SaaS cloud platform-based pet store management system

Abstract

The invention discloses a pet store management system based on an SaaS cloud platform. The system comprises a plurality of data acquisition modules arranged in a pet store and a cloud platform server communicating with the data acquisition modules. The data acquisition module is equipped with a monitoring camera to acquire pet video stream data and a gas sensor to acquire environmental odor data. The core functions of the cloud platform server are as follows: a data processing module performs image recognition on a video stream to extract pet behavior characteristics, and analyzes peculiar smell data to obtain a concentration value; the pet health management module is used for generating health early warning through a behavior characteristic matching abnormal model, and sending out environment cleaning reminding when the peculiar smell exceeds the standard; and the member linkage module stores associated member and pet health archives, actively pushes associated service or commodity information to the user terminal according to the archive state, and realizes intelligent linkage of pet health monitoring and store service.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of store management, in particular to a pet store management system based on a SaaS cloud platform. BACKGROUND

[0002] With the development of social economy and the improvement of people's living standards, the pet consumption market continues to expand, and pet stores, as important places to provide pet retail, beauty, and boarding services, directly affect user experience and store efficiency. The management efficiency and service quality of traditional pet stores rely heavily on manual observation and experience-based judgment, especially in pet health monitoring and environmental management.

[0003] Currently, stores usually use manual patrols to observe the mental state, appetite, and excretion of pets. This method is not only inefficient and lacks continuity, but also prone to misjudgment or missed judgment due to staff fatigue or experience differences, making it impossible to achieve timely and accurate health abnormality warnings. At the same time, store environmental odors are mainly handled by store employees' sense of smell, lacking objective quantitative indicators and precise positioning capabilities, resulting in insufficient timeliness and targeting of environmental cleaning, affecting the quality of the store environment and pet welfare. In addition, existing automated solutions mostly use single sensors, making it difficult to achieve comprehensive and accurate perception of pet behavior and health status in complex scenarios, especially in situations where pets block each other and light conditions change, resulting in a significant decrease in recognition accuracy. At the same time, existing solutions lack a multi-source information fusion mechanism, making it impossible to correlate behavior data, environmental data, and health status for analysis, and even less likely to achieve intelligent environmental control and precise service recommendations based on analysis results.

[0004] Based on the above reasons, there is an urgent need to design a pet store management system that can achieve intelligent perception, multi-modal data fusion analysis, and automated response. By integrating computer vision, gas sensing, and precise positioning technologies, the system can address the shortcomings of existing solutions in pet health monitoring, environmental management, and customer service, improving the level of intelligent management and service quality of pet stores. SUMMARY

[0005] To address the above problems, the present application provides a pet store management system based on a SaaS cloud platform, aiming to achieve precise monitoring of pet health status, intelligent management of store environment, and active linkage of member services through multi-source information perception and intelligent analysis technology.

[0006] The present application provides a pet store management system based on a SaaS cloud platform, comprising: a plurality of data acquisition modules arranged in a pet store and a cloud platform server in communication connection with the data acquisition modules; The data acquisition module includes a monitoring camera for acquiring pet video stream data and a gas sensor for acquiring environmental odor data. The cloud platform server includes: The data processing module is used to perform image recognition processing on the collected video stream data to extract pet behavior characteristics, and to analyze and process environmental odor data to obtain odor concentration values. The pet health management module is used to match the pet's behavioral characteristics with a preset pet abnormal behavior model. If the match is successful, a pet health abnormality warning is generated. When the odor concentration value exceeds a preset threshold, an environmental cleaning reminder is generated. The member linkage module is used to store and associate member information and the pet health records of their respective pets. The pet health records contain health abnormality warnings and historical behavior data. Based on the status of the pet health records, the module proactively pushes service or product information related to the pet's health status to the corresponding user terminal.

[0007] Furthermore, the image recognition processing of the acquired video stream data specifically includes: The target image frames are extracted from the video stream data at preset time intervals, and noisy image frames, including those that are too dark, too bright, or have incomplete pet outlines, are removed. The RGB three-channel data of the selected target image frame is combined with the grayscale data after the image is converted to grayscale to form four-channel input data, which is then input into the improved convolutional neural network model. The improved convolutional neural network model has a multi-level detection network. It amplifies and focuses the feature map of the pet target step by step, and uses deformable convolutional kernels to extract features from the occluded area. Finally, it outputs pet behavior features, including the pet's movement trajectory, posture, eating and excretion behavior.

