Method for monitoring feeding behavior of pet

Abstract

The disclosure relates to monitoring pet feeding behavior under control of at least one processor. An example method includes obtaining load data from a load sensor of a pet bowl while a pet interacts with contents of the pet bowl. The load sensor can have a sensitivity of + / -50 grams or more, and load data is acquired at a sampling rate of 10 to 150 samples per second. The example method can also include sequentially grouping the load data at a time increment of 0.01 to 5 seconds, wherein a single time increment includes a plurality of samples; and identifying a feeding behavior occurring within one or more of the time increments based on load data generated by interaction of the pet with the pet bowl or the pet bowl contents.

Description

[0001] Cross-reference to related applications

[0002] This application claims the benefit and priority of U.S. Provisional Application Serial No. 63 / 580554, filed September 5, 2023, the disclosure of which is incorporated herein by reference in its entirety. Background Technology

[0003] Pet food bowls are used by various pet owners or animal caregivers to provide their animals with food or water (or other liquids). Eating or drinking behavior can provide clues about healthy feeding habits, but in some cases, it can be a tool for detecting potential animal health problems. For example, some visual indicators associated with an animal's eating or drinking behavior can be used to provide information about the animal's health, including the onset of physical, behavioral, or mental health problems. Unfortunately, these visually noticeable symptoms may not become apparent until the middle to late stages of an illness or health problem and often do not provide sufficient information for appropriate intervention. Furthermore, animal caregivers (such as pet owners) often lack the behavioral knowledge to link eating or drinking behavior to health problems.

[0004] There are already some efforts to track animal eating and drinking behaviors, including the use of cameras, scales, etc. While these devices may help track some basic information, such as the amount of food or water consumed and the time of consumption, they often provide insufficient information to assess subtle changes in eating and / or drinking behaviors that may provide clues to an animal's health. Attached Figure Description

[0005] Figure 1 An example system for monitoring pet eating behavior according to this disclosure is illustrated schematically, which includes a smart pet food bowl.

[0006] Figure 2 An example smart pet food bowl according to this disclosure is shown, which includes a food bowl support and multiple food bowl liners.

[0007] Figure 3A and Figure 3B An example smart pet food bowl feeder according to this disclosure is shown.

[0008] Figure 4 This is a flowchart illustrating an example data collection and processing system and method for monitoring pet eating behavior according to this disclosure.

[0009] Figure 5 This is a flowchart illustrating an example data collection and processing system and method for monitoring pet drinking behavior according to this disclosure.

[0010] Figures 6 to 9This is a more specific flowchart illustrating an example artificial intelligence or machine learning system and method for monitoring pet eating behavior according to this disclosure.

[0011] Figure 10 This is a chart illustrating examples of the classification of eating activities according to this disclosure.

[0012] Figure 11 This is a flowchart illustrating an example health insight system and method according to this disclosure.

[0013] Figure 12 It is a chart showing examples of estimating, identifying, and filtering pre- and post-meal phases according to this disclosure.

[0014] Figures 13 to 15 It is a chart showing a feeding session with pet interaction and / or time increments as identified in accordance with this disclosure.

[0015] Figure 16 This is a chart illustrating the characteristics of example load data signals obtained from a pet licking food using a smart pet bowl, according to this disclosure.

[0016] Figure 17 This is a chart illustrating the characteristics of example load data signals obtained from a dog licking water using a smart pet food bowl, according to this disclosure.

[0017] Figure 18 It is a chart showing example data collected from dogs according to this disclosure, illustrating the licking segment, the licking segment, and touching the food bowl.

[0018] Figure 19 This is a flowchart illustrating an example system and method for processing data collected during animal-intelligent pet food bowl interactions using a repeating model with normalization logic, according to this disclosure. Detailed Implementation

[0019] This disclosure relates to animal health and behavior monitoring, and more specifically, to devices, systems, methods, and computer program products for determining, monitoring, processing, recording, and transmitting various physiological and behavioral parameters of animals via networks.

[0020] According to an example of this disclosure, a method for monitoring pet feeding behavior under the control of at least one processor may include: acquiring load data from a load sensor of a pet food bowl while the pet interacts with the contents of the pet food bowl; sequentially grouping the load data in time increments of 0.01 seconds to 5 seconds, wherein a single time increment contains multiple samples; and identifying feeding behavior occurring within one or more of the time increments based on the load data generated from the pet's interaction with the pet food bowl or the contents of the pet food bowl. The load sensor in this example may have a sensitivity of + / - 50 grams or finer, and the load data may be acquired at a sampling rate of 10 to 150 samples per second.

