Pet washing and caring system and method based on visual recognition and large model linkage

Abstract

The application discloses a pet washing and protecting system and method based on visual identification and large model linkage, and the method comprises the following steps: S1, placing a pet in a washing and protecting cabin, and acquiring multi-angle pet time sequence images; S2, sending the acquired pet time sequence images and the equipment state parameters of the washing and protecting cabin to a large model intelligent terminal; S3, the large model intelligent terminal generates a washing and protecting strategy and a washing and protecting time; S4, sending the generated washing and protecting strategy and washing and protecting time to the washing and protecting cabin; S5, analyzing the washing and protecting strategy and performing overrun checking, and generating a control instruction after the checking is passed, and periodically returning the cabin temperature and humidity and pet action feedback within the washing and protecting time. The application analyzes the pet breed, body shape, hair quality / hair length and skin state based on the acquired pet time sequence images through the large model intelligent terminal, and fuses the breed priori knowledge base, generates stage customized parameters for each pet, and ensures the customized implementation of the corresponding pet washing and protection.

Description

Technical Field

[0001] This invention relates to the field of intelligent pet grooming technology, and in particular to a pet grooming system and method based on visual recognition and large-scale model linkage. Background Technology

[0002] With the rapid development of the pet economy, pet grooming services are gradually evolving from traditional manual operations to automation and intelligence. Existing automatic pet grooming equipment typically uses fixed grooming programs, performing fixed cleaning steps according to pre-set fixed times, water pressures, water temperatures, detergents, and dosages.

[0003] However, pets of different breeds, sizes, coat types, and skin conditions have significantly different requirements for grooming parameters. Furthermore, pets may exhibit behaviors such as quieting, struggling, or stress during grooming. Therefore, using fixed, preset cleaning procedures cannot dynamically adjust grooming strategies based on the pet's real-time feedback. This not only results in poor grooming outcomes but also lacks appropriate measures to address pet stress, potentially leading to pet injury and equipment damage. In light of this, this solution proposes a pet grooming system and method based on visual recognition and large-scale model integration to address these issues. Summary of the Invention

[0004] The purpose of this invention is to provide a pet grooming system and method based on visual recognition and large model linkage, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a pet grooming method based on visual recognition and large model linkage, comprising the following grooming operation steps:

[0006] S1. Place the pet in the grooming chamber and start shooting according to the resting shooting frequency or the action shooting frequency after detecting changes in the pet's posture to obtain multi-angle time-series images of the pet.

[0007] S2, the obtained pet time-series images are encoded and sent to the large model smart terminal along with the equipment status parameters of the grooming cabin;

[0008] S3, the large model intelligent terminal analyzes pet breed, estimates body size, analyzes fur quality / length and infers skin condition based on pet time-series images, and generates washing and care strategies and washing and care times by combining the breed prior knowledge base and fur condition discrimination; the washing and care strategy includes independent control parameters for the pre-rinse stage, cleaning stage, rinsing stage and drying stage, and each stage parameter is set with a limited safety threshold.

[0009] S4, send the generated washing and care strategy and washing and care time to the washing and care chamber;

[0010] S5 analyzes the washing and care strategy and performs over-limit verification. After the verification is passed, it generates control commands and drives the corresponding spraying, drying and temperature control. At the same time, it periodically transmits the temperature and humidity inside the cabin and the pet's behavior feedback during the washing and care time. If abnormal stress is detected, it will automatically pause and adjust the parameters.

[0011] Preferably, the following start / stop operation is used when acquiring the pet time-series images:

[0012] First, set the rest shooting frequency f. 静 and action shooting frequency f 动 Then, the threshold of dynamic changes in the pet during the filming was detected. and stress shutdown threshold The exercise intensity M is calculated every fixed time interval. t ;

[0013] When M t < When the pet is motionless, start filming at the resting filming frequency;

[0014] when ≤M t < When the pet is moving, switch the action shooting frequency to start shooting;

[0015] When M t ≥ If the test is repeated more than N times, it indicates that the pet is stressed, and all photography and grooming activities should be stopped immediately.

