Pet behavior analysis system based on visual intelligence

Abstract

The invention discloses a pet behavior analysis system based on visual intelligence, and relates to the technical field of pet intelligent monitoring. According to the invention, an abnormal pet behavior is marked as a target behavior; determining a scene type according to the video data of the target behavior and the corresponding auxiliary behavior, and determining a behavior cause of the pet based on the target behavior, the auxiliary behavior and the scene type; according to the invention, pet behaviors are comprehensively described by means of multi-dimensional monitoring data, so that one-sidedness of a single data dimension is avoided; the preset reference data covers a general scene, and the scientificity of abnormal behavior judgment is improved through hierarchical analysis and weighted fusion design of the behavior recognition model; according to the method, the target behavior is associated with the auxiliary behavior to expand the data dimension, accurate positioning of the behavior inducement is realized in combination with scene recognition, the problem that the core inducement is difficult to lock by single behavior analysis is solved, and systematic and reliable technical support is provided for pet behavior interpretation.

Description

Technical Field

[0001] This invention belongs to the field of pet intelligent monitoring technology, specifically a pet behavior analysis system based on visual intelligence. Background Technology

[0002] Pet behavior analysis aims to interpret the physiological needs, emotional states, or environmental adaptation issues behind pet behaviors, avoiding misinterpretations of instinctive reactions or ignoring potential health / psychological signals. The analysis process involves observing behavioral scenarios, investigating health and environmental factors, and combining these with species instincts to make scientific judgments, ultimately accurately addressing problematic behaviors, ensuring pet welfare, and fostering a more harmonious and stable relationship between humans and pets.

[0003] Current pet behavior analysis technologies often rely on single-dimensional data collection and analysis, either by monitoring pet movement characteristics through video or by collecting pet physiological data through physiological devices. This results in an incomplete portrayal of pet behavior and is prone to misjudgment due to data bias. Moreover, it often focuses on isolated judgments of single abnormal behaviors, making it impossible to pinpoint the core causes of abnormal behaviors and understand the underlying physiological or psychological needs of pets, thus failing to provide effective support for pet behavior intervention.

[0004] This invention provides a pet behavior analysis system based on visual intelligence to solve the above-mentioned technical problems. Summary of the Invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes a pet behavior analysis system based on visual intelligence.

[0006] To achieve the above objectives, a first aspect of the present invention provides a pet behavior analysis system based on visual intelligence, comprising:

[0007] Central processing module: used to mark abnormal pet behaviors as target behaviors; determine the scene type based on video data of the target behavior and corresponding auxiliary behaviors; wherein, pet behaviors include motion characteristics, vocal characteristics, or physiological characteristics, and auxiliary behaviors are temporally correlated with the target behavior; and,

[0008] Used to determine the behavioral triggers of pets based on target behavior, auxiliary behavior, and scenario type; among which, behavioral triggers include physiological triggers or psychological triggers.

[0009] In one possible embodiment, generating pet behavior based on monitoring data includes:

[0010] Extract monitoring data; the monitoring data includes video data, audio data, and physiological data;

[0011] Motion features are extracted from video data, sound features from audio data, and physiological features from physiological data; these motion features, sound features, and physiological features are then aligned and integrated into pet behavior.

[0012] In one possible embodiment, determining whether a pet's behavior is abnormal includes:

[0013] Extract the pet's preset baseline data; the preset baseline data includes action baseline data, vocal baseline data, and physiological baseline data;

[0014] The pet's behavior and preset benchmark data are input into the behavior recognition model, which outputs the judgment result. The behavior recognition model includes a data preprocessing layer, an independent analysis layer, and a weighted fusion layer. The independent analysis layer includes a motion analysis unit, a sound analysis unit, and a physiological analysis unit.

[0015] In one possible embodiment, determining the scene type based on video data of the target behavior and the corresponding auxiliary behavior includes:

[0016] Identify the auxiliary behaviors corresponding to the target behavior; where auxiliary behaviors refer to pet behaviors that are temporally related to the target behavior.

[0017] Video data of target behavior and auxiliary behavior are extracted, and the video data is analyzed through a scene recognition model to obtain the scene type. The scene recognition model includes a data input layer, a data preprocessing layer, a feature extraction layer, and a classification output layer. The feature extraction layer is used to extract features from the video data.

[0018] In one possible embodiment, determining the auxiliary behavior corresponding to the target behavior includes:

[0019] Extract several pet behaviors before or after the target behavior;

[0020] Using the target behavior as a baseline, analyze whether several pet behaviors are temporally related to the target behavior, either forward or backward. If so, mark the pet behavior as an auxiliary behavior. The temporal correlation is determined from two perspectives: time span and time interval.