[0021] In another example, a non-transitory machine-readable storage medium embodies instructions. When executed, the instructions cause the processor to perform a method for monitoring a pet's feeding behavior. The method for monitoring feeding behavior may include: acquiring load data from a load sensor of the pet food bowl as the pet interacts with the contents of the food bowl; sequentially grouping the load data in time increments of 0.01 seconds to 5 seconds, where a single time increment contains multiple samples; and identifying feeding behavior occurring within one or more of the time increments based on the load data generated from the pet's interaction with the pet food bowl or the contents of the food bowl. The load sensor in this example may have a sensitivity of + / - 50 grams or finer, and the load data may be acquired at a sampling rate of 10 to 150 samples per second.

[0022] Regarding these and other related methods for monitoring pet eating behavior, and method steps related to instructions from non-transitory machine-readable storage media, additional steps may include: excluding load data when identifying eating behavior if it is determined that the load data is caused by human-computer interaction, accidental triggering, or unintentional interaction with the pet food bowl or its contents. More specifically, in some examples, eating behavior may be count-based eating behavior selected from licking, licking, or biting. In other examples, eating behavior may be duration-based behavior selected from pet touching the food bowl, moving the food bowl, sniffing food, pausing, eating, licking, licking, or biting. Regarding count-based eating behavior, these "counts" may be based on individual micro-events of eating behavior over a single time increment or over a period of time spanning multiple time increments. Individual micro-events may include, for example, a single lick, a single lick, or a single bite. In some examples, count-based eating behavior may be used to characterize a period of time spanning one or more time increments in which count-based behavior has not occurred. Regarding duration-based eating behavior, in some examples, these behaviors can be characterized by sequentially mapping the time increments in which the duration-based behavior occurred or did not occur. More specifically, the method may include notifying the pet's caregiver (e.g., pet owner or foster parent) about the eating behavior or changes in the pet's eating behavior. This notification may include warning the caregiver that the change in the pet's eating behavior may be related to potential health or behavioral problems. These methods may also include acquiring auxiliary data from auxiliary sensors associated with the pet's food bowl. Exemplary auxiliary sensors may include proximity sensors, cameras, microphones, accelerometers, gyroscopes, inertial measurement unit sensors, radar, or combinations thereof. In some more specific examples, parameters for a dog water bowl may include: a sensitivity of + / -4 grams or finer, 15 to 75 samples per second (sampling rate), and a time increment of at least 0.4 seconds; parameters for a dog food bowl may include: a sensitivity of + / -4 grams or finer, 15 to 75 samples per second (sampling rate), and a time increment of at least 0.3 seconds. The parameters for cat water bowls can be as follows: a sensitivity of + / - 2 grams or finer, a sampling rate of 15 to 75 samples per second, and a time increment of at least 0.4 seconds; the parameters for dog food bowls can be as follows: a sensitivity of + / - 2 grams or finer, a sampling rate of 15 to 75 samples per second, and a time increment of at least 0.3 seconds. In other examples, the method may include generating a feeding behavior model for the pet. The feeding behavior model may include identifying the pet's feeding behavior based on feeding frequency, characteristic feeding behaviors of the pet, or a combination thereof.

[0023] Based on the methods for monitoring pet eating behavior, non-transitory machine-readable storage media, and related examples of smart pet bowls and / or systems disclosed herein, various details related to these methods, storage media, instructions, smart pet bowls, and / or systems may overlap to some extent. For example, the processor and / or memory may be located on an onboard system of the smart pet bowl or remotely relative to the smart pet bowl, which may relate to various usage options for the system and / or method. For example, the sampling rate, continuous-time increment, or both may be controlled onboard by the smart pet bowl and / or by a client device via a computer network. In some examples, the smart pet bowl may be able to exclude load data from humans, accidental triggers, or unintended interactions with the smart pet bowl or its contents. More specifically, the sensitivity, sampling rate, and continuous-time increment may be established at levels sufficient to identify count-based eating behaviors (e.g., licking, biting, or chewing). In other examples, sensitivity, sampling rate, and continuous time increments can be established at levels sufficient to allow counting of individual micro-events of feeding behavior within a single time increment or over uninterrupted time periods spanning multiple time increments. Individual micro-events can include, for example, a single lick, a single nibble, or a single bite. More specifically, sensitivity, sampling rate, and continuous time increments can be established at levels sufficient to identify duration-based feeding behavior. Examples of duration-based feeding behavior can include a pet touching the food bowl, moving the food bowl, sniffing the food, pausing, eating, licking, nibbling, or combinations thereof. Sensitivity, sampling rate, and continuous time increments can also be established at levels sufficient to allow sequential mapping of time increments in which duration-based feeding behavior occurred or did not occur. In some examples, auxiliary sensors can be included, such as proximity sensors, cameras, microphones, accelerometers, gyroscopes, inertial measurement unit sensors, or combinations thereof. Dogs and cats are examples of pets from which this technology can be utilized.