[0016] Preferably, the motion intensity M t The calculation includes the following steps:

[0017] First, obtain the rectangular activity regions of the pet in the image at the current and previous sampling times;

[0018] Calculate the area of ​​the rectangular active region at the current time and the area of ​​the rectangular active region at the previous sampling time, respectively;

[0019] Calculate the absolute value of the difference between the two areas;

[0020] Find the center points of the current rectangular active region and the rectangular active region at the previous sampling time, respectively;

[0021] Calculate the distance between the two center points and divide it by the height of the current rectangular active area;

[0022] The motion intensity M is obtained by adding the absolute value of the area difference and the calculated value of the center point distance divided by the rectangle height through weight allocation. t .

[0023] Preferably, the specific operation steps of S2 are as follows:

[0024] S201, determine the sending trigger condition, and decide whether to send based on the shooting start and stop operation, and send the complete pet time sequence image and device status parameters or only the device status parameters;

[0025] S202, Encode the pet time-series images that need to be sent;

[0026] S203 packages the encoded pet time-series images and the status parameters of the grooming cabin equipment and sends them to the large model intelligent terminal.

[0027] Preferably, the triggering conditions include the following:

[0028] When the shooting mode is in rest shooting mode and the last transmission is less than 30 seconds, only the pet time sequence image and device status parameters captured in rest shooting mode are sent.

[0029] When the shooting mode is in silent shooting mode and more than 30 seconds have passed since the last transmission, only one image of the pet from any angle will be sent.

[0030] When the shooting mode is motion shooting and the time since the last large model intelligent terminal inference has not exceeded the throttling window, the pet time sequence image will not be sent temporarily;

[0031] When the shooting mode is motion shooting and the time since the last large model intelligent terminal inference exceeds the throttling window, send the complete pet time series image for the corresponding time period;

[0032] When switching shooting frequencies, send complete time-series images of the pet for the corresponding time period;

[0033] When the pet is in a stressed state, send the last set of complete pet time-series images and stop all grooming activities.

[0034] Preferably, step S3 specifically includes the following steps:

[0035] S301, the large model intelligent terminal receives and parses the encoded pet time-series image data and the status parameters of the pet grooming cabin equipment;

[0036] S302 identifies pet breeds based on parsed pet time-series image data and combined with image classification models; estimates pet body length, shoulder height, chest circumference, and weight range through pet outline or limb characteristics; outputs coat quality classification by analyzing pet hair texture, density, curliness, and average length; and outputs abnormal locations and severity by detecting erythema, dandruff, eczema, parasites, or external injuries in skin areas.

[0037] S303, query the breed prior knowledge base, retrieve the corresponding breed's skin sensitivity, suitable water temperature range, hair washing and care precautions and susceptible diseases based on the identified breed, compare the contents of the prior knowledge base with the actual analysis results, and correct the judgment of the pet's condition.

[0038] S304 generates a washing and care strategy that includes a pre-rinse stage, a cleaning stage, a rinsing stage, and a drying stage. If shampoo is used in the cleaning stage, the water temperature in the rinsing stage is increased and the number of water changes is increased. If the pet has sensitive skin, the water temperature in all stages is lowered to the lower limit, and the speed of the roller brush is reduced.

[0039] Preferably, the pre-rinsing stage is set with a water temperature of 25℃-40℃, a water pressure of 50-400kPa, a spray angle, and a duration of ≤3min;

[0040] The cleaning stage settings include the type and amount of bath liquid used;

[0041] The rinsing stage is set with a water temperature of 30℃-42℃;

[0042] The drying stage is set with an air temperature of 35℃-42℃, an air speed of 10-60m / s, and a total time of ≤30min.

[0043] Preferably, step S5 specifically includes the following operation steps:

[0044] S501, analyze the washing and care strategy and perform over-limit verification. Check whether the parameters of the pre-rinse stage, cleaning stage, rinsing stage and drying stage are within the preset safety threshold range. If any parameter exceeds the safety threshold, it is marked as invalid and an alarm is triggered. At the same time, the corresponding strategy is refused to be executed.

[0045] S502, after the over-limit verification is passed, a control command is generated, which converts the parameters of each stage into drive signals and sends them to the corresponding execution equipment in the washing and care chamber.

[0046] S503 collects temperature, humidity, and pet's current activity level (M) inside the bathing chamber every 1-2 seconds during the bathing / care time. t The data is transmitted to a large-scale intelligent terminal. If the analysis indicates pet stress or abnormal temperature or humidity, the bathing and grooming process will be stopped immediately.

[0047] The pet grooming system based on visual recognition and large model linkage is used to execute the pet grooming method based on visual recognition and large model linkage, including a grooming chamber, a camera module, an equipment status acquisition module, a communication module, a large model intelligent terminal, and an execution control module.

