customer behavior time series sequence based on the customer behavior characteristics, input the customer behavior time series sequence into a pre-trained behavior analysis model to determine whether there is abnormal behavior, and when abnormal behavior is detected, record the abnormal behavior type, occurrence time and location information into an abnormal event list, trigger a corresponding security warning mechanism according to the danger level of the abnormal behavior in the abnormal event list, and call a corresponding emergency response strategy from a preset emergency response solution library according to the store scene status label and the abnormal behavior type;

[0184] Step S103: Perform statistical analysis on the store scene status label and the customer behavior characteristics in the time dimension to obtain store operation data, input the store operation data into an adaptive optimization model to generate store management strategy optimization suggestions, dynamically update the management strategies in the preset scene management strategy library according to the management strategy optimization suggestions, and simultaneously send the store operation data, the abnormal event list and the management strategy optimization suggestions to a cloud management platform for storage, and the cloud management platform generates a store operation analysis report based on the received data.

[0185] From the above description, it can be seen that the computer program product provided in the embodiment of the present application collects store data through multiple cameras and sensors to extract customer behavior characteristics, product placement characteristics and environmental parameters. Innovatively construct scene recognition models and behavior analysis models to achieve intelligent recognition of scene states and real-time monitoring of abnormal behaviors. Dynamically match management strategies based on scene feature vectors, and formulate detailed management sub-strategies based on time periods, customer flow and environmental conditions. At the same time, analyze operational data through adaptive optimization models to achieve dynamic updating and optimization of management strategies. This method breaks through the limitations of traditional fixed management models and provides a comprehensive technical solution for unmanned operations of smart stores.

[0186] It should be understood by those skilled in the art that embodiments of the present invention may be provided as methods, devices, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0187] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (apparatus), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0188] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0189] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0190] The present invention uses specific embodiments to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A scenario-adaptive unattended management method for smart stores, characterized in that: The method comprises: Collect multi-channel camera video data and sensor data in the store, perform target detection on the video data to extract customer behavior characteristics and product placement characteristics, perform feature extraction on the sensor data to obtain store environment parameters, input the customer behavior characteristics, the product placement characteristics and the store environment parameters into a pre-trained scene recognition model to obtain a store scene state label, perform semantic analysis on the store scene state label to obtain the main feature parameters of the current scene, wherein the main feature parameters are obtained by using an adaptive weight allocation mechanism and analyzing the contribution of different features in historical data to scene state judgment, which can have a significant impact on scene state judgment At least one of the spatiotemporal dimension features, personnel flow features and environmental state features in the scene state label that is affected by the main feature parameters is used to construct a scene feature vector according to the main feature parameters, the scene feature vector is similarly calculated with the scene description vector in the preset scene management strategy library to select the management strategy with the highest similarity, the time period information, passenger flow information and environmental state information in the scene state label are matched with the preset time period rule library, passenger flow threshold rule library and environmental parameter rule library to obtain corresponding time period management sub-strategies, passenger flow management sub-strategies and environmental management sub-strategies, and the sub-strategies are combined to form a complete scene management strategy; Constructing a customer behavior time series sequence based on the customer behavior characteristics, inputting the customer behavior time series sequence into a pre-trained behavior analysis model to determine whether there is abnormal behavior, and recording the abnormal behavior type, occurrence time and location information into an abnormal event list when abnormal behavior is detected, triggering a corresponding security warning mechanism according to the danger level of the abnormal behavior in the abnormal event list, and calling a corresponding emergency response strategy from a preset emergency response solution library according to the store scene status label and the abnormal behavior type; A statistical analysis of the store scene status label and the customer behavior characteristics in the time dimension is performed to obtain store operation data, and the store operation data is input into an adaptive optimization model to generate store management strategy optimization suggestions. The management strategies in the preset scene management strategy library are dynamically updated according to the management strategy optimization suggestions. At the same time, the store operation data, the abnormal event list and the management strategy optimization suggestions are sent to a cloud management platform for storage. The cloud management platform generates a store operation analysis report based on the received data.

