Intelligent pet cage based on Internet of Things

Abstract

The intelligent pet cage based on the Internet of Things comprises a base and a necklace dismounting and mounting mechanism, the periphery of the base is fixedly connected with a fence, the fence is provided with an access door, and the top of the fence is fixedly connected with a top plate; the necklace dismounting and mounting mechanism is used for dismounting and mounting a necklace on the neck of a pet and comprises a shell, a hole is formed in the front side of the shell, an operation assembly is arranged in the hole and used for dismounting or mounting the necklace on the neck, a first electric push rod is arranged in the shell, and the output end of the first electric push rod is fixedly connected with a pet basin. A pet granary is arranged at the top of the pet basin and used for adding pet food into the pet basin, and the first electric push rod is used for pushing the pet basin to the hole. Through fusion of necklace monitoring, computer vision and machine learning, precise health monitoring, intelligent behavior analysis and automatic management are realized, pet raising experience is effectively improved, pet health risks are reduced, and the system has significant market application value.

Description

Technical Field

[0001] The present invention relates to the technical field of pet cages, and in particular to an intelligent pet cage based on the Internet of Things. Background Art

[0002] As people in modern society pay more and more attention to pet health management, smart pet devices have gradually become an important auxiliary tool in the process of pet breeding. At present, smart pet cages on the market usually have basic functions such as remote video monitoring and automatic feeding, which can facilitate pet management to a certain extent. However, these devices still have many shortcomings in terms of intelligence, especially in terms of pet behavior analysis and health monitoring, lacking accurate personalized behavior recognition capabilities. Existing smart pet cages and monitoring equipment usually use traditional cameras for video monitoring, and users can view the status of pets remotely, but these devices often only provide basic video playback functions and cannot conduct in-depth analysis of pet behavior patterns. For example, existing equipment cannot effectively identify the behavioral characteristics of pets such as eating, activity, and sleep, and it is difficult to determine whether their behavior patterns are abnormal. In addition, these devices lack the support of intelligent algorithms and cannot be combined with pets' physiological data (such as heart rate, body temperature, etc.) for comprehensive analysis, resulting in limited health monitoring capabilities.

[0003] Due to the lack of accurate and personalized behavior recognition, current smart pet devices have obvious defects in health monitoring. For example, a pet's reduced food intake, reduced exercise, or abnormal behavior may be an early sign of health problems, but traditional devices cannot accurately capture these subtle changes or provide real-time warnings. For pet owners, it is often difficult to detect abnormal conditions of pets in a timely manner by relying solely on video surveillance, and the best time for early intervention may be missed.

[0004] In addition, existing devices mainly rely on traditional video monitoring methods to identify pet behavior, lacking in-depth analysis based on artificial intelligence (AI) and Internet of Things (IoT) technologies. For example, existing smart pet cages rarely use computer vision technology to analyze pets' postures and movement patterns, nor do they combine multi-sensor data fusion methods, resulting in insufficient accuracy in behavior monitoring. At the same time, traditional equipment's data processing methods mostly rely on cloud computing, which has certain delays and affects real-time performance, and long-term data transmission may bring privacy and security risks. Therefore, how to build a set of smart pet cages based on IoT and AI technologies that can accurately identify pets' personalized behaviors, analyze health status in real time, and provide intelligent warnings has become a technical problem that needs to be solved urgently in the current technical field. Summary of the invention

[0005] The purpose of the present invention is to provide a smart pet cage based on the Internet of Things, which can accurately identify the personalized behavior of pets, analyze the health status in real time and provide intelligent early warning.

[0006] To achieve the above object, the present invention proposes the following technical solution: a smart pet cage based on the Internet of Things, comprising:

[0007] A base, wherein the base is fixedly connected with a fence on all sides, the fence is provided with an entry and exit door, and the top of the fence is fixedly connected with a top plate;

[0008] A collar disassembly and assembly mechanism, the collar disassembly and assembly mechanism is used to disassemble and assemble the collar on the neck of a pet, the collar disassembly and assembly mechanism includes a shell, a hole is opened on the front side of the shell, the hole is provided with an operating component, the operating component is used to remove or install the collar on the neck, a first electric push rod is provided in the shell, the output end of the first electric push rod is fixedly connected to a pet bowl, a pet food bin is provided on the top of the pet bowl, the pet food bin is fixedly connected to the shell, a solenoid valve is provided at the bottom of the pet food bin, the pet food bin is used to add pet food into the pet bowl, and the first electric push rod is used to push the pet bowl to the hole.

[0009] Furthermore, in the present invention, the operating assembly includes an operating box, a second electric push rod is arranged in the operating box, the second electric push rod is provided with a clamping arm, and the clamping arm is used to clamp the collar.

[0010] Furthermore, in the present invention, the collar is used to monitor the heart rate, body temperature, and acceleration of the pet, and the collar includes two left and right semicircular collars, and the collar is provided with a heart rate sensor, a body temperature sensor, and a pressure acceleration sensor.

[0011] Furthermore, in the present invention, positioning components are provided at both ends of the semicircular collar on the left, and both ends of the semicircular collar on the right are fixedly connected with special-shaped locking rods, the positioning component includes a lock box and a rotating lock arm, the rotating lock arm is movably connected to the lock box through a rotating shaft, the upper part of the rotating lock arm is movably connected to a movable rod, one end of the movable rod can be inserted into the positioning arm and is fixedly connected with a reset spring, the reset spring is arranged in the inner cavity of the positioning arm, the positioning arm is movably connected to the lock box, a limiting rod is fixedly connected to the movable rod, a limiting hole matched with the limiting rod is opened on the positioning arm, and the limiting rod is inserted into the limiting hole.