[0021] In one possible embodiment, the behavioral triggers of the pet are determined based on the target behavior, auxiliary behavior, and scenario type, including:

[0022] The trigger identification model is invoked; the trigger identification model includes a data input layer, a data preprocessing layer, a core classification layer, and a result output layer. The core classification layer is used to determine the behavioral trigger.

[0023] After standardizing the target behavior, auxiliary behavior, and scenario type, the data is input into the trigger recognition model to obtain the pet's behavioral triggers.

[0024] In one possible embodiment, determining whether a pet's behavior is abnormal includes:

[0025] Extract the pet's individual baseline data; the individual baseline data is constructed based on monitoring data of the pet's health status;

[0026] Pet behavior and personality baseline data are input into the behavior recognition model, which outputs the judgment result. The behavior recognition model includes a data preprocessing layer, an independent analysis layer, and a weighted fusion layer. The independent analysis layer includes a motion analysis unit, a sound analysis unit, and a physiological analysis unit.

[0027] In one possible embodiment, constructing personality baseline data for the pet includes:

[0028] When a pet is found to be healthy after a check-up, the pet's monitoring data is extracted as reference data.

[0029] Based on preset baseline data, the pet's individual baseline data is generated according to reference data; the individual baseline data contains the same data items as the preset baseline data.

[0030] In one possible embodiment, after identifying the behavioral triggers, intervention recommendations are generated based on the pet's physical examination data, including:

[0031] Extract the pet's physical examination data and extract abnormal items from the data as candidate items;

[0032] Match the related items of the behavioral triggers in the candidate options to obtain the target items; generate intervention suggestions based on the behavioral triggers and target items.

[0033] Compared with the prior art, the beneficial effects of the present invention are:

[0034] 1. This invention generates pet behavior data including action, sound, and physiological features through monitoring data. Based on preset benchmark data and a behavior recognition model containing independent analysis and weighted fusion layers, it determines whether the pet behavior is abnormal. Abnormal pet behavior is marked as the target behavior, and then associated with auxiliary behaviors that have a time correlation. The scene type is determined by a scene recognition model containing a feature extraction layer. Finally, the cause recognition model outputs the behavior cause by combining the target behavior, auxiliary behaviors, and scene type. This invention comprehensively portrays pet behavior with multi-dimensional monitoring data, avoiding the one-sidedness of a single data dimension. The preset benchmark data covers common scenarios, and the hierarchical analysis and weighted fusion design of the behavior recognition model improves the scientific nature of abnormal behavior judgment. By associating target behavior with auxiliary behaviors, the data dimensions are expanded, and the scene recognition enables accurate positioning of behavior causes, solving the problem that single behavior analysis is difficult to pinpoint the core cause. This provides systematic and reliable technical support for interpreting pet behavior.

[0035] 2. This invention employs personalized benchmark data constructed based on monitoring data of pets' health status. It generates personalized benchmark data consistent with preset benchmark data items based on reference data of the pet's health status, and then uses a behavior recognition model with the same structure to determine abnormal pet behavior. This invention constructs exclusive personalized benchmark data for individual differences in pet breeds, growth experiences, etc., compensating for the insufficient coverage of general preset benchmark data. The personalized benchmark data is constructed based on the pet's own health status, making it more closely aligned with the pet's actual behavioral characteristics. Combined with the hierarchical analysis logic of the behavior recognition model, it significantly reduces the probability of misjudgment in niche breeds or rare behavioral scenarios, greatly improving the accuracy and specificity of abnormal behavior judgment, and making behavior analysis more adaptable to individual pet characteristics.

[0036] 3. This invention extracts pet health checkup data and filters abnormal items as candidate items, matches target items associated with behavioral triggers, and generates intervention suggestions by combining behavioral triggers and target items. This invention deeply integrates behavioral triggers with pet health checkup data, using abnormal items in the health checkup data to pinpoint specific health problems associated with the triggers, avoiding the limitation of relying solely on behavioral triggers to locate specific influencing factors. By pre-setting association relationships, the efficiency of target item matching is improved. The generated intervention suggestions rely on both the root cause analysis of behavioral triggers and the health evidence from health checkup data, ensuring the pertinence and operability of the suggestions. It realizes a closed loop from behavioral abnormality identification, trigger analysis to intervention guidance, effectively helping to solve pet health problems in a timely manner and ensuring pet welfare. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of the present invention 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 the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is a schematic diagram illustrating the working steps of the pet behavior analysis system in Embodiment 1 of the present invention;