2. The scenario-adaptive unattended management method for smart stores according to claim 1 is characterized in that: The customer behavior characteristics, the commodity placement characteristics and the store environment parameters are input into the pre-trained scene recognition model to obtain a store scene status label, and the store scene status label is semantically analyzed to obtain the main characteristic parameters of the current scene, including: The customer behavior features, the product placement features and the store environment parameters are feature-fused in a preset ratio to construct a multimodal input matrix, the multimodal input matrix is ​​normalized, the normalized feature data is input into a scene recognition model based on a convolutional neural network and a long short-term memory network, and the store scene status label is output through the scene recognition model; The text information in the store scene status label is semantically segmented and word vector mapped through a natural language processing algorithm, and the spatiotemporal dimension features, personnel flow features and environmental status features in the scene label are extracted. The features are weighted and calculated based on preset feature item weights to obtain the main feature parameters of the current scene.

3. The scenario-adaptive unattended management method for smart stores according to claim 1 is characterized in that: The step of constructing a scene feature vector according to the main feature parameters, and calculating the similarity between the scene feature vector and a scene description vector in a preset scene management strategy library to select a management strategy with the highest similarity includes: The main feature parameters are subjected to feature dimension reduction and quantization processing according to a preset feature vector construction rule to generate a scene feature vector of fixed dimension, the numerical features in the scene feature vector are standardized, and the category features in the scene feature vector are subjected to one-hot encoding conversion to generate a standardized scene feature vector; The cosine similarity calculation method is used to calculate the similarity score between the standardized scene feature vector and each scene description vector in the preset scene management strategy library, and the similarity score is screened based on a preset similarity threshold, and the management strategy corresponding to the scene description vector with the highest similarity score that exceeds the similarity threshold is selected as the matching strategy for the current scene.

4. The scenario-adaptive unattended management method for smart stores according to claim 1 is characterized in that: The time period information, passenger flow information and environmental status information in the scene state tag are matched with the preset time period rule base, passenger flow threshold rule base and environmental parameter rule base to obtain corresponding time period management sub-strategies, passenger flow management sub-strategies and environmental management sub-strategies, and the sub-strategies are combined to form a complete scene management strategy, including: Parse the scene state label into time period information, passenger flow information and environment state information, divide the time period information according to a preset time granularity and compare it with the rule items in the time period rule base to obtain a time period management sub-strategy, perform density calculation on the passenger flow information and compare it with the rule items in the passenger flow threshold rule base to obtain a passenger flow management sub-strategy, perform parameter classification on the environment state information and compare it with the rule items in the environment parameter rule base to obtain an environment management sub-strategy; Based on the preset policy priority, the time period management sub-strategy, the passenger flow management sub-strategy and the environment management sub-strategy are subjected to conflict detection and elimination processing, and the conflict-free sub-strategies are combined and optimized through a policy fusion algorithm to generate a complete scene management strategy. Access control instructions, video surveillance instructions and environment adjustment instructions are generated according to the scene management strategy.

5. The scenario-adaptive unattended management method for smart stores according to claim 1 is characterized in that: The step of constructing a customer behavior time series sequence based on the customer behavior characteristics and inputting the customer behavior time series sequence into a pre-trained behavior analysis model to determine whether there is abnormal behavior includes: Sorting the customer behavior features according to timestamps and constructing a behavior feature sequence in segments, dividing the behavior feature sequence into time windows and performing feature completion processing to obtain a fixed-length behavior time series sequence, and normalizing and standardizing the posture features, motion trajectory features, and interaction features in the behavior time series sequence respectively; The standardized behavior time series sequence is input into the abnormal behavior recognition model constructed based on the recurrent neural network. The time series correlation features and local spatial features in the behavior sequence are extracted through the abnormal behavior recognition model. The extracted features are subjected to abnormal detection and classification based on the preset behavior pattern feature library, and the abnormal behavior judgment result is output.