[0012] Furthermore, in the present invention, the special-shaped locking rod includes a rod body, the top of the rod body is fixedly connected to a first vertebra, the upper part of the first vertebra is conical, the middle part of the rod body is fixedly connected to a second vertebra, the upper and lower parts of the second vertebra are both conical, and the volume of the second vertebra is greater than that of the first vertebra.

[0013] Further, in the present invention, when the special-shaped lock rod is inserted into the positioning assembly, a first state is defined. The first vertebral body of the special-shaped lock rod contacts the rotating lock arm. At this time, the first vertebral body pushes the rotating lock arm open, and the rotating lock arm rotates by a certain angle, defined as X°. When the first vertebral body continues to move upward until the top of the rear rotating lock arm, at this time, the reset spring pushes the rotating lock arm to reset through the movable rod, and the rotating lock arm plays a supporting role for the first vertebral body, thus completing the locking;

[0014] When the special-shaped lock rod is inserted into the positioning assembly and reaches the first state, and continues to move upward, the second state can be achieved. Define the second state. The second vertebral body of the special-shaped lock rod contacts the rotating lock arm. At this time, the second vertebral body pushes the rotating lock arm open, and the rotating lock arm rotates by a certain angle, defined as Y°. Y is greater than X. When the second vertebral body continues to move upward until the top of the rear rotating lock arm, at this time, the rotating lock arm will not reset. The special-shaped lock rod moves downward until the second vertebral body contacts the bottom of the rotating lock arm. At this time, the second vertebral body drives the rotating lock arm to rotate in the reverse direction to form a reset, thus completing the unlocking movement.

[0015] Further, in the present invention, it includes a high-definition camera and a control system. The high-definition camera is arranged at the bottom of the top plate. The control system is an edge computing device. The edge computing device is used for local computer vision processing to reduce cloud latency. The collar transmits the monitored pet heart rate, body temperature, and acceleration data to the edge computing device, and the edge computing device is used to run the intelligent pet health monitoring and warning method.

[0016] Further, in the present invention, the intelligent pet health monitoring and warning method includes the following process:

[0017] Step 1, collect the daily behavior data of the pet through the high-definition camera and the collar, including visual behavior and physiological data;

[0018] YOLO object detection, where the YOLO loss function is as follows:

[0019] L = λ 1 L coord + λ 2 L obj + λ 3 L noobj + λ 4 L class + λ 5 L temp ;

[0020] Where: L coord : The target box coordinate loss, which improves the target positioning accuracy;

[0021] L obj: The target has losses, reducing the missed detection rate;

[0022] L noobj : Background error loss, reducing false detection cases;

[0023] L class : Class classification loss, ensuring that different behaviors of pets can be accurately classified;

[0024] L temp , Temporal sequence consistency loss: A newly introduced optimization term used to consider the continuity of pet behaviors and prevent short-term behaviors from being misjudged;

[0025] Among them: B t : Detection box parameters for the t-th frame. This loss term ensures stable prediction of the same target in consecutive frames and reduces jitter errors;

[0026] Perform OpenPose pose analysis, analyze the key points of the pet's bones, calculate the motion trajectory, acceleration, and pose changes, extract its activity patterns. The key point detection formula is as follows:

[0027] v i，j (t) = p j (t) - p i (t);

[0028] p i (t): At time t, the coordinates of the i-th key point, v i，j (t): The relative vector between key points, used to calculate limb postures;

[0029] Calculation of motion speed and acceleration:

[0030]

[0031] If is high for a long time, it indicates that the pet has abnormal jitter. If is close to zero, it indicates that the pet's movement has decreased, indicating that it is likely to be in a sick state;

[0032] Run OpenPose to detect the key points of the pet's head, limbs, and tail, calculate the pet's motion trajectory, and combine the results of YOLO to comprehensively judge the behavior pattern;

[0033] Step 2, Combine the biological data of heart rate and body temperature to further judge whether the pet is abnormal. Collect the PPG or ECG of the pet through the collar 6 to obtain the heart rate signal and analyze the anxiety situation. The heart rate variability (HRV) calculation formula:

[0034]

[0035] Among them: RRi is the heart - beat interval, is the average heart - beat interval. If the HRV decreases, it indicates that the pet is prone to an anxious state;

[0036] If the body temperature is higher than the threshold, push an alarm message to the owner's mobile phone. Combining with the behavior data of YOLO + OpenPose, further analyze whether medical treatment is needed;

[0037] Step 3: Integrate computer vision and sensor data, and use machine learning to predict health risks,

[0038] First, perform data fusion. Input X = {x 1 , x 2 ,..., x n}, where x 1 = motion trajectory data, x 2 = behavior pattern data, x 3 = heart rate data, x 4 = body temperature data;

[0039] Use the Kalman filter to fuse multiple sensor data: Let the measurement value be z k , the true state be x k , and the estimation equation: x k = Ax k-1 + Bu k + w k ;

[0040] x k : System state vector;

[0041] A: State transition matrix, describing the law of system change over time;

[0042] x k-1 : The system state at the previous moment;

[0043] B: Control input matrix, indicating the external influence on the system;

[0044] u k : External input;

[0045] w k : Process noise, representing the uncertainty in sensor measurements;

[0046] z k = Hx k + v k ;

[0047] z k : Sensor observation value;

[0048] H: Observation matrix, defining how the sensor measures the state variables;

[0049] v k : Measurement noise, representing the error existing in the sensor;

[0050] Eliminate the noise through Kalman filtering to improve data reliability;

[0051] Adopt a deep time series prediction model:

[0052] h t = σ(W h h t-1 + W x x t + b h );

[0053]

[0054] y t = W y h t + b y ;

[0055] x t : Input data at the current moment;

[0056] h t : Hidden state of the LSTM, storing long-term memory;

[0057] W h ,W x ,W y : Neural network weight matrix, learning the long-term change patterns of the input data;

[0058] b h ,b y : Bias term, helping to adjust the activation threshold of the neuron;

[0059] σ: Activation function;

[0060] c t : Cell state, storing the long-term health information of the pet;

[0061] f t ,i t , Internal gating mechanism in LSTM: f t Forget gate: Decide whether to forget the health status at the previous moment, i t Input gate: Decide whether to accept the new health data at the current moment, New information at the current moment, updating the cell state together with the input gate;

[0062] Input x t : Including YOLO behavior data, OpenPose motion data, heart rate, and body temperature in the past seven days, output yt : The pet health status score for the next 24 hours is 0 - 100. The scoring thresholds are as follows: 80 - 100 indicates healthy, 50 - 79 indicates mild abnormality, and 0 - 49 indicates severe abnormality, and an alarm is triggered.