[0039] Figure 2 This is a schematic diagram of the system principle of the pet behavior analysis system in Embodiment 1 of the present invention;

[0040] Figure 3 This is a schematic diagram illustrating the structural principle of the behavior recognition model in Embodiment 1 of the present invention;

[0041] Figure 4 This is a schematic diagram of the scene recognition model in Embodiment 1 of the present invention;

[0042] Figure 5This is a schematic diagram of the cause identification model in Embodiment 1 of the present invention;

[0043] Figure 6 This is a schematic diagram of the method for judging pet behavior based on personality benchmark data in Embodiment 2 of the present invention;

[0044] Figure 7 This is a schematic diagram of the method steps for generating intervention suggestions based on behavioral triggers in Embodiment 3 of the present invention. Detailed Implementation

[0045] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] Example 1:

[0047] Please see Figures 1-2 The first aspect of the present invention provides a visual intelligence-based pet behavior analysis system, including a central processing module: for marking abnormal pet behaviors as target behaviors; determining scene types based on video data of the target behaviors and corresponding auxiliary behaviors; and for determining the behavioral triggers of the pet based on the target behaviors, auxiliary behaviors, and scene types.

[0048] This application extracts pet behavior based on monitoring data, which mainly includes video data, audio data, and physiological data. Video data is collected through cameras installed in the pet's activity space. If the activity space is large, multiple cameras can be used in tandem to collect video data. The video data is primarily used to identify the pet's movement characteristics and the scene features of its activity space. Audio data is collected through sound sensors and is mainly used to identify the pet's vocal characteristics. Physiological data is collected through smart pet devices and wearable pet devices and is mainly used to identify the pet's physiological characteristics.

[0049] After cameras, sound sensors, and wearable pet devices collect monitoring data, it can be sent to a database for storage via the network. During pet analysis, the central analysis module can extract the monitoring data from the database and analyze pet behavior based on it. Of course, if immediate analysis of pet behavior is required, the monitoring data can also be sent directly to the central processing module.

[0050] In the process of identifying behavioral triggers, it is necessary to generate pet behavior data based on monitoring data and determine whether the pet behavior is abnormal. Pet behavior includes the pet's movement characteristics, vocal characteristics, and physiological characteristics. When identifying whether a pet behavior is abnormal, the movement characteristics, vocal characteristics, and physiological characteristics need to be compared item by item with the corresponding preset benchmark data. The comparison results are then weighted and fused to determine whether the pet behavior is abnormal.

[0051] In the process of identifying behavioral triggers, if a pet's behavior is abnormal, it is marked as the target behavior. This target behavior then serves as a baseline to determine the corresponding scene type. Since the target behavior may correspond to a short amount of video data, it may be difficult to accurately identify the scene type. Therefore, by using the target behavior as the baseline and several pet behaviors preceding and following it as auxiliary behaviors, the video data corresponding to both the target and auxiliary behaviors can be used to identify the scene type. Identifying the scene type based on multiple consecutive pet behaviors is more accurate.

[0052] In the process of identifying behavioral triggers, the behavioral triggers corresponding to abnormal pet behaviors are determined by combining the target behavior, auxiliary behaviors, and scene type. The target behavior and auxiliary behaviors can be sorted in chronological order to obtain a behavior sequence; after standardizing the behavior sequence and scene type, they are input into a pre-trained trigger identification model to obtain the behavioral trigger.

[0053] This application first determines whether the pet's behavior is abnormal. If it is abnormal, it uses this as a benchmark to determine the scene type. The abnormal behavior, which includes action features, sound features, and physiological features, is combined with the corresponding scene type, and the cause of the pet's behavior is determined through the cause recognition model. This application limits the pet's behavior to scene type, which can achieve accurate positioning of the cause of the pet's behavior and avoid the inability to determine the core cause by identifying a single pet behavior.

[0054] In a preferred embodiment, generating pet behavior based on monitoring data includes: extracting monitoring data; extracting motion features from video data, extracting sound features from sound data, and extracting physiological features from physiological data; and aligning and integrating the motion features, sound features, and physiological features into pet behavior.

[0055] Generating pet behavior based on monitoring data essentially involves using video analysis technology, voice recognition technology, and other techniques to extract the current behavioral characteristics of a pet, i.e., pet behavior.