[0008] By combining an improved convolutional neural network with a multi-level detection network and deformable convolution, the accuracy of behavior recognition for small targets and occluded pets in complex scenes is significantly improved, and the model can also effectively adapt to various posture changes of pets. The four-channel input data, which combines RGB and grayscale, not only preserves color information but also enhances brightness stability, enabling the model to reliably extract key behavioral features such as the pet's fine movement trajectory, specific posture, and eating and excretion under different lighting conditions, providing more comprehensive data support for health assessment.

[0009] Furthermore, the specific steps of analyzing and processing the environmental odor data include: A concentration gradient model is constructed based on the concentration data collected by gas sensors to locate the sub-region where the odor source is located. The gas sensors are deployed according to the functional zones of the pet store. The odor type is determined by using a pre-defined mapping table between odor types and concentration thresholds.

[0010] By deploying a network of gas sensors in different zones and using a concentration gradient model, centimeter-level precision in odor source location is achieved, significantly shortening the response time to environmental issues. Combined with an intelligent odor type identification mapping table, it can not only distinguish between excrement odors and disease-related odors, but also identify chemical odors such as disinfectants, providing multi-dimensional decision-making basis for environmental management and significantly improving the targeting and efficiency of odor treatment.

[0011] Furthermore, when generating an environmental cleaning reminder, the pet health management module determines whether the type of odor is related to the pet's own health. If the sub-region where the odor source is located overlaps with the real-time activity area of ​​a pet, and the odor type is a specific type that indicates a health abnormality, then the odor data will be synchronized to the corresponding pet's health record. When the pet's behavioral characteristics match the abnormal behavior model, the associated odor data will serve as an auxiliary basis for improving the confidence level of health abnormality warnings.

[0012] An intelligent correlation mechanism between odor data and individual pet health status has been established. When a specific health-related odor is detected and coincides with the pet's activity area in time and space, the system automatically builds an odor-health correlation map, providing multimodal data support for subsequent health warnings. This cross-validation mechanism greatly improves the accuracy and reliability of health anomaly judgment and reduces the possibility of false alarms from a single sensor.

[0013] Furthermore, the data acquisition module also includes a UWB positioning submodule, which consists of a UWB positioning base station deployed in the store and a lightweight UWB tag worn by the pet, used to collect the pet's three-dimensional coordinate data in real time.

[0014] The UWB positioning submodule enables centimeter-level real-time tracking of pet locations, providing not only precise spatial coordinate data but also calculating movement speed and activity trajectory. This provides a rich spatiotemporal data foundation for behavior analysis, area control, and health management, effectively solving problems such as occlusion, lighting changes, and blind spots in pure visual monitoring.

[0015] Furthermore, the cloud platform server also includes an electronic fence management module, used for: Based on the physical layout of the stores, dynamic electronic fence boundaries are drawn in the system. It receives and processes pet coordinate data collected by the UWB positioning submodule in real time; When a pet enters the boundary of the electronic fence, it triggers the corresponding area's surveillance camera to capture an image frame; The system uses image recognition algorithms to determine whether a pet entering the boundary area is being carried by a store employee. If the system identifies that a pet is not being carried by a store employee, an alert is triggered.

[0016] The electronic fence management module enables intelligent monitoring and tiered early warning of pet safety areas. When a pet approaches a dangerous area, the system will issue a primary alarm. Upon entering a dangerous area, it will immediately trigger image verification and emergency warning, forming a multi-layered protection system that significantly reduces the risk of pets getting lost or having accidents, while also alleviating the monitoring burden on store staff.