[0063] Reinforcement learning is adopted to adjust the health scoring model based on the pet's historical data. The reward function: R(s, a) = -(y t - y baseline ) 2 ; Let the AI adjust the alarm threshold according to the historical data to reduce false alarms.

[0064] Furthermore, in the present invention, in step 1, continuous frame data of the RGB camera is collected, the YOLOv8 / YOLO-NAS model is run to detect the pet's body parts, the behavior data within different time periods is statistically analyzed, the staying time in front of the food bowl is detected. If it is lower than the normal threshold, there may be a problem of decreased appetite. If the YOLO detects that the tongue frequently touches the paw, it may indicate anxiety or skin problems. If the pet keeps moving at high speed or stays still for too long, it may indicate health problems.

[0065] Furthermore, in the present invention, in step 1, various behaviors of the pet such as eating, licking the paw, walking, and lying prone are detected. YOLOv8 or YOLO-NAS is adopted to improve the detection accuracy of small targets. Multi-scale feature fusion (FPN + PAN) is used to ensure that pets of different sizes can be detected. GhostNet lightweight CNN is adopted to reduce the computational amount so that it can run on low-power devices. RGB + IR data fusion is adopted, combined with an infrared camera, and the night monitoring effect is more stable.

[0066] Beneficial effects: The technical solution of this application has the following technical effects:

[0067] Through the collar disassembly and assembly mechanism of the present invention, the pet collar can be automatically disassembled and assembled, reducing manual intervention and improving the convenience of use. The granary system controlled by an electric push rod + solenoid valve realizes precise feeding, ensuring that the pet's head can reach the hole (506), thereby facilitating the disassembly and assembly of the pet collar. The collar (6) monitors the pet's heart rate, body temperature, and acceleration, and can obtain physiological data in real time to achieve personalized health management. The high-definition camera combined with YOLO + OpenPose computer vision technology can detect the pet's behavior patterns, including eating, licking the paw, walking, lying prone, etc., and identify abnormal behaviors such as decreased appetite, anxiety, and disease warning. Heart rate variability (HRV) calculation helps analyze the pet's anxiety state and provides a reference for the owner's mental health. RGB + IR data fusion enables more accurate night monitoring and ensures all-weather health management.

[0068] Adopt a dual-state locking structure of special-shaped lock rod + rotating lock arm + return spring, which can automatically lock and unlock the collar to ensure safety. The edge computing device directly processes data, reduces cloud computing latency, and improves the system response speed. Combine Kalman filtering to eliminate sensor data noise and improve measurement accuracy. Adopt the LSTM deep learning model to predict the health status in the next 24 hours and provide scientific pet-raising decisions for the owner. Reinforcement learning adaptively adjusts the health scoring algorithm to reduce the false alarm rate and improve the system reliability.

[0069] Adopt YOLO-NAS / YOLOv8 + FPN + PAN to improve the detection accuracy of small targets and ensure that pets of different sizes can be recognized. GhostNet lightweight CNN reduces the computing cost and makes the device suitable for low-power scenarios.

[0070] Through the integration of collar monitoring + computer vision + machine learning, the present invention realizes precise health monitoring, intelligent behavior analysis, and automated management, effectively improves the pet-raising experience, reduces the pet health risk, and has significant market application value.

[0071] It should be understood that all combinations of the foregoing concepts and additional concepts described in greater detail below can be regarded as part of the inventive subject matter of the present disclosure as long as such concepts do not conflict with each other.

[0072] The foregoing and other aspects, embodiments, and features of the teachings of the present invention can be more fully understood from the following description in conjunction with the accompanying drawings. Other additional aspects of the present invention, such as the features and / or beneficial effects of exemplary embodiments, will be apparent in the following description or will be learned through the practice of specific embodiments according to the teachings of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] The drawings are not intended to be drawn to scale. In the drawings, each identical or nearly identical component shown in each figure can be represented by the same reference numeral. For clarity, not every component is labeled in each figure. Now, embodiments of various aspects of the present invention will be described by way of example and with reference to the drawings, wherein:

[0074] Figure 1 It is a schematic structural diagram of the present invention.

[0075] Figure 2 It is a schematic structural diagram of the collar disassembly and assembly mechanism of the present invention.

[0076] Figure 3 It is a schematic structural diagram of the operation component of the present invention.

[0077] Figure 4 It is a schematic structural diagram of the collar of the present invention.

[0078] Figure 5This is a schematic structural diagram of the collar of the present invention.

[0079] Figure 6 This is a schematic structural diagram of the positioning component of the present invention.