[0056] Pet monitoring data, including video, audio, and physiological data, is extracted from the database. Video analytics is used to analyze the video data to determine the pet's movement characteristics. Next, the time range corresponding to these dynamic characteristics is determined. Based on this time range, audio analysis is used to extract vocal features from the audio data and physiological features from the physiological data. These movement, audio, and physiological characteristics are then combined to form a single pet behavior pattern.

[0057] It is worth noting that pet behavior should include motor characteristics, vocal characteristics, and physiological characteristics. However, if the three cannot be aligned, pet behavior should include at least one characteristic, such as pet behavior containing only motor characteristics, or pet behavior containing both motor and vocal characteristics.

[0058] Motion characteristics are primarily used to identify a pet's body posture and assist in recognizing its emotional state. Therefore, they include basic movements and emotional movements. Basic movements refer to the pet's overall posture, while emotional movements refer to the postures of key parts of the pet's body. For example, basic movements of a pet dog include eating, resting, and playing. Emotional movements include head movements such as raising, lowering, tilting, shaking, and nodding; ear movements such as erecting and folding back; and tail movements such as high-frequency swaying from side to side, low-speed stiff swaying, and drooping or tucked tail.

[0059] Vocal characteristics are used to identify a pet's emotions and physical condition, and different vocal characteristics can be set for different pets. For example, the vocal characteristics of a pet dog include high-frequency short barks, low-frequency long barks, continuous rapid barks, intermittent barks, coughing, and screaming.

[0060] Physiological characteristics are used to identify a pet's emotions and physical condition, and different physiological parameters can be set for different pets. For example, the physiological characteristics of a pet dog include heart rate, body temperature, food intake, defecation frequency, sleep duration, and sleep quality.

[0061] In one example, using video data of a pet dog eating, video analysis technology is used to identify the eating action from the video data and treat it as a motion feature. Simultaneously, the time range corresponding to this motion sequence is determined, and audio analysis technology is used to extract the dog's vocal features from the sound data within this time range, as well as physiological features from the physiological data within the same time range. The motion feature, vocal feature, and physiological features are then correlated to form a single pet behavior.

[0062] In another example, if after determining the time range of the action sequence, the audio analysis technique extracts no vocal features of the pet dog within that time range, but extracts physiological features within that time range, then the pet's behavior only includes action features and physiological features, but no vocal features.

[0063] In a preferred embodiment, determining whether a pet's behavior is abnormal includes: extracting preset baseline data of the pet; inputting the pet's behavior and the preset baseline data into a behavior recognition model, and the behavior recognition model outputting a judgment result.

[0064] Since pet behavior contains a variety of features, the process of comparing it with the corresponding preset benchmark data is quite complex. Therefore, it is possible to determine whether a pet's behavior is abnormal based on an artificial intelligence model. That is, a pre-trained behavior recognition model is used to compare the action features, sound features, and physiological features of the pet's behavior with the preset benchmark data, and then the comparison results are weighted and fused to output the judgment result of the pet's behavior.

[0065] Figure 3 This is a schematic diagram illustrating the structural principle of a behavior recognition model. The behavior recognition model includes a data input layer, a data preprocessing layer, an independent analysis layer (including motion analysis units, sound analysis units, and physiological analysis units), a weighted fusion layer, and a result output layer. The data input layer receives monitoring data from the pet and preset baseline data, including motion baseline data, sound baseline data, and physiological baseline data. The data preprocessing layer performs data cleaning, data synchronization, and feature preprocessing. In the independent analysis layer, the motion analysis unit analyzes whether motion features are abnormal based on the motion baseline data, the sound analysis unit analyzes whether sound features are abnormal based on the sound baseline data, and the physiological analysis unit analyzes whether physiological data is abnormal based on the physiological baseline data. The weighted fusion layer weights and fuses the analysis results from the first three units and outputs a judgment result. If the judgment result shows an anomaly, the target behavior is considered abnormal; if the judgment result shows a normal result, the target behavior is considered normal. The result output layer outputs the final judgment result.

[0066] In the behavior recognition model, each unit of the independent analysis layer can use a publicly available artificial intelligence algorithm. After training the artificial intelligence model according to the analysis purpose of each unit, it is concatenated with other units or processing layers to obtain the behavior recognition model. The specific training process will not be described in detail here.

[0067] In one example, taking a pet dog as an example, the pet's behavior, including movement characteristics, vocal characteristics, and physiological characteristics, is generated based on the monitoring data. Then, the preset benchmark data corresponding to the pet is extracted. The preset benchmark data includes movement benchmark data, vocal benchmark data, and physiological benchmark data.