[0048] The washing and care chamber is equipped with a variety of washing and care devices for performing washing and care operations;

[0049] The camera module is installed on the inner wall of the pet grooming chamber and is used to continuously collect multi-angle image sequences of the pet according to a preset time interval or trigger signal.

[0050] The equipment status acquisition module is deployed at the control terminal of the washing and care chamber and various washing and care equipment, and is used to collect the status parameters of the washing and care chamber and the washing and care equipment.

[0051] The large-scale intelligent terminal is deployed in the cloud or at the edge. It integrates a multimodal recognition network and includes at least a breed classification branch, a body size estimation branch, a coat quality / length analysis branch, and a skin sensitivity inference branch. It is used to output washing and care parameters based on multi-angle image sequences of pets, including but not limited to water temperature range, water pressure level, detergent dosage, air temperature curve, and washing and care time.

[0052] The communication module is used to transmit data collected by the camera module and the device status acquisition module to the large-scale intelligent terminal;

[0053] The execution control module receives the washing and care strategy from the large-scale intelligent terminal through the communication module, and generates washing and care instructions to send to various washing and care devices in the washing and care chamber for execution.

[0054] Preferably, the various washing and care devices include, but are not limited to, adjustable spray devices, drying devices, temperature control devices, and bath liquid application devices.

[0055] The technical effects and advantages of this invention are as follows:

[0056] (1) This invention uses a large model intelligent terminal to comprehensively analyze the pet breed, body shape, coat quality / length and skin condition based on the acquired pet time-series images, and integrates the breed prior knowledge base to generate phased customized parameters for each pet, ensuring the customized implementation of pet grooming; at the same time, the acquisition of pet time-series images is activated by shooting at the resting shooting frequency or the action shooting frequency after detecting changes in pet posture, which not only ensures the capture of key action frames, but also greatly reduces the communication and computing load.

[0057] (2) After the large-scale intelligent terminal of this invention outputs the washing and care strategy, it needs to perform over-limit verification to eliminate the risk of over-limit and ensure the safe and orderly washing and care arrangement. At the same time, during the pet washing and care process, the pet's status is detected and fed back in real time. If a risk is detected, the machine is stopped immediately to ensure the pet's safety and avoid equipment damage. Attached Figure Description

[0058] Figure 1 This is a flowchart illustrating the operation of the washing and care method of the present invention.

[0059] Figure 2 This is a logical block diagram of the washing and care system of the present invention. Detailed Implementation

[0060] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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.

[0061] Example 1: The present invention provides as follows Figure 1 The pet grooming method shown here, based on visual recognition and large model linkage, includes the following grooming steps:

[0062] S1. Place the pet to be bathed and cared for inside the bathing and caring chamber. Start shooting according to the resting shooting frequency or the action shooting frequency after detecting changes in the pet's posture. Through multiple camera modules arranged inside the bathing and caring chamber, obtain multi-angle time-series images of the pet from multiple shooting positions inside the bathing and caring chamber.

[0063] Specifically, the following start / stop operation is used when acquiring pet time-series images:

[0064] First, set the rest shooting frequency f. 静 and action shooting frequency f 动 Among them, the resting shooting frequency refers to the shooting frequency when the pet is not moving or remains in the previous posture, and the action shooting frequency refers to the shooting frequency when the pet is moving; then, the dynamic change threshold of the pet during the shooting period is detected. and stress shutdown threshold The exercise intensity M is calculated every fixed time interval. t ;

[0065] When M t < When the pet is motionless, start filming at the resting filming frequency;

[0066] when ≤M t < When the pet is moving, switch the action shooting frequency to start shooting;

[0067] When M t ≥ If the test is repeated more than N times, it indicates that the pet is stressed, and all photography and grooming activities should be stopped immediately.