[0080] In the figure, the meanings of the reference numerals are as follows: 1, base; 2, fence; 3, entrance / exit door; 4, top plate; 5, collar disassembly and assembly mechanism; 501, housing; 502, operation component; 5021, operation box; 5022, second electric push rod; 5023, clamping arm; 503, first electric push rod; 504, pet basin; 505, pet granary; 506, hole; 6, collar; 601, lock box; 602, special-shaped lock rod; 6021, rod body; 6022, first vertebral body; 6023, second vertebral body; 603, rotating lock arm; 604, movable rod; 605, positioning arm; 606, return spring; 607, limit hole; 608, limit rod. Detailed implementation manners

[0081] In order to better understand the technical content of the present invention, specific embodiments are hereby given and described in conjunction with the accompanying drawings as follows. In the present disclosure, aspects of the present invention are described with reference to the accompanying drawings, and many illustrative embodiments are shown in the drawings. The embodiments of the present disclosure do not necessarily define all aspects of the present invention. It should be understood that the various concepts and embodiments introduced above, as well as those concepts and embodiments described in more detail below, can be implemented in any of many ways, because the concepts and embodiments disclosed in the present invention are not limited to any implementation manner. Additionally, some aspects of the present invention can be used alone, or in any suitable combination with other aspects of the present invention that are disclosed.

[0082] Embodiment 1

[0083] As Figure 1-6 shown, this embodiment provides an Internet of Things-based intelligent pet cage, including a base 1 and a collar disassembly and assembly mechanism 5. The four sides of the base 1 are fixedly connected with a fence 2. An entrance / exit door 3 is arranged on the fence 2. The top of the fence 2 is fixedly connected with a top plate 4, providing a basic framework for the pet cage to ensure the activity range of the pet and prevent the pet from escaping at the same time. The base 1 provides support to improve the overall stability; the fence 2 surrounds to prevent the pet from escaping, and the entrance / exit door 3 can control the entry and exit of the pet; the top plate 4 can prevent external objects from falling into the pet cage to ensure the safety of the pet;

[0084] The collar disassembly and assembly mechanism 5 is used to disassemble and assemble the collar 6 on the pet's neck, realizing the automatic disassembly and assembly of the pet collar 6, reducing manual intervention, and improving the intelligent level. The collar disassembly and assembly mechanism 5 includes a housing 501. A hole 506 is provided on the front side of the housing 501. An operating component 502 is arranged in the hole 506. The operating component 502 disassembles or installs the collar 6 on the neck. A first electric push rod 503 is arranged in the housing 501. The output end of the first electric push rod 503 is fixedly connected with a pet basin 504. A pet food bin 505 is arranged on the top of the pet basin 504. The pet food bin 505 is fixedly connected with the housing 501. A solenoid valve is arranged at the bottom of the pet food bin 505. The pet food bin 505 is used to add pet food into the pet basin 504. The first electric push rod 503 is used to push the pet basin 504 to the hole 506.

[0085] In this embodiment, the operating component 502 includes an operating box 5021. A second electric push rod 5022 is arranged in the operating box 5021. The second electric push rod 5022 is provided with a clamping arm 5023. The clamping arm 5023 is used to clamp the collar 6.

[0086] When in use, when the collar needs to be automatically installed, the solenoid valve of the pet food bin 505 is opened, so that the pet food falls into the pet basin 504. Then the first electric push rod 503 is used to push the pet basin 504 to the hole 506. The pet will stretch its head into the hole 506 to eat the pet food in the pet basin 504. At this time, the second electric push rod 5022 pushes the clamping arm 5023 to drive the collar 6 to be stuck around the pet's neck. When removing the collar 6, similarly, the second electric push rod 5022 pushes the clamping arm 5023 to clamp the collar 6, and then separates the collar 6. It can be applicable to collars 6 of different sizes according to the different body sizes of pets. It can prevent the risks of pet struggling and escaping caused by manual disassembly. Through mechanical components such as electric push rods, automatic disassembly and assembly are realized, avoiding the inconvenience of manual operation. It can be combined with the pet health monitoring system to remove or install the collar when necessary.

[0087] Embodiment 2

[0088] On the basis of Embodiment 1, the collar 6 is used to monitor the pet's heart rate, body temperature, and acceleration. The collar 6 includes two left and right semi-circular collars. A heart rate sensor, a body temperature sensor, and a pressure acceleration sensor are arranged on the collar 6. The collar 6 can monitor data such as the pet's heart rate, body temperature, and acceleration in real time, helping to identify the pet's health status. It can perform multi-parameter monitoring: including multiple health indicators such as heart rate, body temperature, and acceleration; the heart rate sensor and the body temperature sensor can provide accurate data; it can be linked with the intelligent pet health monitoring system to give early warnings of health risks in a timely manner.

[0089] To ensure the stability of the collar, facilitate disassembly and assembly, and improve the convenience of use, this embodiment further optimizes the collar structure. Positioning components are provided at both ends of the left semi-circular collar, and special-shaped lock rods 602 are fixedly connected to both ends of the right semi-circular collar. The positioning component includes a lock box 601 and a rotating lock arm 603. The rotating lock arm 603 is movably connected to the lock box 601 through a rotating shaft. The upper part of the rotating lock arm 603 is movably connected to a movable rod 604. One end of the movable rod 604 can be inserted into the positioning arm 605 and is fixedly connected to a return spring 606. The return spring 606 is arranged in the inner cavity of the positioning arm 605. The positioning arm 605 is movably connected to the lock box 601. A limiting rod 608 is fixedly connected to the movable rod 604, and a limiting hole 607 adapted to the limiting rod 608 is formed in the positioning arm 605. The limiting rod 608 is inserted into the limiting hole 607.

[0090] The special-shaped lock rod 602 includes a rod body 6021. A first cone 6022 is fixedly connected to the top of the rod body 6021. The upper part of the first cone 6022 is conical. A second cone 6023 is fixedly connected to the middle of the rod body 6021. The upper and lower parts of the second cone 6023 are both conical. The volume of the second cone 6023 is larger than the volume of the first cone 6022.