[0024] Regarding the sensitivity of the load sensor, one example prototype has been found to operate effectively at a signal-to-noise ratio of 4 grams. Therefore, in this particular example, any signal generated by a load greater than 4 grams provides an effective tool for identifying pet feeding behavior. In other systems, signal-to-noise ratios as low as 0.1 grams or less, 0.25 grams or less, or 0.5 grams or less provide systems that are even more sensitive for identifying pet feeding behavior, particularly those behaviors that do not exert excessive force on the pet food bowl disclosed herein. Regarding further details on the sampling rate, while a range of 10 to 150 samples per second has been found to be effective for identifying pet feeding behavior, a moderate range of approximately 15 to approximately 75 samples per second provides sufficient and accurate information to properly characterize a variety of feeding behaviors.

[0025] In some examples, the smart pet food bowl can be a dog water bowl with a load sensor having a sensitivity of + / - 4 grams or less. For dog water bowls, the sampling rate can be 15 to 75 samples per second, with a time increment of at least about 0.4 seconds, or at least about 0.5 seconds. In some examples, the dog water bowl can have a sensitivity of + / - 0.25 grams or finer, with a sampling rate of 20 to 75 samples per second, or 35 to 65 samples per second. In other examples, the smart pet food bowl can be a dog food bowl with a load sensor having a sensitivity of + / - 4 grams or less. For dog food bowls, the sampling rate can be 15 to 75 samples per second, with a time increment of at least about 0.3 seconds, or at least about 0.333 seconds (or 1 / 3 of a second). In some examples, the dog water bowl can have a sensitivity of + / - 0.25 grams or finer, with a sampling rate of 20 to 75 samples per second, or 35 to 65 samples per second.

[0026] In other examples, the smart pet food bowl can be a cat water bowl with a load sensor having a sensitivity of + / - 2 grams or less. For cat water bowls, the sampling rate can be 15 to 75 samples per second, with a time increment of at least about 0.4 seconds, or at least about 0.5 seconds. In some examples, the cat water bowl can have a finer sensitivity of + / - 0.25 grams or more, with a sampling rate of 20 to 75 samples per second, or 35 to 65 samples per second. In other examples, the smart pet food bowl can be a cat food bowl with a load sensor having a sensitivity of + / - 2 grams or less. For cat food bowls, the sampling rate can be 15 to 75 samples per second, with a time increment of at least about 0.3 seconds, or at least about 0.333 seconds (or 1 / 3 of a second). In some examples, the cat water bowl can have a finer sensitivity of + / - 0.25 grams or more, with a sampling rate of 20 to 75 samples per second, or 35 to 65 samples per second.

[0027] More specifically, the method for monitoring pet eating behavior and / or the non-transitory machine-readable storage medium used for monitoring pet eating behavior can be implemented in a multi-pet household. For example, the method and / or the non-transitory machine-readable storage medium may include identifying the pet to be monitored when the pet is in a multi-pet household. As mentioned, the smart pet bowl may be equipped with one or more auxiliary sensors, such as a proximity sensor, camera, microphone, accelerometer, gyroscope, inertial measurement unit sensor, and / or radar. In other examples, load data during eating and / or drinking may be used to identify the pet, as multiple animals may have different eating behavior characteristics and may be distinguishable in some examples. Therefore, the smart pet bowl disclosed herein can provide personalized insights for individual animals, such as individual pets in a multi-pet household.

[0028] Additional features and advantages of the intelligent pet food bowl and system for monitoring pet eating behavior (e.g., eating and / or drinking behavior) disclosed in this invention are described in the following detailed description and will become apparent therefrom. Not all features and advantages described herein are included; in particular, many additional features and advantages will be apparent to those skilled in the art based on the accompanying drawings and description. Furthermore, it should be noted that the language used in this specification has been chosen primarily for readability and instructional purposes and does not limit the scope of the subject matter of this invention.

[0029] Based on the examples in this article, Figure 1 A smart pet food bowl 110 and a system 100 for monitoring animal eating behavior (e.g., eating and / or drinking) are illustrated by way of example. In some examples, monitoring animal eating behavior can be used for animal health monitoring.