[0068] After standardization, the motion features and motion baseline data are input into the motion analysis unit to obtain the motion feature analysis results. Similarly, after standardization, the sound data and sound baseline data are input into the sound analysis unit to obtain the sound feature analysis results. Likewise, after standardization, the physiological data and physiological baseline data are input into the physiological analysis unit to obtain the physiological feature analysis results. The analysis results from these three units are then weighted and fused according to preset weights to obtain a judgment result. If the judgment result indicates abnormal pet behavior, then it is considered abnormal behavior, and the judgment result is output.

[0069] In a preferred embodiment, determining the scene type based on video data of the target behavior and the corresponding auxiliary behavior includes: determining the auxiliary behavior corresponding to the target behavior; extracting video data of the target behavior and the auxiliary behavior; and analyzing the video data through a scene recognition model to obtain the scene type.

[0070] To address the problem that existing technologies only analyze whether pet behavior is abnormal, but cannot accurately determine the behavioral triggers of abnormal pet behavior, this application, when determining that a pet's behavior is abnormal, identifies the corresponding scenario type. By combining the scenario type with the abnormal pet behavior, the behavioral triggers are determined. This not only improves the reliability of the behavioral triggers but also allows for targeted treatment based on the triggers, thereby ensuring the pet's health at all times.

[0071] Once a pet behavior is marked as the target behavior, several pet behaviors preceding or following the target behavior are extracted as auxiliary behaviors. The video data corresponding to the target behavior and auxiliary behaviors are used as the data basis for scene type recognition. The video data is analyzed through a scene recognition model to obtain the corresponding scene type.

[0072] Figure 4 This is a schematic diagram of the scene recognition model. The scene recognition model includes a data input layer, a data preprocessing layer, a feature extraction layer, and a classification output layer. The feature extraction layer is built based on the C3D model. First, a basic scene model is built based on the C3D model. Then, the basic scene model is trained using video data with already labeled scene types. The trained basic scene model serves as the scene recognition model.

[0073] Scene types include solitary scenes, interactive scenes, exercise scenes, and feeding scenes, and other scene types can be set according to the pet's activity environment. Of course, each of the above scene types can also be subdivided, such as interactive scenes can be subdivided into scenes of interaction with people and scenes of interaction with other pets, and exercise scenes can be subdivided into running, jumping, pouncing, climbing, etc.

[0074] In one example, after determining the target behavior and its corresponding auxiliary behavior, the video data corresponding to the target behavior and auxiliary behavior are extracted, and the video data is sorted according to the order in which the pet's behavior occurs to obtain a video sequence. The video sequence is then analyzed using video analytics techniques to determine its scene type.

[0075] It should be noted that in real-time pet behavior analysis scenarios, since subsequent video data has not yet been collected, the auxiliary behaviors are all pet behaviors that precede the target behavior; in timed pet behavior analysis scenarios, the auxiliary behaviors may also include pet behaviors that follow the target behavior.

[0076] In a preferred embodiment, determining the auxiliary behavior corresponding to the target behavior includes: extracting several pet behaviors before or after the target behavior; using the target behavior as a reference point, analyzing whether the several pet behaviors have a temporal correlation with the target behavior in a forward or backward manner; if so, marking the pet behavior as an auxiliary behavior.

[0077] After identifying the target behavior, use the target behavior as a baseline to extract pet behaviors before or after the target behavior. If the pet behaviors before or after the target behavior are temporally related to the target behavior, they are considered as auxiliary behaviors.

[0078] When determining whether a pet's behavior is related to the target behavior, a time threshold can be set. Specifically, time threshold one and time threshold two are set. If the target behavior corresponds to several auxiliary behaviors, the time interval between adjacent pet behaviors shall not exceed time threshold one, and the time interval between the last pet behavior and the target behavior shall not exceed time threshold two.

[0079] In one example, suppose the target behavior is A. The pet behaviors before A, from closest to furthest, are B1, B2, B3, B4, and B5; and the pet behaviors after A, from closest to furthest, are C1, C2, C3, C4, and C5.

[0080] If the time interval between A and B1, B1 and B2, B2 and B3, and B3 and B4 is not greater than time threshold one, the time interval between B4 and B5 is greater than time threshold one, and the time interval between A and B5 does not exceed time threshold two, then B1, B2, B3, and B4 can be used as auxiliary behaviors.

[0081] If the time interval between A and C1, C1 and C2, C2 and C3, C3 and C4, and C4 and C5 is not greater than time threshold one, but the time interval between A and C3 exceeds time threshold two, then C1, C2, and C3 can be used as auxiliary behaviors.