[0091] When it is necessary to lock the collar on the pet's neck, the second electric push rod 5022 pushes the clamping arm 5023 to clamp the two sides of the collar 6 closer. At this time, the special-shaped lock rod 602 is inserted into the positioning component. In the first state, the first cone 6022 of the special-shaped lock rod 602 contacts the rotating lock arm 603. At this time, the first cone 6022 pushes the rotating lock arm 603 open, and the rotating lock arm 603 rotates a certain angle, defined as X°. At this time, the movable rod 604 compresses the return spring 606. When the first cone 6022 continues to move upward until it reaches the top of the rotating lock arm 603, at this time, the limiting rod 608 is located at the top of the limiting hole 607. Then the return spring resets. The return spring 606 pushes the rotating lock arm 603 to reset through the movable rod 604. The first cone 6022 moves to the top of the rotating lock arm 603, and the rotating lock arm 603 plays a supporting role for the first cone 6022, thereby completing the locking.

[0092] When it is necessary to remove the collar from the pet's neck, the second electric push rod 5022 pushes the clamping arm 5023 to clamp the two sides of the collar 6 and continue to move closer, that is, when continuing to move upward Figure 6, Prepare to enter the second state. At this time, the second vertebral body 6023 of the special-shaped lock rod 602 contacts the rotating lock arm 603. Since the volume of the second vertebral body 6023 is larger than that of the first vertebral body 6022, the second vertebral body 6023 pushes the rotating lock arm 603 open at this time, and the rotating lock arm 603 rotates by a certain angle, defined as Y°, where Y is greater than X. When the second vertebral body 6023 continues to move upward until it reaches the top of the rotating lock arm 603, the rotating lock arm 603 will not reset at this time. The special-shaped lock rod 602 moves downward until the second vertebral body 6023 contacts the bottom of the rotating lock arm 603. At this time, the second vertebral body 6023 drives the rotating lock arm 603 to rotate in the reverse direction to form a reset, thereby completing the unlocking movement. Then the collar can be removed from the pet's neck.

[0093] In Embodiments 1 and 2, a second electric push rod + clamping arm is used to realize the automatic disassembly and assembly of the collar. Combining the precise mechanical locking structure of a special-shaped lock rod + rotating lock arm + return spring enables the collar to be stably disassembled and assembled, avoiding pet discomfort or escape caused by manual disassembly and assembly. There is no solution in the prior art that combines pet feeding with the automatic disassembly and assembly of the collar. In the prior art, the collar is usually used as an independent device and cannot be intelligently disassembled and assembled in combination with pet behavior. However, in this solution, through feeding induction + automatic mechanical locking, the collar can be intelligently disassembled, reducing pet discomfort and improving the degree of intelligence. That is to say, by inducing the pet to reach into the position of the collar, non-compulsory intelligent wearing is realized, improving the user experience. Reducing manual intervention, reducing the pet's stress response, and improving safety and convenience.

[0094] Embodiment 3

[0095] On the basis of Embodiment 2, this embodiment further includes a high-definition camera and a control system. The high-definition camera is arranged at the bottom of the top plate 4, and the control system is an edge computing device. The edge computing device is used for local computer vision processing, such as NVIDIA Jetson AGX Orin, to reduce cloud latency. The collar 6 transmits the monitored pet heart rate, body temperature, and acceleration data to the edge computing device, and the edge computing device is used to run the intelligent pet health monitoring and warning method.

[0096] The intelligent pet health monitoring and warning method includes the following process:

[0097] Step 1, collect the daily behavior data of the pet through the high-definition camera and the collar 6, including visual behavior and physiological data. In Step 1, various behaviors of the pet such as eating, licking paws, walking, and lying down are detected. YOLOv8 or YOLO-NAS is used to improve the detection accuracy of small targets. Multi-scale feature fusion (FPN + PAN) is used to ensure that pets of different sizes can be detected. GhostNet lightweight CNN is used to reduce the computational load so that it can run on low-power devices. RGB + IR data fusion is used, combined with an infrared camera, to make the night monitoring effect more stable.

[0098] Collect consecutive frame data from an RGB camera, run the YOLOv8 / YOLO-NAS model to detect the pet's body parts, count the behavior data within different time periods, detect the staying time in front of the food bowl, and if it is lower than the normal threshold, there may be a problem of decreased appetite. If YOLO detects that the tongue frequently touches the paw, it may indicate anxiety or skin problems. If the pet keeps moving at high speed or stays still for too long, it may indicate health problems.

[0099] YOLO object detection, where the YOLO loss function is as follows:

[0100] L = λ 1 L coord + λ 2 L obj + λ 3 L noobj + λ 4 L class + λ 5 L temp ;

[0101] Where: L coord : Coordinate loss of the target box, which improves the target localization accuracy;

[0102] L obj : Target existence loss, which reduces the missed detection rate;

[0103] L noobj : Background error loss, which reduces the false detection situation;

[0104] L class : Class classification loss, which ensures that different behaviors of the pet can be accurately classified;

[0105] L temp , Temporal sequence consistency loss: A newly introduced optimization term used to consider the continuity of the pet's behavior and prevent short-term behaviors from being misjudged;

[0106] Where: B t : Detection box parameters of the t-th frame. This loss term ensures the stability of the prediction of the same target in consecutive frames and reduces the jitter error;

[0107] The above process can perform more stable behavior recognition, reduce single-frame misjudgment, and ensure that only continuously occurring behaviors will be recorded as abnormal. It can improve the detection robustness and reduce the false detection rate by combining historical frame information in low-light, occlusion, etc. Traditional YOLO mainly focuses on single-frame object detection, and after introducing the temporal consistency loss (L temp ), the object detection can consider the temporal changes of behaviors and improve the accuracy of behavior analysis.