[0030] The smart pet food bowl 110 can be, in particular, a single unit having a food bowl holder 112 and one or more load sensors 114A positioned to acquire load data regarding the interaction between the animal and / or person and the smart pet food bowl (or its contents). In the example shown, the food bowl holder is bowl-shaped, so pet food and / or water can be added directly into the bowl-shaped food bowl holder. However, it should be understood that the food bowl holder can have this shape, or even different shapes adapted to receive the food bowl liner 150, which is shaped to receive pet food and / or water. In this case, the food bowl holder is bowl-shaped. A separate food bowl liner is shown in dashed lines to indicate that such a liner may be included to interact with the load sensors of the food bowl holder. This could allow more specialized food bowls to be inserted into a common food bowl holder, as described below. Figure 2 Those shown in the image. Regardless of the method, the load sensors can be placed in a location that does not interfere with the animal's natural behavior. The load sensors can automatically detect various feeding behaviors of the animal as it eats.

[0031] More specifically, an enlarged example load sensor 114B is shown, with sampling circuitry that can be used to collect load data during animal and / or human interaction with the smart pet bowl. Other load sensors and / or circuitry can be used, and / or multiple load sensors can be used. Regardless of which load sensor(s) are used, the number of load sensors, the placement of the load sensors, etc., the load sensors and associated circuitry or logic can be configured to be sufficiently sensitive (sensitivity) and fast enough (sampling rate) to provide load data in relatively short time increments (0.1 seconds to 5 seconds) suitable for distinguishing between count-based feeding behavior and / or duration-based feeding behavior, and even, in some examples, individual micro-behaviors. The smart pet bowl can also be able to exclude load data from humans, accidental triggering, or unintentional interaction with the smart pet bowl or its contents.

[0032] Count-based feeding behaviors can include, for example, licking, licking, and / or biting. These count-based feeding behaviors can be counted using individual micro-events associated with feeding behaviors within a single time increment or over a continuous period of time spanning multiple time increments. Therefore, a single micro-event can include a single lick, a single lick, or a single bite. Duration-based feeding behaviors can include, for example, touching the food bowl, moving the food bowl, sniffing food, pausing, eating, licking, and / or licking, etc. These duration-based feeding behaviors can be established at a level sufficient to allow for sequential mapping of the time increments in which duration-based feeding behaviors occur or do not occur. Load sensitivity can be fixed based on the architecture of the load sensor 114B, or can be modified, for example, via selected circuitry. More specifically, the sampling rate, continuous time increments, or both can be controlled on-board by the smart pet food bowl. In other examples, the sampling rate, continuous time increments, or both can be controlled by a client device via a computer network.

[0033] Regarding the system 100 for monitoring pet eating behavior, the smart pet food bowl 110 can be used in the context of other networked computers and / or server systems via wireless or wired connections. For example, in addition to the smart pet food bowl, the system for monitoring pet eating behavior may also include a client device 130 and an analysis server system 120, each communicating via a network 140. In this example, the smart pet food bowl (which in some examples may include a bowl holder 112 and a bowl liner 150) can hold pet food or water (or other liquids). In some aspects, the analysis server system may be implemented using a single server. In other aspects, the analysis server system may be implemented using multiple servers. In still other examples, the client device may interact with and utilize the analysis server system, and vice versa.

[0034] Client device 130 may include, for example, a desktop computer, laptop computer, smartphone, tablet, and / or any other user interface suitable for communicating with the smart pet bowl. The client device may acquire various data from one or more smart pet bowls 110, provide data and insights about one or more animals via one or more software applications, and / or provide data and / or insights to the analytics server system 120, as described herein. The software applications may provide data on animal weight and behavior, track changes in that data over time, and / or provide health information about the animal, as described herein. In some embodiments, the software application acquires data from the analytics server system for processing and / or display.

[0035] The analytics server system 120 can acquire data from various client devices 130 and / or smart pet bowls 110, as described herein. The analytics server system can provide data and insights about one or more animals and / or transmit data and / or insights to the client devices. These insights may include, but are not limited to, insights about animal weight and behavior, changes in data over time, and / or health information about the animals, as described herein. In several embodiments, the analytics server system acquires data from multiple client devices and / or smart pet bowls, identifies animal groups in the acquired data based on one or more characteristics of the animals, and determines insights for these animal groups. Insights into the animal cohorts can be used to provide recommendations for specific animals that share characteristics with the cohort. In some examples, the analytics server system provides a portal (e.g., a website) for veterinarians to access information about specific animals.