[0082] In this application, time threshold one and time threshold two are mainly used to determine whether a behavior can be used as an auxiliary behavior based on the correlation between the pet's behavior and the target behavior. Time threshold one and time threshold two need to be set according to the different habits of different pets. For example, for a pet's head, time threshold one can be set to 10 seconds and time threshold two can be set to 60 seconds.

[0083] In a preferred embodiment, determining the pet's behavioral triggers based on the target behavior, auxiliary behavior, and scenario type includes: calling a trigger recognition model; and inputting the target behavior, auxiliary behavior, and scenario type into the trigger recognition model after standardization to obtain the pet's behavioral triggers.

[0084] This application determines behavioral triggers based on pet behavior and its corresponding scenario type. After determining the scenario type, abnormal pet behavior, auxiliary behavior, and scenario type are integrated as inputs to the trigger identification model. The trigger identification model can then identify the corresponding behavioral triggers, which include physiological and psychological triggers.

[0085] It should be noted that when pet behaviors (including target behaviors and auxiliary behaviors) are input into the cause recognition model as data items, not only the behavior labels of the pet behaviors need to be input, but also the abnormal labels of the pet behaviors. That is, it is also necessary to first determine whether the auxiliary behaviors are abnormal behaviors.

[0086] Figure 5 This is a schematic diagram of the incentive recognition model. The incentive recognition model includes a data input layer, a data preprocessing layer, a core classification layer, and a result output layer. The core classification layer can be constructed based on a gradient boosting tree model. During training, several training datasets are first acquired. Each training data point in the training dataset includes the target behavior, auxiliary behavior, scene type, and labeled behavioral incentives. The training datasets are then used to complete the training and will serve as the incentive recognition model.

[0087] In one instance, after calling a pre-trained trigger recognition model, the target behavior and auxiliary behavior are sorted according to the order in which they occur to obtain a behavior sequence. At the same time, for anomalies in the pet's behavior in the behavior sequence, a behavior judgment sequence is generated based on the analysis results. After standardizing the behavior sequence, behavior judgment sequence, and scene model, the results are input into the trigger recognition model to obtain the output behavior trigger.

[0088] It is worth noting that this application determines the pet's behavioral triggers based on the target behavior, auxiliary behavior, and scenario type, rather than the behavioral triggers corresponding to the target behavior. This is because behavioral triggers are the identification results of a series of pet behaviors in a specific scenario type, not the identification results of a single pet behavior; the target behavior is merely used as the entry point. Since it does not identify the behavioral triggers corresponding to the target behavior, the input items of the trigger identification model do not need to be specially labeled for the target behavior.

[0089] Example 2: Compared to Example 1, this example determines whether a pet's behavior is abnormal based on a personalized baseline, rather than based on preset baseline data. By identifying abnormal behavior through specifically set personalized baseline data, the accuracy of the judgment results can be improved.

[0090] Existing solutions typically use general, pre-set baseline data to classify the behavior of different pets. However, these pre-set baseline data are difficult to cover individual differences such as pet breed, growth history, and health status. Even with training using massive amounts of data, there will still be insufficient coverage of niche breeds or rare behavioral scenarios. Therefore, the accuracy of using general models to analyze the behavior of different pets is low, and misjudgments are prone to occur.

[0091] Figure 6 This is a diagram illustrating a method for judging pet behavior based on personality baseline data.

[0092] In a preferred embodiment, determining whether a pet's behavior is abnormal includes: extracting the pet's personality baseline data; wherein the personality baseline data is constructed based on monitoring data of the pet's health status; inputting the pet's behavior and personality baseline data into a behavior recognition model, and the behavior recognition model outputting a judgment result.

[0093] In Example 1, when analyzing whether pet behavior (including auxiliary behaviors) is abnormal, the process essentially involves independently analyzing motion features, vocal features, and physiological features using preset benchmark data, followed by weighted fusion. The fusion result is then used to determine whether the pet behavior is abnormal. However, this preset benchmark data is a general benchmark obtained from a large number of pet analyses. Using a general benchmark as a basis cannot guarantee the coverage of pet behavior recognition scenarios.

[0094] This application extracts the pet's individual baseline data and uses this individual baseline data to replace the general preset baseline data as the basis for judging whether the pet's behavior is abnormal. This allows for targeted judgment of the pet, thereby avoiding misjudgment or omission due to insufficient coverage of pet breed, pet behavior and other scenarios.

[0095] In a preferred embodiment, constructing the pet's personality baseline data includes: extracting the pet's monitoring data as reference data when the pet is found to be healthy after examination; and generating the pet's personality baseline data based on the reference data and the preset baseline data.