[0017] Furthermore, the process of matching the pet's behavioral characteristics with a preset abnormal pet behavior model, and generating a pet health abnormality warning if a match is successful, specifically includes: Receive consecutive multi-frame images from a video stream and extract fused feature vectors containing spatial and temporal information through a three-dimensional residual convolutional neural network; wherein, during the processing of the three-dimensional residual convolutional neural network, a motion-based attention mechanism is introduced to assign differentiated weights to feature maps at different spatiotemporal locations in consecutive frames in order to focus on key regions related to behavior. The extracted fusion feature vectors are input into a bidirectional long short-term memory network to model and filter the temporal dependencies of behavioral features from both forward and backward directions, so as to capture the bidirectional dependencies of behavioral context and obtain the final features. The final features are input into a fully connected layer and a Softmax classifier to calculate the matching confidence of the pet behavior with multiple preset abnormal behavior patterns; when the highest confidence exceeds a preset threshold, a corresponding pet health abnormality warning is generated.

[0018] By integrating a deep learning architecture that combines 3D convolution and bidirectional LSTM, the spatiotemporal features and behavioral context information in video sequences are fully explored. This not only identifies obvious abnormal behaviors but also detects subtle early signs of abnormality. The introduction of an attention mechanism enables the model to focus on key behavioral segments, significantly improving the accuracy of abnormal behavior recognition and early warning capabilities.

[0019] Furthermore, the introduced motion-based attention mechanism is the ActionNet mechanism. This mechanism models the spatiotemporal importance of the input features, the inter-channel dependencies, and the motion changes between adjacent frames through parallel spatiotemporal attention sub-modules, channel attention sub-modules, and motion attention sub-modules, respectively. The output features of the three sub-modules are added together and then fused to optimize the feature extraction process.

[0020] By employing the ActionNet mechanism to process spatiotemporal, channel, and motion features in parallel, multi-dimensional feature optimization is achieved, enabling the model to more accurately capture key health-related behavioral features. This improves recognition performance while maintaining computational efficiency, making it particularly suitable for real-time processing of multi-channel video data from pet stores.

[0021] Furthermore, based on the status of the pet's health record, service or product information related to the pet's health status is proactively pushed to the corresponding user terminal, specifically including: Establish and maintain an association mapping table, which defines the correspondence between various pet health statuses and recommended services or products; The system monitors changes in the pet's health record in real time, and automatically triggers a service recommendation process when new health abnormality warnings appear or historical behavior data indicates a deteriorating health trend. Based on the specific health status at the time of triggering, query the associated mapping table to generate a personalized push plan containing recommended service or product information; Based on this push notification scheme, prompt messages are proactively sent to the corresponding user terminals via a message interface.

[0022] By establishing an intelligent mapping relationship between health status and service recommendations, personalized and precise member service recommendations are achieved. The automatic triggering mechanism based on real-time health data ensures that the most valuable service information is pushed to users at the best time, which not only improves customer experience and satisfaction, but also creates significant value-added service revenue for stores.

[0023] Furthermore, the system also includes an environmental cleaning module, used to automatically clean the environment upon receiving an environmental cleaning reminder, specifically including: Receive environmental cleaning reminders from the pet health management module, which include information on the type, concentration, and source location of odors. Based on the received reminder information, control commands are automatically generated to intelligently start and stop the ventilation system, air purifier or disinfection atomization device in the corresponding area; The system generates work orders that include specific cleaning locations and recommended actions, and automatically assigns them to store staff's mobile terminals.

[0024] The environmental cleaning module enables an intelligent environmental management system that automatically selects the most appropriate treatment method (ventilation, purification, or disinfection) based on the type and concentration of odors, and generates work orders containing specific locations and treatment suggestions. This significantly improves environmental management efficiency, ensures that the store's environmental quality is always kept at its best, and significantly reduces manual management costs.