[0036] The smart pet bowl can transmit data to client device 130 and / or analysis server system 120 for processing and / or analysis. In some examples, the smart pet bowl can communicate directly with a non-networked client device 115 without sending data over network 140. The term "non-networked" client device does not imply that the device is not connected to the cloud or other networks, but simply refers to the presence of a wireless or wired connection that can be established directly with the smart pet bowl. For example, the smart pet bowl and the non-networked client device can communicate via Bluetooth. In some examples, the smart pet bowl can directly process load sensor data. In other examples, the smart pet bowl can utilize load sensor data to determine whether the smart pet bowl is balanced (especially in the case of multiple load sensors), unbalanced, and / or properly calibrated. In some cases, one or more adjustable load sensors positioned to interact with the contents of the smart pet bowl can be automatically or manually adjusted. In other cases, this balance and / or calibration can be fixed, as the load sensors are typically positioned in locations that do not directly contact the ground and do not bear the weight of the bowl support and / or bowl liner (if included).

[0037] Figure 1 Any computing device shown (e.g., client device 130, analytics server system 120, smart pet food bowl 110, non-network client device 115, etc.) may include a single computing device, multiple computing devices, a cluster of computing devices, etc. A computing device may include one or more physical processors communicatively coupled to storage devices, input / output devices, etc. As used herein, a processor may also be referred to as a central processing unit (CPU). The client device may be accessed by an animal owner, veterinarian, or any other user.

[0038] Additionally, as used herein, a processor may include one or more means capable of executing instructions that encode arithmetic, logic, and / or I / O operations. In one illustrative example, the processor may implement a von Neumann architecture model and may include an arithmetic logic unit (ALU), a control unit, and / or multiple registers. In some aspects, the processor may be a single-core processor, which is typically capable of executing one instruction at a time (or processing a single instruction pipeline); and / or a multi-core processor, which can execute multiple instructions simultaneously. In some examples, the processor may be implemented as a single integrated circuit, two or more integrated circuits, and / or may be a component of a multi-chip module, wherein the individual microprocessor dies are included in a single integrated circuit package and thus share a single socket. As described herein, memory refers to volatile or non-volatile storage devices, such as RAM, ROM, EEPROM, or any other device capable of storing data. Input / output devices may include network devices (e.g., network adapters or any other components that connect a computer to a computer network), peripheral component interconnect (PCI) devices, storage devices, disk drives, audio or video adapters, photo / video cameras, printing devices, keyboards, displays, etc. In some respects, computing devices provide interfaces, such as APIs or web services, that provide some or all of the data to other computing devices for further processing. Access to the interface can be opened and / or secured using any of a variety of technologies, such as by using client authorization keys, depending on the specific application requirements of this disclosure.

[0039] Network 140 may include a LAN (Local Area Network), WAN (Wide Area Network), telephone network (e.g., Public Switched Telephone Network (PSTN)), Session Initiation Protocol (SIP) network, wireless network, point-to-point network, star network, token ring network, hub network, wireless networks (including protocols such as EDGE, 3G, 4G LTE, Wi-Fi, 5G, WiMAX, etc.), the Internet, etc. Various authorization and authentication technologies can be used to protect communication security, such as username / password, OAuth, Kerberos, SecureID, digital certificates, etc. It should be understood that the network connection shown in the example pet eating behavior monitoring system 100 is illustrative, and any means of establishing one or more communication links between computing devices can be used.

[0040] Systems for monitoring feeding behavior can, for example, automatically track feeding frequency, type of food (food and / or water), feeding behavior based on count-based and / or duration-based feeding interactions, changes in feeding behavior, etc. For example, by collecting historical information associated with the collected payload data, monitoring these or other parameters (or optionally combining multiple other sensor data and / or other biological data of the animal (e.g., age / life stage, sex, reproductive status, body condition, weight data, etc.) can be directly used to identify when changes occur that may be associated with underlying health or behavioral conditions affecting the animal.

[0041] Smart pet bowls and systems that monitor pet eating behavior can advantageously provide early indicators of potential health conditions, including but not limited to the animal's physical, behavioral, and mental health. Examples of physical health include, but are not limited to, kidney health, urinary tract health, metabolic health, and digestive health. More specifically, animal diseases that may be associated with behavioral eating data obtained from the use of smart pet bowls and related systems include, but are not limited to, diabetes, chronic kidney disease, and hyperthyroidism. Based on these potential health conditions, proactive notifications can be provided to pet owners and / or veterinarians for further diagnosis and treatment.