[0096] When a pet's health check indicates it is in good health, the monitoring data from that state can be used as reference data to construct unique personality baseline data for the pet. This personality baseline data contains the same data items as the preset baseline data. It's important to note that for a pet to be considered healthy, both psychologically and physiologically, this "healthy" state is crucial. Using the preset baseline data as a foundation ensures consistency between the personality baseline data and the data items within it, allowing for adaptation to various artificial intelligence models.

[0097] It is important to note that a pet's health status refers to both physical and mental well-being; otherwise, the baseline characteristics of the personality baseline data cannot be guaranteed. This is because a pet's physical and mental health can influence each other. For example, if a pet is physically healthy but mentally unhealthy, its physiological data may appear abnormal, leading to misjudgments of the pet's behavior in the personality baseline data generated based on this.

[0098] It is also important to note that when constructing individual baseline data, it should be based on monitoring data from the same growth stage of the pet. For example, reference data should be generated using monitoring data collected from the pet when it is an adult and in a healthy state. Avoid using individual baseline data from different growth stages to judge abnormal pet behavior.

[0099] In one example, assuming the pet dog's health check data at times T1 and T2 indicate it is in good health, monitoring data within one week after times T1 and T2 (assuming no damage to its health) can be used as reference data. Personalized baseline data can be generated based on this reference data from these two time periods. For instance, one of the pet dog's eating actions from the two time periods can be selected as the baseline data for that action, and the range of heart rate fluctuations within those two time periods can be used as the baseline data for heart rate indicators.

[0100] Example 3: Compared with Example 1 or Example 2, after analyzing the behavioral causes of abnormal pet behavior, this application analyzes the correlation items in conjunction with the pet's physical examination data, and generates intervention suggestions based on the correlation items, which can intervene in a timely manner to eliminate the pet's health condition.

[0101] Figure 7 This diagram illustrates the steps involved in generating intervention recommendations based on behavioral triggers.

[0102] In a preferred embodiment, after identifying the behavioral trigger, intervention suggestions are generated by combining the pet's physical examination data, including: extracting the pet's physical examination data and extracting abnormal items from the physical examination data as candidate items; matching the candidate items with the related items of the behavioral trigger to obtain the target item; and generating intervention suggestions based on the behavioral trigger and the target item.

[0103] Existing solutions focus on physiological data when analyzing pet behavior. After identifying the behavioral triggers, they cannot pinpoint the specific influencing factors, nor can they determine whether the behavioral trigger is a common phenomenon in the corresponding type of pet or a phenomenon unique to this pet. Consequently, it is difficult to provide effective intervention directions for abnormal pet behavior.

[0104] After identifying the behavioral trigger for the abnormal behavior, this application retrieves the pet's physical examination data and matches the behavioral trigger with the health data to determine the specific cause of the abnormal behavior. If the behavioral trigger is a physiological reason, the application combines the health data to identify the physiological problem leading to the abnormal behavior; if the behavioral trigger is a psychological reason, the application combines the health data to identify the psychological problem leading to the abnormal behavior.

[0105] In one example, assuming the target behavior's movement characteristics indicate an abnormal gait in a pet dog, a cause identification model identifies it as a physiological trigger. The dog's recent physical examination data is retrieved, and abnormalities are extracted, including mild osteophyte formation in the right hind leg knee joint and a history of right hind leg sprain. Both of these are associated with the abnormal gait and are therefore selected as target items. Then, corresponding medications and care methods are matched based on the target items, and the medications and care methods are converted into intervention recommendations.

[0106] It should be noted that, in order to improve matching efficiency, the correlation between behavioral triggers and each physical examination item in the physical examination data can be established in advance; intervention suggestions can also be pre-matched for the physical examination items in the physical examination data, that is, the measures that should be taken if the physical examination items are abnormal.

[0107] It should be understood that the descriptions of technical features, technical solutions, beneficial effects, or similar language in this application do not imply that all features and advantages can be achieved in any single embodiment. Rather, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution, or beneficial effect is included in at least one embodiment. Therefore, the descriptions of technical features, technical solutions, or beneficial effects in this specification do not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions, and beneficial effects described in this embodiment can be combined in any suitable manner. Those skilled in the art will understand that embodiments can be implemented without one or more specific technical features, technical solutions, or beneficial effects of a particular embodiment. In other embodiments, additional technical features and beneficial effects may be identified in specific embodiments that do not embody all embodiments.