[0025] Compared with existing technologies, the beneficial effects of this invention are as follows: This system, based on a SaaS cloud platform architecture, effectively solves the problems of long deployment cycles, high maintenance costs, and poor scalability of existing local deployment systems. Stores do not need to build local servers; they only need to deploy data collection devices to connect, significantly reducing initial investment and subsequent maintenance costs. It also supports real-time data aggregation and unified management across multiple stores. Simultaneously, through a combined data collection module of "surveillance camera + gas sensor," it overcomes the limitations of fragmented data in existing technologies, simultaneously acquiring pet video streams and environmental odor data and connecting them to the cloud platform, providing multi-dimensional basic data for subsequent management. The cloud platform's data processing module transforms the collected data into behavioral characteristics and quantitative concentration values, which are then analyzed through pet health... The pet health management module automatically issues health anomaly warnings and provides environmental cleaning reminders, eliminating the need for manual monitoring and improving the timeliness of detecting pet abnormalities. It also avoids indiscriminate cleaning, enhancing targeted environmental management and significantly reducing reliance on manual labor and operational costs. Furthermore, the membership linkage module deeply integrates member information with pet health records, recording health anomaly warnings and historical behavioral data in real time. Based on the record status, it proactively pushes pet health-related services or products, breaking through the limitations of existing membership management systems where information is fragmented, service delivery is passive, and accuracy is low. This improves user experience and trust, helps stores deliver services precisely, enhances user stickiness and business conversion capabilities, and achieves comprehensive optimization of pet store management efficiency, service quality, and cost-effectiveness. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of this drawing or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this drawing. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0027] Figure 1 This is a schematic diagram of the system principle of the present invention. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described and illustrated below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. All other embodiments obtained by those skilled in the art based on the embodiments provided by this invention without inventive effort are within the scope of protection of this invention.

[0029] This invention provides a pet store management system based on a SaaS cloud platform, such as... Figure 1 As shown, it specifically includes: Multiple data acquisition modules are deployed in the pet store, and a cloud platform server is connected to the data acquisition modules.

[0030] The data acquisition module includes a surveillance camera for collecting pet video stream data and a gas sensor for collecting environmental odor data.

[0031] Specifically, the surveillance cameras are high-definition network cameras, evenly deployed in all areas of the store to ensure no blind spots; gas sensors are deployed according to functional zones and can detect gases such as NH3, H2S, and VOCs.

[0032] Furthermore, the data acquisition module also includes a UWB positioning submodule, which consists of UWB positioning base stations deployed in the store and lightweight UWB tags worn by the pets, used to collect the pets' three-dimensional coordinate data in real time.

[0033] The cloud platform server includes the following modules: The data processing module is used to perform image recognition processing on the collected video stream data to extract pet behavior characteristics, and to analyze and process environmental odor data to obtain odor concentration values.

[0034] Furthermore, the following steps are used to extract pet behavioral characteristics: The target image frames are extracted from the video stream data at preset time intervals, and noisy image frames, including those that are too dark, too bright, or have incomplete pet outlines, are removed.

[0035] Specifically, the image frame acquisition adopts a fixed time interval sampling strategy, and the sampling interval is set according to the video frame rate, which significantly reduces the computational load while ensuring the continuity of the behavior.

[0036] The following methods are used to remove noisy image frames: calculate the average brightness value of the image, set two thresholds, and remove image frames whose average brightness value exceeds this range; use a background subtraction algorithm combined with edge detection, and determine that the contour is incomplete when the detected contour area is less than the set threshold or the main boundary of the contour exceeds the image boundary; and remove blurred frames by calculating the Laplacian variance of the image and setting a threshold.

[0037] The RGB three-channel data of the selected target image frame is combined with the grayscale data after the image is converted to grayscale to form four-channel input data, which is then input into the improved convolutional neural network model.

[0038] Specifically, a weighted method is used to convert RGB images to grayscale images using the formula: Gray = 0.299 × R + 0.587 × G + 0.114 × B, preserving brightness information that conforms to human visual perception. The RGB three-channel data and the single-channel grayscale data are concatenated along the channel dimension to form a four-channel input tensor. The mean and variance are standardized for each channel using the mean and variance parameters from the ImageNet dataset to accelerate model convergence.

[0039] Grayscale images are far less sensitive to changes in lighting than RGB images. In environments with fluctuating light levels, colors can change drastically, but the outlines and motion trajectories of objects remain relatively stable in grayscale images. Providing the model with both types of information simultaneously, allowing it to learn to fuse RGB and grayscale image information, enables the model to utilize color information without becoming overly reliant on it, thus adapting to more complex lighting environments.

[0040] The improved convolutional neural network model has a multi-level detection network. It amplifies and focuses the feature map of the pet target step by step, and uses deformable convolutional kernels to extract features from the occluded area. Finally, it outputs pet behavior features, including the pet's movement trajectory, posture, eating and excretion behavior.