[0068] Among them, the exercise intensity M t The calculation includes the following steps:

[0069] First, obtain the rectangular activity regions of the pet in the image at the current time and the previous sampling time, i.e., the pet's rectangular activity region in the image at the current time is denoted as R. t The height of the rectangle is denoted as h. t The rectangular area of ​​activity of the pet in the image at the previous sampling time is denoted as R. t-1 ;

[0070] Calculate the areas of the rectangular active region at the current time and the rectangular active region at the previous sampling time, respectively, where the area of ​​the rectangular active region at the current time is denoted as S. t The area of ​​the rectangular active region at the previous sampling time is denoted as S. t-1 ;

[0071] The absolute value of the difference between two areas is expressed as:

[0072] (1);

[0073] Find the center points of the current rectangular active region and the rectangular active region at the previous sampling time, respectively. The center point of the current rectangular active region is denoted as C. t The center point of the rectangular active region at the previous sampling time is denoted as C. t-1 ;

[0074] The calculation of the distance between the two center points, divided by the height of the current rectangular active area, is expressed as:

[0075] (2);

[0076] Among them, (x) t y t (x) represents the coordinates of the center point of the current rectangular active area; (x) t-1 y t-1 () represents the coordinates of the center point of the rectangular active region at the previous sampling time;

[0077] The motion intensity M is obtained by adding the absolute value of the area difference and the calculated value of the center point distance divided by the rectangle height through weight allocation. t The calculation is expressed as:

[0078] (3);

[0079] in, and The weights of area change and distance from center point are respectively represented by , and .

[0080] By employing a shooting method that combines resting and active shooting frequencies, the communication and computational load is reduced. Motion intensity quantization drives two-level frame rate switching. When the pet is resting, the shooting frequency is significantly reduced, minimizing redundant image transmission and processing. When the pet is active, the frame rate is increased to ensure key postures are captured, guaranteeing rapid communication processing and accurate acquisition of temporal images of the pet within the grooming chamber. Simultaneously, a stress response is determined through N consecutive threshold exceedances, effectively filtering single-frame detection noise and brief disturbances to avoid accidental shutdowns. Furthermore, genuine, continuous struggling is promptly identified and triggers shutdown, ensuring pet safety. The quantification of motion intensity integrates area change and center point distance. The area change reflects the change in posture scale, the center point distance calculation reflects translational motion, and the normalization across body size and distance is achieved by dividing by the current rectangle height, making the threshold universal.

[0081] S2, the obtained pet time-series images are encoded and sent to the large model smart terminal along with the equipment status parameters of the grooming cabin;

[0082] Specifically, the specific operation steps of S2 are as follows:

[0083] S201, determine the sending trigger condition, and decide whether to send based on the shooting start and stop operation, and send the complete pet time sequence image and device status parameters or only the device status parameters;

[0084] The triggering conditions include the following:

[0085] When the shooting mode is in silent shooting mode, and the time since the last transmission is less than 30 seconds, only the pet time-series images captured during silent shooting and the device status parameters are transmitted. In this case, the pet time-series images captured during silent shooting refer to the time from the last transmission to the current time. Press f... 静 The sequence consists of all captured frames, with a heartbeat interval set to thirty seconds, according to f. 静 With a typical value of 0.5-1fps, it can accumulate 1-2 frames to tens of frames, balancing the input continuity and communication overhead of large-scale intelligent terminals.

[0086] When the shooting mode is at rest and more than 30 seconds have passed since the last transmission, only one image of the pet at any angle is sent. Since the change in the pet's posture is negligible in the rest shooting mode, a single frame is sufficient to represent the current state, and continuing to send the sequence would cause redundancy.

[0087] When the shooting state is motion shooting and the time since the last inference of the large model intelligent terminal has not exceeded the throttling window, the pet time sequence image is not sent temporarily. In the settings of this solution, the throttling window is set to 3-5 seconds. Based on the time consumption of a single round of inference of the large model intelligent terminal and the frame rate of motion state, the motion segment images are submitted in segments. The images that are not sent temporarily are stored in the local cache and sent together with the subsequent frames after the throttling window expires.

[0088] When the shooting mode is motion capture, and the time since the last large-scale intelligent terminal inference exceeds the throttling window, a complete pet time-series image for the corresponding time period is sent. The complete pet time-series image for the corresponding time period is from the end of the last large-scale inference to the time of this triggering. The local cache contains images stored by pressing f. 动 All frames captured;

[0089] When switching shooting frequencies, a complete pet time-series image for the corresponding time period is sent. This time-series image is a frame sequence from Δt1 before the switch to Δt2 after the switch, where Δt1 and Δt2 take 1-2 sampling periods.

[0090] When the pet is in a state of stress, send the last set of complete pet time-series images and stop all grooming activities to avoid causing injury to the pet after stress.