[0108] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any other combination thereof. When implemented using a software program, it can be implemented entirely or partially in the form of a computer program product. This computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device containing one or more servers, data centers, etc., that can be integrated with the medium. The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state disks (SSDs)).

[0109] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A pet behavior analysis system based on visual intelligence, characterized in that, include: Central processing module: used to mark abnormal pet behaviors as target behaviors; The scene type is determined based on video data of the target behavior and corresponding auxiliary behaviors; wherein, pet behavior includes movement characteristics, vocal characteristics, or physiological characteristics, and the auxiliary behaviors are temporally correlated with the target behavior; and, Used to determine the behavioral triggers of a pet based on the target behavior, auxiliary behavior, and scenario type; wherein, the behavioral triggers include physiological triggers or psychological triggers.

2. The pet behavior analysis system based on visual intelligence according to claim 1, characterized in that, The pet behaviors are generated based on monitoring data, including: Extract monitoring data; the monitoring data includes video data, audio data, and physiological data; Motion features are extracted from the video data, sound features are extracted from the sound data, and physiological features are extracted from the physiological data; the motion features, sound features, and physiological features are then aligned and integrated into pet behavior.

3. The pet behavior analysis system based on visual intelligence according to claim 1, characterized in that, Determining whether the pet's behavior is abnormal includes: Extract the pet's preset baseline data; the preset baseline data includes action baseline data, vocal baseline data, and physiological baseline data; The pet behavior and the preset benchmark data are input into the behavior recognition model, and the behavior recognition model outputs the judgment result; wherein, the behavior recognition model includes a data preprocessing layer, an independent analysis layer and a weighted fusion layer, and the independent analysis layer includes an action analysis unit, a sound analysis unit and a physiological analysis unit.

4. The pet behavior analysis system based on visual intelligence according to claim 1, characterized in that, The scene type is determined based on the video data of the target behavior and the corresponding auxiliary behavior, including: Identify the auxiliary behaviors corresponding to the target behavior; wherein, the auxiliary behavior refers to pet behaviors that are temporally related to the target behavior; Video data of the target behavior and the auxiliary behavior are extracted, and the video data is analyzed through a scene recognition model to obtain the scene type. The scene recognition model includes a data input layer, a data preprocessing layer, a feature extraction layer, and a classification output layer. The feature extraction layer is used to extract features from the video data.

5. A pet behavior analysis system based on visual intelligence according to claim 4, characterized in that, Determining the auxiliary behavior corresponding to the target behavior includes: Extract several pet behaviors before or after the target behavior; Using the target behavior as a reference point, analyze whether several pet behaviors have a temporal correlation with the target behavior in a forward or backward manner; if so, mark the pet behavior as an auxiliary behavior; wherein, the temporal correlation is determined from two perspectives: time span and time interval.

6. The pet behavior analysis system based on visual intelligence according to claim 1, characterized in that, Based on the target behavior, auxiliary behavior, and scenario type, the pet's behavioral triggers are determined, including: The trigger identification model is invoked; the trigger identification model includes a data input layer, a data preprocessing layer, a core classification layer, and a result output layer. The core classification layer is used to determine the behavioral trigger. The target behavior, auxiliary behavior, and scene type are standardized and then input into the cause recognition model to obtain the pet's behavioral causes.

7. The pet behavior analysis system based on visual intelligence according to claim 1, characterized in that, Determining whether the pet's behavior is abnormal includes: Extract the pet's individual baseline data; the individual baseline data is constructed based on monitoring data of the pet's health status; The pet's behavior and the personality baseline data are input into the behavior recognition model, and the behavior recognition model outputs a judgment result; wherein, the behavior recognition model includes a data preprocessing layer, an independent analysis layer and a weighted fusion layer, and the independent analysis layer includes an action analysis unit, a sound analysis unit and a physiological analysis unit.

8. A pet behavior analysis system based on visual intelligence according to claim 7, characterized in that, The personality baseline data for the pet is constructed, including: When a pet is found to be healthy after a check-up, the pet's monitoring data is extracted as reference data. Based on preset baseline data, the pet's personality baseline data is generated according to the reference data; wherein, the personality baseline data contains the same data items as the preset baseline data.

9. A pet behavior analysis system based on visual intelligence according to claim 1, characterized in that, After identifying the behavioral triggers, intervention recommendations are generated based on the pet's physical examination data, including: Extract the pet's physical examination data, and extract abnormal items from the physical examination data as candidate items; Match the associated items of the behavioral triggers in the candidate options to obtain the target items; generate intervention suggestions based on the behavioral triggers and the target items.

Images