[0041] Specifically, the improved convolutional neural network model uses ResNet-50 as its backbone. The first few layers of the network process low-resolution, large-area images for initial pet localization; the later layers gradually zoom in and focus on the previously localized area for refined pet analysis and feature extraction. This ensures that the pet can be clearly identified regardless of its size in the image.

[0042] Deformable convolution is an advanced convolutional technique where the sampling point positions are not fixed but adaptively shifted according to the content of the current image. Employing deformable convolution kernels to extract features from occluded regions enables the system to achieve high-precision pet behavior recognition in complex environments. This approach fully considers various challenges in practical deployments, including changes in lighting, occlusion issues, and small object detection, ensuring the system's practicality and reliability.

[0043] The following steps are used to obtain the odor concentration value: A concentration gradient model is constructed based on the concentration data collected by gas sensors to locate the sub-region where the odor source is located. The gas sensors are deployed according to the functional zones of the pet store. Because the sources and types of odors vary from area to area, gas sensors are deployed according to the functional zones of the store, such as the sample collection area, beauty operation area, toilet area, rest area, and retail shelf area. One or more gas sensors are deployed in each area to form a sensor network covering the entire store.

[0044] The system receives readings from all sensors in the network in real time. Each reading contains two key pieces of information: the gas concentration value and the sensor's location identifier. The system maintains a virtual digital map of the stores in the background, mapping the location of each sensor to a coordinate point on the map and using the concentration value reported by that point as its weight. Then, through a spatial interpolation algorithm, it calculates a continuously distributed, smooth concentration value across the entire map, generating a concentration gradient model similar to a topographic map.

[0045] The system can identify the most likely odor source center by finding the extreme points in this concentration gradient model. By analyzing the gradient of concentration diffusion and decay from the center to the surrounding areas, the odor source can be accurately defined as a sub-region.

[0046] The odor type is determined by using a pre-defined mapping table between odor types and concentration thresholds.

[0047] Before system deployment, a knowledge base is pre-established through experiments and data analysis, namely a mapping table between odor types and concentration thresholds. For example, for excrement odors, the characteristic gases are set as ammonia (NH3) and hydrogen sulfide (H2S). The system is set to determine the current odor type as excrement odor when the concentration of ammonia or hydrogen sulfide continuously exceeds the preset concentration threshold. When a sensor in a certain sub-area detects that the gas concentration exceeds the standard, the system analyzes the component characteristic signal of the gas and matches the component characteristic signal with the data in the mapping table. By comparing the gas type and concentration level, the system can determine which type the current odor is most likely to belong to.

[0048] The pet health management module is used to match the pet's behavioral characteristics with a preset pet abnormal behavior model. If the match is successful, a pet health abnormality warning is generated. When the odor concentration value exceeds a preset threshold, an environmental cleaning reminder is generated.

[0049] Furthermore, feature behavior matching is performed using the following methods: Receive consecutive multi-frame images from a video stream and extract fused feature vectors containing spatial and temporal information using a three-dimensional residual convolutional neural network (Res3D). In the processing of the three-dimensional residual convolutional neural network, a motion-based attention mechanism is introduced to assign differentiated weights to feature maps at different spatiotemporal locations in consecutive frames in order to focus on key regions related to behavior.

[0050] Specifically, the motion-based attention mechanism introduced is the ActionNet mechanism. This mechanism models the spatiotemporal importance of input features, inter-channel dependencies, and motion changes between adjacent frames through parallel spatiotemporal attention sub-modules, channel attention sub-modules, and motion attention sub-modules, respectively. The output features of the three sub-modules are added together and then fused to strengthen key features and weaken redundant information, thereby optimizing the feature extraction process.

[0051] Abnormal pet behavior is a series of continuous actions, therefore it is necessary to extract both spatial features (such as pet posture and body part positions) and temporal features (such as the trend of posture changes over frames). A 3D residual convolutional neural network is employed, which can capture both spatial and temporal features simultaneously. To improve the specificity and robustness of feature extraction, a motion-based attention mechanism, ActionNet, is introduced, whose structure includes three parallel sub-modules: Spatiotemporal attention submodule: Focuses on the importance of different time and spatial locations in the video, highlighting key areas related to behavior.