[0091] S202, encode the pet time-series images to be sent; the encoding format uses the JPEG image compression standard to encode single-frame images, and for continuous frame sequences, H.265 video encoding is used to combine multiple frames into a GOP to further reduce the bit rate; when images from multiple cameras are sent simultaneously, each camera image is encoded independently, and the camera ID and shooting location label are appended to the header of the encoded data packet.

[0092] S203, the encoded pet time-series images and the status parameters of the grooming chamber equipment are packaged and sent to the large model smart terminal. The equipment status parameters include at least the current grooming stage identifier, current water temperature ℃, water pressure kPa, brush head speed rpm, air volume m³ / h, air temperature ℃, humidity in the chamber %RH, real-time pet weight reading kg, current shooting frequency status, and cumulative grooming time s.

[0093] S3, the large model intelligent terminal analyzes pet breed, estimates body size, analyzes fur quality / length and infers skin condition based on pet time-series images, and generates washing and care strategies and washing and care times by combining the breed prior knowledge base and fur condition discrimination; the washing and care strategy includes independent control parameters for the pre-rinse stage, cleaning stage, rinsing stage and drying stage, and each stage parameter is set with a limited safety threshold.

[0094] Specifically, S3 includes the following operational steps:

[0095] The S301 large-scale intelligent terminal receives and parses encoded pet time-series image data and grooming chamber equipment status parameters. It decodes the received JPEG encoded image data, restoring it to an RGB three-channel image matrix with the same resolution as before encoding. Multiple camera images are aligned according to their acquisition timestamps; if multiple images exist simultaneously, they are arranged into an image sequence according to camera ID. Equipment status parameters are extracted, including grooming stage, water temperature, water pressure, brush head speed, airflow, air temperature, chamber humidity, pet weight, shooting frequency, and the most recent M... t The values ​​are calculated, and each parameter is validated for validity.

[0096] S302 identifies pet breeds based on parsed pet time-series image data and combined with image classification models; estimates pet body length, shoulder height, chest circumference, and weight range through pet outline or limb characteristics; outputs coat quality classification by analyzing pet hair texture, density, curliness, and average length; and outputs abnormal locations and severity by detecting erythema, dandruff, eczema, parasites, or external injuries in skin areas.

[0097] S303, query the breed prior knowledge base, retrieve the corresponding breed's skin sensitivity, suitable water temperature range, hair washing and care precautions, and common diseases based on the identified breed, compare the content in the prior knowledge base with the actual analysis results, and correct the judgment of the pet's condition. The breed prior knowledge base is stored in a relational database, and each record includes: breed name, skin sensitivity level, suitable water temperature range ℃, suitable water pressure range kPa, upper limit of brush speed rpm, list of common skin diseases, and washing and care contraindications;

[0098] S304 generates a washing and care strategy that includes a pre-rinse stage, a cleaning stage, a rinsing stage, and a drying stage. If shampoo is used in the cleaning stage, the water temperature in the rinsing stage is increased and the number of water changes is increased. If the pet has sensitive skin, the water temperature in all stages is lowered to the lower limit, and the speed of the roller brush is reduced.

[0099] The pre-rinse stage settings include water temperature of 25℃-40℃, water pressure of 50-400kPa, spray angle of 45°-90°, and duration of ≤3min. When the pet is deemed to have sensitive skin, the water temperature should be set to the lower limit of 25℃-30℃ and the water pressure to the lower limit of 50-100kPa.

[0100] The cleaning stage settings include the type and amount of bath liquid. In subsequent extended applications, when the cleaning chamber is equipped with a brushing structure for pet rolling and rubbing, the cleaning stage settings also include setting the rolling brush speed to 100-400 rpm and the rubbing time to ≤10 min. When abnormalities are detected on the pet's skin, a hypoallergenic bath liquid is used. If a brushing structure is provided, the rolling brush speed is reduced to 100-200 rpm.

[0101] The rinsing stage is set with a water temperature of 30℃-42℃; if bath liquid is used in the cleaning stage, the rinsing water temperature is increased to 38℃-42℃ to promote the dissolution of bath liquid residue, and the number of water changes is increased to ≥3 times.

[0102] The drying stage is set with an air temperature of 35℃-42℃, an air speed of 10-60m / s, and a total time of ≤30min; among them, the air temperature for curly wool varieties should not exceed 45℃ to prevent frizz, and the air speed for double-layer wool varieties should not be less than 2 m / s to ensure that the bottom layer is thoroughly dried.