[0108] Perform OpenPose pose analysis to analyze the key points of the pet's bones, calculate the movement trajectory, acceleration, and pose changes, extract its activity patterns. The key point detection formula is as follows:

[0109] v i，j (t) = p j (t) - p i (t);

[0110] p i (t): At time t, the coordinates of the i-th key point, v i，j (t): The relative vector between key points, used to calculate the limb pose;

[0111] Calculation of movement speed and acceleration:

[0112]

[0113] If is high for a long time, it indicates that the pet has abnormal jitter. If is close to zero, it indicates that the pet's movement has decreased, indicating that it is prone to being in a sick state;

[0114] Run OpenPose to detect the key points of the pet's head, limbs, and tail to calculate the pet's movement trajectory, and combine the results of YOLO to comprehensively judge the behavior pattern.

[0115] The above process can detect abnormal movement patterns (such as anxiety, trembling, lameness). Combining speed / acceleration analysis can reduce misjudgment. For example, being stationary is not abnormal and historical data needs to be combined for judgment. Traditional OpenPose is only used for human pose estimation, while here we introduce time series analysis (speed + acceleration), which can capture long-term trends and improve the reliability of abnormal behavior detection. This method is similar to trajectory prediction in motion analysis and is used for pet health monitoring for the first time.

[0116] Step 2: Combine the biological data of heart rate and body temperature to further judge whether the pet is abnormal. Collect the PPG or ECG of the pet through the collar 6 to obtain the heart rate signal and analyze the anxiety situation. The formula for heart rate variability (HRV) is:

[0117]

[0118] Where: RR i is the heart beat interval, is the average heart beat interval. If the HRV decreases, it indicates that the pet is prone to being in an anxious state;

[0119] If the body temperature is higher than the threshold, push an alarm message to the owner's mobile phone, and combine the behavior data of YOLO + OpenPose to further analyze whether medical treatment is needed.

[0120] The above process can reflect the stress level better than the static heart rate. It can detect psychological states such as anxiety and stress, rather than just diseases. Traditional pet health monitoring mainly focuses on body temperature and resting heart rate. The introduction of HRV into pet monitoring in this invention is more accurate.

[0121] Adopt YOLOv8 / YOLO-NAS object detection + OpenPose skeleton key point recognition, combined with infrared RGB data, to improve the detection stability under night and low-light conditions. Through multi-modal data fusion (vision + physiological monitoring), and using time series analysis, combined with heart rate, body temperature, and acceleration, a more accurate health status assessment is achieved, rather than single static monitoring. Compared with traditional single vision monitoring or collar sensor monitoring, this solution can provide more accurate health warnings and reduce false negatives and false positives.

[0122] Step 3, integrate computer vision and sensor data, and use machine learning to predict health risks.

[0123] First, perform data fusion. Input X = {x 1 , x 2 ,..., x n}, where x 1 = motion trajectory data, x 2 = behavior pattern data, x 3 = heart rate data, x 4 = body temperature data;

[0124] Use Kalman filter to fuse multiple sensor data: Let the measurement value be z k , the true state be x k , and the estimation equation: x k = Ax k-1 + Bu k + w k ;

[0125] x k : system state vector;

[0126] A: state transition matrix, describing the law of system change over time;

[0127] x k-1 : the system state at the previous moment;

[0128] B: control input matrix, indicating the external influence on the system;

[0129] u k : external input;

[0130] w k : process noise, representing the uncertainty in sensor measurement;

[0131] z k = Hx k + v k ;

[0132] z k : Sensor observation value;

[0133] H: Observation matrix, defining how the sensor measures the state variable;

[0134] v k : Measurement noise, representing the error existing in the sensor;

[0135] Eliminate noise through Kalman filtering to improve data reliability; can improve data fusion accuracy and reduce sensor errors. Can dynamically adjust the measurement weights to improve adaptability.

[0136] Adopt a deep time series prediction model:

[0137] h t = σ(W h h t-1 + W x x t + b h );

[0138]

[0139] y t = W y h t + b y ;

[0140] x t : Input data at the current moment;

[0141] h t : Hidden state of LSTM, storing long-term memory;

[0142] W h ,W x ,W y : Neural network weight matrices, learning the long-term change patterns of the input data;

[0143] b h ,b y : Bias terms, helping to adjust the activation thresholds of neurons;

[0144] σ: Activation function;

[0145] c t : Cell state, storing the long-term health information of the pet;

[0146] f t ,i t , Internal gating mechanism of LSTM: f t Forget gate: determines whether to forget the health state at the previous moment, i t Input gate: determines whether to accept the new health data at the current moment, The new information at the current moment, together with the input gate, updates the cell state;

[0147] Input x t : includes YOLO behavior data, OpenPose motion data, heart rate, and body temperature in the past seven days, and outputs y t : The pet health status score for the next 24 hours is 0 - 100. Scoring threshold: 80 - 100 is healthy, 50 - 79 is mildly abnormal, 0 - 49 is severely abnormal, and an alarm is triggered;

[0148] The above process can capture long - term trends and reduce single - point misjudgment. Based on historical data, it can predict the health status of different pets individually. Traditional health monitoring only sets thresholds based on rules, while LSTM allows for personalized modeling to adapt to the health patterns of different pets.

[0149] Adopt LSTM time - series analysis to establish a personalized health model to predict the pet health status, rather than relying solely on fixed thresholds to determine whether it is abnormal. Combine Kalman filtering to eliminate noise, avoid single - frame misjudgment, and improve data reliability. Adopt reinforcement learning to automatically optimize the health scoring model, improve adaptability, and reduce false alarms. It is more intelligent than the existing fixed - threshold - based monitoring method and can adapt to the changes in the health status of different pet individuals.