[0052] Channel Attention Submodule: Convolution generates multiple feature channels, each of which may capture different features. This submodule learns the dependencies between different channels, enhances the channel responses that are useful for the current behavior, and suppresses useless channels. The motion attention submodule is specifically designed to capture motion changes between adjacent frames, enhancing dynamic behavioral information. The output features of these three sub-modules are added and fused together to comprehensively optimize the feature representation, highlight key information, and suppress redundancy.

[0053] The extracted fusion feature vectors are input into a bidirectional long short-term memory network to model and filter the temporal dependencies of behavioral features in both forward and backward directions, so as to capture the bidirectional dependencies of behavioral context and obtain the final features.

[0054] The final features are input into a fully connected layer and a Softmax classifier to calculate the matching confidence of the pet behavior with multiple preset abnormal behavior patterns; when the highest confidence exceeds a preset threshold, a corresponding pet health abnormality warning is generated.

[0055] Specifically, the fully connected layer maps the final feature vector to a higher-dimensional feature space; the Softmax classifier receives the output of the fully connected layer and converts it into a probability distribution, which represents the confidence (probability) that the currently observed behavior belongs to each preset abnormal behavior pattern. The system reads the confidence vector output by Softmax, selects the highest confidence value, and compares this highest confidence value with a preset threshold. When the highest confidence value exceeds the preset threshold, and the behavior category corresponding to this highest confidence value belongs to abnormal behavior, the system generates a corresponding pet health abnormality warning.

[0056] Furthermore, when generating environmental cleaning reminders, the pet health management module determines whether the type of odor is related to the pet's own health. If the sub-region where the odor source is located overlaps with the real-time activity area of ​​a pet, and the odor type is a specific type that indicates a health abnormality, such as skin bacterial metabolism odor or oral disease odor, then the odor data will be synchronized to the corresponding pet's health record. When the pet's behavioral characteristics match the abnormal behavior model, the associated odor data will serve as an auxiliary basis for improving the confidence level of health abnormality warnings.

[0057] The behavioral model outputs a preliminary confidence score. The system automatically retrieves the pet's health record to check for any relevant odor data recorded in the past few days. The odor data is then input into the system's preset algorithm model, such as a Bayesian network, to obtain the final comprehensive confidence score, thereby improving the accuracy and reliability of health abnormality judgment.

[0058] Furthermore, this system also includes an environmental cleaning module, which automatically cleans the environment upon receiving an environmental cleaning reminder, specifically including: Receive environmental cleaning reminders from the pet health management module, which include information on the type, concentration, and source location of odors. Based on the received reminder information, control commands are automatically generated to intelligently start and stop the ventilation system, air purifier, or disinfection atomization device in the corresponding area, ensuring that the spread of odors is suppressed and initial treatment is carried out as soon as possible without human intervention, thus ensuring the timeliness of environmental quality. The system generates work orders that include specific cleaning locations and recommended actions, and automatically assigns them to store staff's mobile terminals.

[0059] The electronic fence management module is used to prevent pets from entering dangerous areas or leaving safe areas without permission, thereby avoiding pets getting lost, injured, or causing other safety accidents. Specifically, it employs the following methods: Based on the physical layout of the stores, dynamic electronic fence boundaries are drawn in the system. It receives and processes pet coordinate data collected by the UWB positioning submodule in real time; When a pet enters the boundary of the electronic fence, it triggers the corresponding area's surveillance camera to capture an image frame; The system uses image recognition algorithms to determine whether a pet entering the boundary area is being carried by a store employee. If the system identifies that a pet is not being carried by a store employee, an alert is triggered.

[0060] Specifically, administrators can virtually delineate areas where pets are prohibited on the digital map of the backend system via a graphical interface. The shape and location of the electronic fence are not fixed; administrators can redraw it on the system map at any time according to adjustments in the store layout.

[0061] The UWB tags worn by pets continuously send signals to base stations deployed within the store. By calculating the time difference of signal arrival, the base stations can determine the precise 3D coordinates of the pet's UWB tag in real time. The electronic fence management module continuously receives this coordinate data and updates the position of each pet on a virtual map in the background in real time, continuously determining whether the current pet's coordinates have entered the polygonal boundary of the electronic fence. Once the UWB detects a boundary breach, it immediately calls upon the surveillance camera closest to the incident location with the best viewing angle and controls that camera to capture one or more images.