[0103] In the implementation of this solution, the pet breed, body size, coat quality and skin are analyzed through multi-angle shooting data, which enables personalized pet washing and care solutions. At the same time, the output parameters of the large model are limited by preset safety thresholds, and the device performs a second verification before execution to prevent parameters exceeding the limit from being executed, thus ensuring the safety of the washing and care.

[0104] S4. The generated washing and care strategy and washing and care time are sent to the washing and care chamber. The washing and care strategy includes the instructions for the operation. It is sent to the control terminal of each device in the washing and care chamber, and the central control of the washing and care chamber performs corresponding operation sequence execution control and real-time monitoring.

[0105] S5 analyzes the washing and care strategy and performs over-limit verification. After the verification is passed, it generates control commands and drives the corresponding spraying, drying and temperature control. At the same time, it periodically transmits the temperature and humidity inside the cabin and the pet's behavior feedback during the washing and care time. If abnormal stress is detected, it will automatically pause and adjust the parameters.

[0106] Specifically, S5 includes the following operation steps:

[0107] S501 analyzes the washing and care strategy and performs over-limit verification. It checks whether the parameters of the pre-rinse stage, cleaning stage, rinsing stage and drying stage are within the preset safety threshold range, including but not limited to detecting water temperature, water pressure, speed and air temperature. If any parameter exceeds the safety threshold, it is marked as invalid and an alarm is triggered. At the same time, the corresponding strategy is refused to be executed and the process is suspended for manual confirmation.

[0108] S502, after the over-limit verification is passed, a control command is generated, which converts the parameters of each stage into drive signals and sends them to the corresponding execution equipment in the washing and care chamber.

[0109] S503 collects temperature, humidity, and pet's current activity level (M) inside the bathing chamber every 1-2 seconds during the bathing / care time. tThe collected data is transmitted to the large-scale intelligent terminal. If the analysis indicates pet stress or abnormal temperature and humidity, the washing and grooming process is immediately stopped, and a safety ventilation and cooling program is initiated to ensure the pet's safety. Continuous monitoring during the washing and grooming process ensures the identification of pet stress behaviors and real-time monitoring of equipment operation, guaranteeing real-time protection and monitoring of the pet during the washing and grooming process.

[0110] Example 2: The present invention provides as follows Figure 2 The pet grooming system shown is based on visual recognition and large model linkage. It is used to execute the pet grooming method based on visual recognition and large model linkage in Embodiment 1. It includes a grooming chamber, a camera module, an equipment status acquisition module, a communication module, a large model intelligent terminal, and an execution control module.

[0111] The bathing and grooming chamber is equipped with various bathing and grooming devices for performing bathing and grooming operations. These devices include, but are not limited to, adjustable spray equipment, air drying equipment, temperature control equipment, and bath liquid application equipment. The bathing and grooming chamber provides a closed working space for pet bathing and grooming, and is waterproof and non-slip, and can restrict the pet's range of movement. A sensor for weighing the pet can be installed on the base of the bathing and grooming chamber for weighing when estimating the pet's size.

[0112] The camera module is installed on the inner wall of the bathing and grooming chamber and is used to continuously collect multi-angle image sequences of the pet according to a preset time interval or trigger signal. The camera module has the functions of being splash-proof and fog-proof, ensuring that the pet images can be acquired even during the bathing and grooming process.

[0113] The equipment status acquisition module is deployed in the washing and care chamber and the control terminal of various washing and care devices to collect status parameters of the washing and care chamber and the washing and care devices. The equipment status acquisition module includes components configured inside the washing and care chamber and at the control terminal of the washing and care devices. Temperature sensors, humidity sensors, and liquid level sensors are arranged inside the washing and care chamber to monitor the temperature, humidity, and liquid level inside the washing and care chamber in real time. The washing and care devices control terminal includes a water pump driver, a heating element, a fan driver, a roller brush motor, and a bath liquid pump. The water pump driver is used for water pressure feedback, the heating element for water temperature feedback, the fan driver for air temperature / air speed feedback, the roller brush motor for speed feedback, and the bath liquid pump for flow rate feedback.

[0114] The large-scale intelligent terminal is deployed in the cloud or at the edge, and integrates a multimodal recognition network, including at least a breed classification branch, a body size estimation branch, a coat quality / length analysis branch, and a skin sensitivity inference branch. It is used to output washing and care parameters based on multi-angle image sequences of pets, including but not limited to water temperature range, water pressure level, detergent dosage, air temperature curve, and washing and care time; it is also equipped with a brushing structure that includes rubbing intensity.