[0042] Further details regarding the notification may include the transmission of information about the animal, which may include notifications related to the animal's behavior that can be generated based on events and / or historical events categorized for the animal. In some embodiments, the notification may be generated based on events of other animals in the same cohort as the animal. The notification may indicate that an event has occurred and / or may indicate one or more inferences about the animal. For example, an animal's drinking behavior may be tracked over time, and if the activity increases over time or the number of events increases, a notification may be generated to alert the pet owner to the change in behavior and encourage the caregiver to take the pet to the veterinarian, as the behavior may be associated with diabetes, advanced kidney failure, or other kidney disease. In some embodiments, the notification is transmitted once a threshold amount of data and / or events has been determined. The notification may be transmitted to a client device associated with the animal's owner and / or the animal's veterinarian, as described herein. In several embodiments, the notification provides an indication requesting the user to confirm that the detected event is correct. In this way, the notification can be used to obtain baseline true labels for the events, which can be used to train and / or retrain one or more machine classifiers. Notifications can also be generated and / or transmitted based on specific animal performance events.

[0043] Regarding the operation of the smart pet bowl described herein, various user interfaces can be provided to ensure the proper installation, configuration, and use of the feeding behavior monitoring system. These user interfaces can provide instructions to the user, request information from the user, and / or provide insights into the behavior and potential problems of one or more animals. When setting up a system to monitor feeding behavior, the initialization and location of the smart pet bowl can be valuable in ensuring the accuracy of the collected load data. In some implementations, the smart pet bowl may operate better in an indoor, temperature-controlled environment away from direct sunlight. In other examples, the smart pet bowl should be placed at least one inch away from all walls or other obstacles, as insufficient space may cause the smart pet bowl to become stuck on obstacles, interfering with data or readings. Furthermore, the smart pet bowl can be kept at a sufficient distance from high-vibration objects (such as washing machines and dryers) or high-flow areas, as vibration can cause misreadings and / or inaccurate readings from the weight sensor. In some implementations, the smart pet bowl may function better on a smooth, level, hard surface, as soft or uneven surfaces may affect the accuracy of the load sensor, although it may still operate to provide usable data under such conditions. In many implementations, the smart pet food bowl can be introduced to the animal slowly to improve the process of its integration into the environment. For example, the smart pet food bowl can be placed in the same room as the litter box for a few days to allow the animal to get used to its presence. Once the animal has adapted to the smart pet food bowl, the volume or brightness of the bowl can be reduced to allow the animal to adjust to the subtle sounds and lights that may be present when the bowl is running, such as dispenser sounds, LED lights, etc.

[0044] In some implementations, multiple user interfaces can be used to configure the feeding behavior monitoring system. These user interfaces may include: a user interface for initiating the smart pet bowl setup process; a user interface for initiating the network setup process; a user interface for connecting to the smart pet bowl via Bluetooth during the setup process; a user interface for confirming the Bluetooth connection to the smart pet bowl during the setup process; a user interface for connecting the smart pet bowl to a local area network; a user interface indicating that the smart pet bowl is ready for use; a user interface for physically locating the smart pet bowl and litter box; and / or a user interface for confirming the completion of the setup process. In some examples, if multiple animals are to use the same smart pet bowl, profiles can be generated for multiple animals. These profiles can be used to establish baseline characteristics for each animal and track the animal's behavior and characteristics over time.

[0045] In some implementations, a user interface for creating an animal profile may be used. Examples of user interfaces for creating an animal profile include: a user interface for the start screen of the animal profile creation process, a user interface for the introduction screen of animal profile creation, a user interface for entering the animal's name, a user interface for entering the animal's sex, a user interface for entering the animal's reproductive status, a user interface for the introduction screen explaining the current physical condition of the captured animal, a user interface for examining a specific part of the animal's body, a user interface for examining the animal's outline, a user interface for examining the animal's waist, and / or a user interface for the end screen of the animal profile creation process.

[0046] Each animal is unique and exhibits unique behaviors. Systems that monitor feeding behavior can utilize multiple machine classifiers to track and differentiate between multiple animals without the need for additional collars or small devices, although additional auxiliary sensors can also be used in some examples. In some implementations, information can be solicited from the user regarding specific events, such as identifying which animal used the smart pet food bowl. This information can be used to confirm the identity of the animal associated with the specific event, which can then be used to retrain the machine classifier and improve the accuracy of future results. For example, regarding an animal's feeding behavior, the system can request confirmation of which animal is associated with the event to ensure the system continues to provide the best available insights. In some implementations, once the system has established a unique profile for a particular animal (e.g., after reaching a confirmation threshold), the frequency of future confirmation requests may decrease.

[0047] In some implementations, a user interface is used for expert advice notifications. The user interface may include: a user interface displaying notifications (instructing that the pet should be monitored due to changes in eating behavior), a user interface requesting confirmation that the smart pet food bowl is configured correctly, a user interface requesting additional information about the pet's weight, a user interface requesting additional information about the pet's appearance, a user interface requesting additional information about the pet's excretions, and / or a user interface providing guidance (contacting a veterinarian if changes in the pet's behavior or condition are concerning).