[0150] Adopt reinforcement learning to adjust the health scoring model based on pet historical data. Reward function: R(s, a) = -(y t -y baseline ) 2 ; Let the AI adjust the alarm threshold according to historical data to reduce false alarms.

[0151] This embodiment comprehensively adopts technologies such as the Internet of Things, computer vision, deep learning, and sensor fusion to construct a highly intelligent pet cage, which can automatically disassemble and assemble collars, intelligently feed, and real - time monitor the pet's health status; combine vision + physiological data + deep learning to provide accurate health warnings, reduce the burden of manual monitoring, and timely detect pet health abnormalities; edge computing + Kalman filtering + LSTM to ensure prediction accuracy, monitor all - weather, and the combination of RGB + IR to ensure stable monitoring during both day and night. This embodiment not only improves the convenience of pet management but also provides strong intelligent protection for the health of pets.

[0152] Although the present invention has been disclosed above in preferred embodiments, it is not intended to limit the present invention. Those of ordinary skill in the art to which the present invention pertains can make various modifications and refinements without departing from the spirit and scope of the present invention. Therefore, the protection scope of the present invention shall be determined by the scope defined in the claims.

Claims

1. A smart pet cage based on the Internet of Things, characterized by: include: A base (1), wherein the base (1) is fixedly connected to a fence (2) on all sides, an entry and exit door (3) is provided on the fence (2), and a top plate (4) is fixedly connected to the top of the fence (2); A collar disassembly and assembly mechanism (5), the collar disassembly and assembly mechanism (5) is used to disassemble and assemble a collar (6) on the neck of a pet, the collar disassembly and assembly mechanism (5) comprises a shell (501), a hole (506) is provided on the front side of the shell (501), an operating component (502) is provided in the hole (506), the operating component (502) is used to disassemble or install the collar (6) on the neck, and a first electric push rod (503) is provided in the shell (501), the first electric push rod (503) is provided in the shell (501), The output end of the electric push rod (503) is fixedly connected to a pet bowl (504), a pet food bin (505) is arranged on the top of the pet bowl (504), the pet food bin (505) is fixedly connected to the housing (501), a solenoid valve is arranged on the bottom of the pet food bin (505), the pet food bin (505) is used to add pet food into the pet bowl (504), and the first electric push rod (503) is used to push the pet bowl (504) to the hole (506).

2. According to claim 1, the smart pet cage based on the Internet of Things is characterized by: The operating assembly (502) comprises an operating box (5021), wherein a second electric push rod (5022) is arranged inside the operating box (5021), and the second electric push rod (5022) is provided with a clamping arm (5023), and the clamping arm (5023) is used to clamp the collar (6).

3. The smart pet cage based on the Internet of Things according to claim 1, characterized in that: The collar (6) is used to monitor the heart rate, body temperature, and acceleration of a pet. The collar (6) comprises two left and right semicircular collars. The collar (6) is provided with a heart rate sensor, a body temperature sensor, and a pressure acceleration sensor.

4. The smart pet cage based on the Internet of Things according to claim 3, characterized in that: Both ends of the semicircular collar on the left are provided with positioning components, and both ends of the semicircular collar on the right are fixedly connected with special-shaped locking rods (602). The positioning component includes a lock box (601) and a rotating lock arm (603). The rotating lock arm (603) is movably connected to the lock box (601) through a rotating shaft. The upper part of the rotating lock arm (603) is movably connected to a movable rod (604), and one end of the movable rod (604) can be inserted into the positioning arm (605). A return spring (606) is fixedly connected in the middle, and the return spring (606) is arranged in the inner cavity of the positioning arm (605). The positioning arm (605) is movably connected to the lock box (601). A limiting rod (608) is fixedly connected to the movable rod (604). A limiting hole (607) adapted to the limiting rod (608) is opened on the positioning arm (605), and the limiting rod (608) is inserted into the limiting hole (607).

5. The smart pet cage based on the Internet of Things according to claim 4, characterized in that: The special-shaped locking rod (602) comprises a rod body (6021), the top of the rod body (6021) is fixedly connected to a first vertebral body (6022), the upper part of the first vertebral body (6022) is conical, the middle part of the rod body (6021) is fixedly connected to a second vertebral body (6023), the upper and lower parts of the second vertebral body (6023) are both conical, and the volume of the second vertebral body (6023) is greater than the volume of the first vertebral body (6022).

6. The smart pet cage based on the Internet of Things according to claim 5, characterized in that: When the special-shaped locking rod (602) is inserted into the positioning assembly, a first state is defined, the first vertebra (6022) of the special-shaped locking rod (602) contacts the rotating locking arm (603), at which time the first vertebra (6022) and the rotating locking arm (603) are pushed open, and the rotating locking arm (603) rotates a certain angle, which is defined as X°, and when the first vertebra (6022) continues to move upward until it reaches the top of the rear rotating locking arm (603), the reset spring (606) pushes the rotating locking arm (603) to reset through the movable rod (604), and the rotating locking arm (603) supports the first vertebra (6022), thereby completing the locking; When the special-shaped locking rod (602) is inserted into the positioning assembly and reaches the first state, and continues to move upward, the second state can be reached. The second state is defined as the second vertebra (6023) of the special-shaped locking rod (602) contacts the rotating locking arm (603). At this time, the second vertebra (6023) and the rotating locking arm (603) are pushed open, and the rotating locking arm (603) is rotated by a certain angle, which is defined as angle Y°, where Y is greater than X. When the second vertebra (6023) continues to move upward until it reaches the top of the rear rotating locking arm (603), the rotating locking arm (603) will not be reset. The special-shaped locking rod (602) moves downward until the second vertebra (6023) contacts the bottom of the rotating locking arm (603). At this time, the second vertebra (6023) drives the rotating locking arm (603) to rotate in the opposite direction, forming a reset, thereby completing the unlocking movement.