[0115] The communication module is used to transmit data collected by the camera module and the device status acquisition module to the large model intelligent terminal. The communication module can be integrated into the main control board of the washing and care cabin as a WiFi / 5G module, or it can be an independent edge gateway box.

[0116] The execution control module receives the washing and care strategy from the large-scale intelligent terminal through the communication module, and generates washing and care instructions to send to various washing and care devices in the washing and care chamber for execution.

[0117] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A pet grooming method based on visual recognition and large-scale model linkage, characterized in that, The following washing and care steps are included: S1. Place the pet in the grooming chamber and start shooting according to the resting shooting frequency or the action shooting frequency after detecting changes in the pet's posture to obtain multi-angle time-series images of the pet. S2, the obtained pet time-series images are encoded and sent to the large model smart terminal along with the equipment status parameters of the grooming cabin; S3, the large model intelligent terminal analyzes pet breed, estimates body size, analyzes fur quality / length and infers skin condition based on pet time-series images, and generates washing and care strategies and washing and care times by combining the breed prior knowledge base and fur condition discrimination; the washing and care strategy includes independent control parameters for the pre-rinse stage, cleaning stage, rinsing stage and drying stage, and each stage parameter is set with a limited safety threshold. S4, send the generated washing and care strategy and washing and care time to the washing and care chamber; S5 analyzes the washing and care strategy and performs over-limit verification. After the verification is passed, it generates control commands and drives the corresponding spraying, drying and temperature control. At the same time, it periodically transmits the temperature and humidity inside the cabin and the pet's behavior feedback during the washing and care time. If abnormal stress is detected, it will automatically pause and adjust the parameters.

2. The pet grooming method based on visual recognition and large model linkage according to claim 1, characterized in that, The following start / stop operation is used when acquiring the pet time-series images: First, set the rest shooting frequency f. 静 and action shooting frequency f 动 Then, the threshold of dynamic changes in the pet during the filming was detected. and stress shutdown threshold The exercise intensity M is calculated once every fixed time interval. t ; When M t < When the pet is motionless, start filming at the resting filming frequency; when ≤M t < When the pet is moving, switch the action shooting frequency to start shooting; When M t ≥ If the test is repeated more than N times, it indicates that the pet is stressed, and all photography and grooming activities should be stopped immediately.

3. The pet grooming method based on visual recognition and large model linkage according to claim 2, characterized in that, The intensity of the movement M t The calculation includes the following steps: First, obtain the rectangular activity regions of the pet in the image at the current and previous sampling times; Calculate the area of ​​the rectangular active region at the current time and the area of ​​the rectangular active region at the previous sampling time, respectively; Calculate the absolute value of the difference between the two areas; Find the center points of the current rectangular active region and the rectangular active region at the previous sampling time, respectively; Calculate the distance between the two center points and divide it by the height of the current rectangular active area; The motion intensity M is obtained by adding the absolute value of the area difference and the calculated value of the center point distance divided by the rectangle height through weight allocation. t .

4. The pet grooming method based on visual recognition and large model linkage according to claim 2, characterized in that, The specific operation steps of S2 are as follows: S201, determine the sending trigger condition, and decide whether to send based on the shooting start and stop operation, and send the complete pet time sequence image and device status parameters or only the device status parameters; S202, Encode the pet time-series images that need to be sent; S203 packages the encoded pet time-series images and the status parameters of the grooming cabin equipment and sends them to the large model intelligent terminal.

5. The pet grooming method based on visual recognition and large model linkage according to claim 4, characterized in that, The triggering conditions include the following: When the shooting mode is in rest shooting mode and the last transmission is less than 30 seconds, only the pet time sequence image and device status parameters captured in rest shooting mode are sent. When the shooting mode is in silent shooting mode and more than 30 seconds have passed since the last transmission, only one image of the pet from any angle will be sent. When the shooting mode is motion shooting and the time since the last large model intelligent terminal inference has not exceeded the throttling window, the pet time sequence image will not be sent temporarily; When the shooting mode is motion shooting and the time since the last large model intelligent terminal inference exceeds the throttling window, send the complete pet time series image for the corresponding time period; When switching shooting frequencies, send complete time-series images of the pet for the corresponding time period; When the pet is in a state of stress, send the last complete set of pet time-series images and stop all grooming activities.