[0062] Image recognition algorithms identify all targets in an image, including people and pets, and analyze the spatial relationship between people and pets. When the pet is identified as being carried by a store employee, the system considers this normal behavior and takes no action; when the pet is identified as not being carried by a store employee, the system considers this an abnormal security event and immediately triggers a final warning.

[0063] The member linkage module is used to store and associate member information and the pet health records of their respective pets. The pet health records contain health abnormality warnings and historical behavior data. Based on the status of the pet health records, the module proactively pushes service or product information related to the pet's health status to the corresponding user terminal.

[0064] Specifically, an association mapping table is established and maintained, which defines the correspondence between various pet health statuses and recommended services or products. This association mapping table is dynamic, and the administrator can update it in real time based on the store's product inventory, service packages, promotional activities, etc., to ensure that the most relevant and available items are recommended.

[0065] The system monitors changes in the pet's health record in real time, and automatically triggers a service recommendation process when new health abnormalities are detected or historical behavioral data indicates a deteriorating health trend.

[0066] Based on the specific health status at the time of triggering, the associated mapping table is queried to generate a personalized push plan containing recommended service or product information.

[0067] Specifically, the system will analyze the specific health status that triggered this recommendation, then query the aforementioned association mapping table to obtain a basic list of recommended items. Based on the list of recommended items, pet profile, and customer consumption history, a personalized push plan will be generated.

[0068] Based on this push notification scheme, prompt messages are proactively sent to the corresponding user terminals via a message interface.

[0069] It should be noted that the present invention is not limited to the above-described embodiments. The above embodiments are merely examples, and any embodiments that have the same structure and perform the same effects as the technical concept within the scope of the present invention are included within the scope of the present invention. Furthermore, various modifications that can be conceived by those skilled in the art to the embodiments, and other ways of constructing by combining some of the constituent elements of the embodiments, without departing from the spirit of the present invention, are also included within the scope of the present invention.

Claims

1. A pet store management system based on a SaaS cloud platform, characterized in that: include: Multiple data acquisition modules deployed within the pet store and a cloud platform server communicating with the data acquisition modules; The data acquisition module includes a monitoring camera for acquiring pet video stream data and a gas sensor for acquiring environmental odor data. The cloud platform server includes: The data processing module is used to perform image recognition processing on the collected video stream data to extract pet behavior characteristics, and to analyze and process environmental odor data to obtain odor concentration values. The pet health management module is used to match the pet's behavioral characteristics with a preset pet abnormal behavior model. If the match is successful, a pet health abnormality warning is generated. When the odor concentration value exceeds a preset threshold, an environmental cleaning reminder is generated. The member linkage module is used to store and associate member information and the pet health records of their respective pets. The pet health records contain health abnormality warnings and historical behavior data. Based on the status of the pet health records, the module proactively pushes service or product information related to the pet's health status to the corresponding user terminal.

2. The pet store management system based on a SaaS cloud platform as described in claim 1, characterized in that, The image recognition processing of the acquired video stream data specifically includes: The target image frames are extracted from the video stream data at preset time intervals, and noisy image frames, including those that are too dark, too bright, or have incomplete pet outlines, are removed. The RGB three-channel data of the selected target image frame is combined with the grayscale data after the image is converted to grayscale to form four-channel input data, which is then input into the improved convolutional neural network model. The improved convolutional neural network model has a multi-level detection network. It amplifies and focuses the feature map of the pet target step by step, and uses deformable convolutional kernels to extract features from the occluded area. Finally, it outputs pet behavior features, including the pet's movement trajectory, posture, eating and excretion behavior.

3. The pet store management system based on a SaaS cloud platform as described in claim 1, characterized in that, The specific steps of analyzing and processing the environmental odor data include: A concentration gradient model is constructed based on the concentration data collected by gas sensors to locate the sub-region where the odor source is located. The gas sensors are deployed according to the functional zones of the pet store. The odor type is determined by using a pre-defined mapping table between odor types and concentration thresholds.