7. The smart pet cage based on the Internet of Things according to claim 1, characterized in that: The invention comprises a high-definition camera and a control system, wherein the high-definition camera is arranged at the bottom of the top plate (4); the control system is an edge computing device, and the edge computing device is used for local computer vision processing to reduce cloud latency; the collar (6) transmits the monitored pet heart rate, body temperature, and acceleration data to the edge computing device, and the edge computing device is used to run the intelligent pet health monitoring and early warning method.

8. The smart pet cage based on the Internet of Things according to claim 7, characterized in that: The intelligent pet health monitoring and early warning method includes the following process: Step 1, collecting daily behavior data of the pet through a high-definition camera and a collar (6), including visual behavior and physiological data; YOLO target detection, where the YOLO loss function is as follows: L=λ1L coord +λ2L obj +λ3L noobj +λ4L class +λ5L temp ; Where: L coord : Target frame coordinate loss, improving target positioning accuracy; L obj : The target is lost and the missed detection rate is reduced; L noobj : Background error loss, reducing false detection; L class : Category classification loss, ensuring that different behaviors of pets can be accurately classified; L temp ,Time series consistency loss: A newly introduced optimization term used to consider the continuity of pet behavior and prevent short-term behaviors from being misjudged; Among them: B t : The detection box parameters of the tth frame. This loss term ensures the stability of the prediction of the same target in consecutive frames and reduces the jitter error; Perform OpenPose posture analysis, analyze the pet's skeleton key points, calculate the motion trajectory, acceleration, posture changes, and extract its activity mode. The key point detection formula is as follows: v i，j (t)=p j (t)-p i (t); p i (t): time t, the coordinates of the i-th key point, v i，j (t): relative vector between key points, used to calculate limb posture; Movement speed and acceleration calculation: like If the value is high for a long time, it means that the pet is shaking abnormally. If it is close to zero, it means that the pet has less exercise, which means that it is prone to illness; Run OpenPose to detect the key points of the pet's head, limbs, and tail to calculate the pet's movement trajectory, and combine the results of YOLO to comprehensively judge the behavior pattern; Step 2, combining heart rate and body temperature biological data, further determine whether the pet is abnormal, collect the pet's PPG or ECG through the collar 6 to obtain the heart rate signal, analyze the anxiety situation, and calculate the heart rate variability (HRV) formula: Among them: RR i is the heartbeat interval, It is the average heartbeat interval. If HRV decreases, it indicates that the pet is prone to anxiety. If the body temperature is higher than the threshold, an alarm message will be pushed to the owner's mobile phone, and combined with the behavioral data of YOLO+OpenPose, further analysis will be performed to determine whether medical treatment is needed; Step 3: Integrate computer vision and sensor data to predict health risks using machine learning. First, data fusion is performed, and input X = {x1, x2, ..., x n }, where x1 = motion trajectory data, x2 = behavior pattern data, x3 = heart rate data, x4 = body temperature data; Use Kalman filtering to fuse multiple sensor data: Let the measurement value be z k , the true state is x k , the estimated equation: k =Ax k-1 +Bu k +w k ; x k : system state vector; A: state transfer matrix, describing the law of system change over time; x k-1 : The system status at the last moment; B: control input matrix, indicating the external influence on the system; u k : External input; w k : Process noise, which represents the uncertainty in the sensor measurement; z k =Hx k +v k ; z k : sensor observation value; H: observation matrix, which defines how the sensor measures the state variables; v k : Measurement noise, indicating the error of the sensor; Eliminate noise and improve data reliability through Kalman filtering; Using deep time series forecasting models: h t =σ(W h h t-1 +W x x t +b h ); y t =W y h t +b y ; x t : Input data at the current moment; h t : The hidden state of LSTM, storing long-term memory; W h , W x , W y : Neural network weight matrix, learning the long-term change pattern of input data; b h , b y : Bias term, which helps adjust the activation threshold of neurons; σ: activation function; c t : Cell status, storing long-term health information of pets; f t ,i t , LSTM internal gate control mechanism: f t Forget gate: decides whether to forget the health status of the previous moment, i t Input gate: decides whether to accept new health data at the current moment. The new information at the current moment updates the cell state together with the input gate; Input x t : Includes YOLO behavior data, OpenPose motion data, heart rate, body temperature, and output y t : The pet health status score for the next 24 hours is 0-100, with score thresholds: 80-100 for health, 50-79 for mild abnormality, and 0-49 for severe abnormality, which triggers an alarm; Reinforcement learning is used to adjust the health score model based on the pet's historical data. The reward function is: R(s, a) = -(y t -y baseline ) 2 ; Let AI adjust the alarm threshold based on historical data to reduce false alarms.

9. The smart pet cage based on the Internet of Things according to claim 8, characterized in that: In step 1, continuous frame data of the RGB camera is collected, the YOLOv8 / YOLO-NAS model is run to detect the body parts of the pet, the behavioral data in different time periods are counted, and the stay time in front of the food bowl is detected. If it is lower than the normal threshold, there may be a problem of decreased appetite. If YOLO detects that the tongue frequently contacts the paws, it may indicate anxiety or skin problems. If the pet continues to move at a high speed or stays still for too long, it may indicate health problems.

10. The smart pet cage based on the Internet of Things according to claim 8, characterized in that: In step 1, various behaviors of the pet, such as eating, licking paws, walking, and lying down, are detected. YOLOv8 or YOLO-NAS is used to improve the accuracy of small target detection. Multi-scale feature fusion (FPN+PAN) is used to ensure that pets of different sizes can be detected. GhostNet lightweight CNN is used to reduce the amount of calculation so that it can run on low-power devices. RGB+IR data fusion is used in combination with infrared cameras to achieve more stable night monitoring effects.

Images