6. The scenario-adaptive unattended management method for smart stores according to claim 1 is characterized in that: When abnormal behavior is detected, the abnormal behavior type, occurrence time and location information are recorded in the abnormal event list, and the corresponding security warning mechanism is triggered according to the danger level of the abnormal behavior in the abnormal event list. At the same time, the corresponding emergency handling strategy is called from the preset emergency handling solution library according to the store scene status label and the abnormal behavior type, including: Determine the type of abnormal behavior detected by the preset behavior classification rules, obtain the spatial coordinate information of the abnormal behavior based on the target tracking algorithm, and write the abnormal event record composed of the abnormal behavior type, timestamp and spatial coordinate information into the abnormal event list, score the abnormal behavior in the abnormal event list according to the preset danger level assessment rules, and determine the warning level based on the danger level score; According to the warning level, the preset security warning rule library is queried to generate a warning instruction, the store scene status label and the abnormal behavior type information are input into the emergency processing strategy selection model constructed based on the decision tree, and the corresponding processing strategy is selected from the preset emergency processing solution library through the emergency processing strategy selection model to generate an emergency processing instruction sequence.

7. The scenario-adaptive unattended management method for smart stores according to claim 1 is characterized in that: The statistical analysis of the store scene status label and the customer behavior characteristics in the time dimension is performed to obtain store operation data, and the store operation data is input into the adaptive optimization model to generate store management strategy optimization suggestions, including: Aggregate the store scene status labels and customer behavior characteristics in segments according to a preset time window, extract multi-dimensional features from the aggregated data to obtain customer flow change trends, customer residence time distribution, product interaction frequency, and environmental parameter change curves, and perform correlation analysis and trend prediction on the various indicators based on a preset data analysis model to generate store operation data; The store operation data is input into a strategy optimization model constructed based on a reinforcement learning algorithm. The execution effect of the existing management strategy is evaluated and scored through the strategy optimization model. The reward value of the strategy execution is calculated based on the preset strategy evaluation index. The policy gradient method is used to tune the parameters of the management strategy to generate strategy optimization suggestions.

8. A scene-adaptive smart store unattended management device, characterized in that: The device comprises: The scene determination module is used to collect video data and sensor data from multiple cameras in the store, perform target detection on the video data to extract customer behavior characteristics and product placement characteristics, perform feature extraction on the sensor data to obtain store environment parameters, input the customer behavior characteristics, the product placement characteristics and the store environment parameters into a pre-trained scene recognition model to obtain a store scene state label, perform semantic analysis on the store scene state label to obtain the main feature parameters of the current scene, wherein the main feature parameters are obtained by using an adaptive weight allocation mechanism and analyzing the contribution of different features in historical data to scene state judgment. At least one of the spatiotemporal dimension features, personnel flow features and environmental state features in the scene state label that has a significant impact is determined, a scene feature vector is constructed according to the main feature parameters, the scene feature vector is similarly calculated with the scene description vector in the preset scene management strategy library to select the management strategy with the highest similarity, the time period information, passenger flow information and environmental state information in the scene state label are matched with the preset time period rule library, passenger flow threshold rule library and environmental parameter rule library respectively to obtain corresponding time period management sub-strategies, passenger flow management sub-strategies and environmental management sub-strategies, and the sub-strategies are combined to form a complete scene management strategy; An emergency response module is used to construct a customer behavior time series sequence based on the customer behavior characteristics, input the customer behavior time series sequence into a pre-trained behavior analysis model to determine whether there is abnormal behavior, and when abnormal behavior is detected, record the abnormal behavior type, occurrence time and location information into an abnormal event list, trigger a corresponding security warning mechanism according to the danger level of the abnormal behavior in the abnormal event list, and call a corresponding emergency response strategy from a preset emergency response solution library according to the store scene status label and the abnormal behavior type; The strategy update module is used to perform statistical analysis on the store scene status label and the customer behavior characteristics in the time dimension to obtain store operation data, input the store operation data into the adaptive optimization model to generate store management strategy optimization suggestions, dynamically update the management strategy in the preset scene management strategy library according to the management strategy optimization suggestions, and send the store operation data, the abnormal event list and the management strategy optimization suggestions to the cloud management platform for storage. The cloud management platform generates a store operation analysis report based on the received data.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the scene-adaptive unmanned smart store management method described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the scene-adaptive smart store unattended management method described in any one of claims 1 to 7 are implemented.

Images