4. The pet store management system based on a SaaS cloud platform as described in claim 3, characterized in that, When generating an environmental cleaning reminder, the pet health management module determines whether the type of odor is related to the pet's own health. If the sub-region where the odor source is located overlaps with the real-time activity area of ​​a pet, and the odor type is a specific type that indicates a health abnormality, then the odor data will be synchronized to the corresponding pet's health record. When the pet's behavioral characteristics match the abnormal behavior model, the associated odor data will serve as an auxiliary basis for improving the confidence level of health abnormality warnings.

5. The pet store management system based on a SaaS cloud platform as described in claim 1, characterized in that, The data acquisition module also includes a UWB positioning submodule, which consists of a UWB positioning base station deployed in the store and a lightweight UWB tag worn by the pet, used to collect the pet's three-dimensional coordinate data in real time.

6. The pet store management system based on a SaaS cloud platform as described in claim 5, characterized in that, The cloud platform server also includes an electronic fence management module, used for: Based on the physical layout of the stores, dynamic electronic fence boundaries are drawn in the system. It receives and processes pet coordinate data collected by the UWB positioning submodule in real time; When a pet enters the boundary of the electronic fence, it triggers the corresponding area's surveillance camera to capture an image frame; The system uses image recognition algorithms to determine whether a pet entering the boundary area is being carried by a store employee. If the system identifies that a pet is not being carried by a store employee, an alert is triggered.

7. The pet store management system based on a SaaS cloud platform as described in claim 1, characterized in that, The process involves matching the pet's behavioral characteristics with a preset abnormal pet behavior model. If a match is successful, a pet health abnormality warning is generated. This specifically includes: Receive consecutive multi-frame images from a video stream and extract fused feature vectors containing spatial and temporal information through a three-dimensional residual convolutional neural network; wherein, during the processing of the three-dimensional residual convolutional neural network, a motion-based attention mechanism is introduced to assign differentiated weights to feature maps at different spatiotemporal locations in consecutive frames in order to focus on key regions related to behavior. The extracted fusion feature vectors are input into a bidirectional long short-term memory network to model and filter the temporal dependencies of behavioral features from both forward and backward directions, so as to capture the bidirectional dependencies of behavioral context and obtain the final features. The final features are input into a fully connected layer and a Softmax classifier to calculate the matching confidence of the pet behavior with multiple preset abnormal behavior patterns; when the highest confidence exceeds a preset threshold, a corresponding pet health abnormality warning is generated.

8. The pet store management system based on a SaaS cloud platform as described in claim 7, characterized in that, The introduced motion-based attention mechanism is the ActionNet mechanism. This mechanism models the spatiotemporal importance of input features, inter-channel dependencies, and motion changes between adjacent frames through parallel spatiotemporal attention sub-modules, channel attention sub-modules, and motion attention sub-modules, respectively. The output features of the three sub-modules are added together and then fused to optimize the feature extraction process.

9. The pet store management system based on a SaaS cloud platform as described in claim 1, characterized in that, Based on the status of the pet's health record, service or product information related to the pet's health status is proactively pushed to the corresponding user terminal, specifically including: Establish and maintain an association mapping table, which defines the correspondence between various pet health statuses and recommended services or products; The system monitors changes in the pet's health record in real time, and automatically triggers a service recommendation process when new health abnormality warnings appear or historical behavioral data indicates a deteriorating health trend. Based on the specific health status at the time of triggering, query the associated mapping table to generate a personalized push plan containing recommended service or product information; Based on this push notification scheme, prompt messages are proactively sent to the corresponding user terminals via a message interface.

10. The pet store management system based on a SaaS cloud platform as described in claim 1, characterized in that, The system also includes an environmental cleaning module, used to automatically clean the environment upon receiving an environmental cleaning reminder, specifically including: Receive environmental cleaning reminders from the pet health management module, which include information on the type, concentration, and source location of odors. Based on the received reminder information, control commands are automatically generated to intelligently start and stop the ventilation system, air purifier or disinfection atomization device in the corresponding area; The system generates work orders that include specific cleaning locations and recommended actions, and automatically assigns them to store staff's mobile terminals.

Images