[0048] The systems and techniques for monitoring feeding behavior described herein offer several advantages over existing systems (though it should be noted that the systems and methods described herein can be used in conjunction with some of these existing monitoring systems in certain situations). Existing monitoring systems typically rely on microchips implanted in animals, RFID-enabled collars, and / or visual image recognition to identify individual animals. These systems can be highly invasive (e.g., requiring veterinary intervention to implant microchips at specific locations within the animal) and are prone to failure (e.g., microchips may migrate to other locations within the animal and become difficult to locate, RFID collars may wear out, be lost, and / or require frequent battery / charging replacements, etc.).

[0049] The system for monitoring pet feeding behavior according to this disclosure addresses some limitations of existing systems, particularly where some of these other systems interfere with normal animal behavior. The system for monitoring feeding behavior disclosed herein can, for example, identify and track animals without relying on external identification such as microchips or RFID collars. For example, the licking or licking characteristics of a first pet may be quite unique compared to a second pet, and the smart pet food bowl and / or system can be used to distinguish between multiple pets. For example, in some examples, the system for monitoring feeding behavior described herein can identify animals and their behavior without relying on image or video information, thereby avoiding the use of cameras or human observers that may influence typical animal behavior. Nevertheless, cameras and / or other sensors can be used in some cases to enhance the functionality of the smart pet food bowl, such as for animal identification, event segmentation, or behavior classification.

[0050] See now Figure 2 An example smart pet bowl 110 is shown, which has a bowl holder 112 and several different bowl liners 150A-150D as options for inserting the bowl holder to interact with a load sensor 114. The load sensor is positioned to transmit load data from interactions with the smart pet bowl (more specifically, interactions with the contents of the smart pet bowl, such as pet food and / or water). Thus, if a bowl liner is used, the bowl liner is configured to contact an area near the load sensor so that the load sensor picks up load and load changes occurring within the bowl liner. As described above, in some examples, the bowl holder may be a smart pet bowl (without a bowl liner), or a combination of the bowl holder and the bowl liner may be a smart pet bowl. In this case, both are possible because the bowl holder is bowl-shaped.

[0051] If the feed bowl liners 150A-150D are available, any of a variety of configurations can be used, making the feed bowl support 112 (or base unit) quite versatile in some cases. For example, different sizes of feed bowl supports may exist based on animal size and / or type, but subsequently, various feed bowl liners may be able to be used with more versatile feed bowl supports. Figure 2The four non-limiting examples of food bowl liners shown can each interact with a load sensor (and / or other sensors) when a particular food bowl liner is placed on a food bowl holder. Therefore, in this example, the bottom surface of the food bowl liner is positioned to transfer load during pet feeding interactions to detect and generate load data for monitoring and / or interpreting health issues. For example, food bowl liner 150A is shown as a deep, large-diameter dish, possibly suitable for providing animals with larger quantities of pet food and / or water (or other liquids). Food bowl liner 150B is a shallow dish, reducing the capacity for adding pet food and / or water, and may be more suitable for young pets, such as teenage or puppy dogs, or medium-sized cats. Food bowl liner 150C is even shallower and has a tapered diameter, possibly more suitable for puppies and / or kittens. Food bowl liner 150D is a larger dish with a maze-like construction at the bottom. This configuration may be suitable for slowing down the feeding process of pets that eat too quickly. Regardless of the type of food bowl liner used, load data can be collected and analyzed as described herein. In some examples, there may be an electrical connection between the food bowl holder and the food bowl liner, allowing the smart pet bowl to know what type of food bowl is placed on it, thus providing more information that can be used to monitor and / or analyze health issues. Alternatively, the smart pet bowl may include a location for the user to input the type of food bowl liner being used.

[0052] Further details regarding the smart pet food bowl 112 include, in addition to the load sensor 114 (and / or other sensors), the smart pet food bowl (or bowl holder) itself may include a processor 116 and a memory 118. The processor and memory may be able to control the load sensor and receive load data from it. The load data may be temporarily or permanently stored in the memory. A data communicator 117 may also be present in the smart pet food bowl and be able to transmit load data to another device. For example, the data communicator may be a wirelessly networked device employing an employee wireless protocol such as Bluetooth or Wi-Fi. The data communicator can send load data to a physical remote device capable of processing the load data, such as... Figure 1 The analysis server system 120. The data communicator can also transmit data via a wired connection and may employ a data port such as a Universal Serial Bus port. Alternatively, the memory slot is capable of accommodating a removable memory card on which payload data can be stored, and the removable memory card can then be physically removed and transferred to another device for uploading or analysis. In one embodiment, the processor and memory can be capable of analyzing payload data without sending the payload data to a physically remote device, such as the analysis server system.