6. The pet grooming method based on visual recognition and large model linkage according to claim 1, characterized in that, S3 specifically includes the following operational steps: S301, the large model intelligent terminal receives and parses the encoded pet time-series image data and the status parameters of the pet grooming cabin equipment; S302 identifies pet breeds based on parsed pet time-series image data and combined with image classification models; estimates pet body length, shoulder height, chest circumference, and weight range through pet outline or limb characteristics; outputs coat quality classification by analyzing pet hair texture, density, curliness, and average length; and outputs abnormal locations and severity by detecting erythema, dandruff, eczema, parasites, or external injuries in skin areas. S303, query the breed prior knowledge base, retrieve the corresponding breed's skin sensitivity, suitable water temperature range, hair washing and care precautions and susceptible diseases based on the identified breed, compare the contents of the prior knowledge base with the actual analysis results, and correct the judgment of the pet's condition. S304 generates a washing and care strategy that includes a pre-rinse stage, a cleaning stage, a rinsing stage, and a drying stage. If shampoo is used in the cleaning stage, the water temperature in the rinsing stage is increased and the number of water changes is increased. If the pet has sensitive skin, the water temperature in all stages is lowered to the lower limit, and the speed of the roller brush is reduced.

7. The pet grooming method based on visual recognition and large model linkage according to claim 6, characterized in that, The pre-rinsing stage is set with water temperature of 25℃-40℃, water pressure of 50-400kPa, spray angle, and duration of ≤3min. The cleaning stage settings include the type and amount of bath liquid used; The rinsing stage is set with a water temperature of 30℃-42℃; The drying stage is set with an air temperature of 35℃-42℃, an air speed of 10-60m / s, and a total time of ≤30min.

8. The pet grooming method based on visual recognition and large model linkage according to claim 2, characterized in that, S5 specifically includes the following operation steps: S501, analyze the washing and care strategy and perform over-limit verification. Check whether the parameters of the pre-rinse stage, cleaning stage, rinsing stage and drying stage are within the preset safety threshold range. If any parameter exceeds the safety threshold, it is marked as invalid and an alarm is triggered. At the same time, the corresponding strategy is refused to be executed. S502, after the over-limit verification is passed, a control command is generated, which converts the parameters of each stage into drive signals and sends them to the corresponding execution equipment in the washing and care chamber. S503 collects temperature, humidity, and pet's current activity level (M) inside the bathing chamber every 1-2 seconds during the bathing / care time. t The data is transmitted to a large-scale intelligent terminal. If the analysis indicates pet stress or abnormal temperature or humidity, the bathing and grooming process will be stopped immediately.

9. A pet grooming system based on visual recognition and large-scale model linkage, used to execute the pet grooming method based on visual recognition and large-scale model linkage as described in any one of claims 1-8, characterized in that, This includes the washing and care cabin, camera module, equipment status acquisition module, communication module, large-scale intelligent terminal, and execution control module; The washing and care chamber is equipped with a variety of washing and care devices for performing washing and care operations; The camera module is installed on the inner wall of the pet grooming chamber and is used to continuously collect multi-angle image sequences of the pet according to a preset time interval or trigger signal. The equipment status acquisition module is deployed at the control terminal of the washing and care chamber and various washing and care equipment, and is used to collect the status parameters of the washing and care chamber and the washing and care equipment. The large-scale intelligent terminal is deployed in the cloud or at the edge. It integrates a multimodal recognition network and includes at least a breed classification branch, a body size estimation branch, a coat quality / length analysis branch, and a skin sensitivity inference branch. It is used to output washing and care parameters based on multi-angle image sequences of pets, including but not limited to water temperature range, water pressure level, detergent dosage, air temperature curve, and washing and care time. The communication module is used to transmit data collected by the camera module and the device status acquisition module to the large-scale intelligent terminal; The execution control module receives the washing and care strategy from the large-scale intelligent terminal through the communication module, and generates washing and care instructions to send to various washing and care devices in the washing and care chamber for execution.

10. The pet grooming system based on visual recognition and large model linkage according to claim 9, characterized in that, The various washing and care equipment mentioned above includes, but is not limited to, adjustable spray equipment, air drying equipment, temperature control equipment, and bath liquid application equipment.

Images