[0053] The smart pet food bowl 110 may include a power source 119. The power source may be a battery, such as a replaceable or rechargeable battery. The power source may be a wired power source plugged into a wall outlet. The power source may be a combination of a battery and a wired power source. The smart pet food bowl may be constructed without a camera or image capture device and may not utilize auxiliary sensors such as proximity sensors, cameras, microphones, accelerometers, gyroscopes, inertial measurement unit sensors, etc. Similarly, the animal may or may not be equipped with wearable sensors, such as RFID collars. Systems for monitoring pet eating behavior may include tracking the animal's activity while using the smart pet food bowl based solely on load sensor data, or in some cases, by combining load sensor data with one or more of these auxiliary sensors. The collected data can then be processed to identify specific animal eating behaviors and, in some cases, can be correlated with the pet's health characteristics. Various events can be determined based on these characteristics and features. In some examples, various machine learning classifiers may be used to determine these events, as described in more detail herein. These events may include, but are not limited to, human-computer interaction, accidental triggering, unintentional interaction with the smart pet food bowl, and / or similar situations.

[0054] Figure 3A and Figure 3B The exhibition showcased an alternative smart pet food bowl 110, which, in addition to holding and Figure 1 and Figure 2 In addition to the same or similar components, an associated automatic feeder 160 may also be included. For example... Figure 3A As shown, the automatic feeder is integrated with a food bowl holder 112, which can be shaped as a pet food receiving container or can receive a separate food bowl liner (not shown). However, as Figure 3B As shown, the automatic feeder is modular. In this example, different food bowl holders or smart pet bowls can be coupled to the automatic feeder. In both examples, the smart pet bowl includes one or more load sensors 114 as described above. Notably, by combining the automatic feeder with the smart pet bowl, the automatic feeder can provide pet food on a schedule and / or can automatically modify pet feeding, such as time, quantity, type, etc., based on data learned from the animal's interaction with the smart pet bowl.

[0055] Now see the example shown and Figures 4 to 10The systems and methods specifically described herein illustrate, through examples, the identification of human-computer interaction, preprocessing (normalizing load values, cleaning data, identifying and / or pruning data, such as feeding interruptions or unreliable data), feeding stages, feeding behaviors, feeding sessions, features, etc., to aid in understanding this disclosure. For example, as previously described, a “feeding session” can conceptually be divided into one or more individual feeding behaviors that are “count-based” (licking, licking, or biting) and / or “duration-based” (touching the food bowl, moving the food bowl, smelling food, pausing, eating, licking, or licking) feeding or drinking behaviors. Notably, for example, licking, licking, or biting can be categorized by count and / or by duration. Furthermore, “features” can be developed in the load data that can be used to identify individual or grouped feeding behaviors based on count and / or duration. The load data can be analyzed in the time domain and / or frequency domain. Time-domain features can include, but are not limited to, mean, median, standard deviation, range, autocorrelation, etc. Frequency domain characteristics may include, for example, median, energy, power spectral density, etc. More specifically, load data can be analyzed as total load, individual load of each load sensor, and / or at the feeding behavior level, with load data separated into individual or group feeding behavior interactions via separation algorithms.

[0056] Temporal and / or frequency domain features can be created as input to a computer network or other system that acts as a machine classifier to categorize feeding behaviors within one or more feeding sessions. The machine classifier can be used to analyze payload data to identify and / or label feeding behaviors over a period of time, such as a 3-second lick followed by a 2-second lick, or in some examples, even to identify micro-events within time frames of feeding behaviors in the payload data, such as a single lick. Based on these labels, feeding behaviors (even individual animals) can be classified or categorized.

[0057] A variety of machine classifiers can be utilized, including but not limited to decision trees (e.g., random forests), k-nearest neighbors, support vector machines (SVMs), neural networks (NNs), recurrent neural networks (RNNs), convolutional neural networks (CNNs), probabilistic neural networks (PNNs), heuristics, regression, lightweight gradient boosting machines (GBMs), and / or similar methods. RNNs may also include (but are not limited to) fully recurrent networks, Hopfield networks, Boltzmann machines, self-organizing maps, learned vector quantization, simple recurrent networks, echo-state networks, long short-term memory networks, bidirectional RNNs, hierarchical RNNs, stochastic neural networks, and / or genetically scaled RNNs. In several implementations, combinations of machine classifiers can be utilized. Using more specific machine classifiers where available and general-purpose machine classifiers at other times can further improve prediction accuracy.

[0058] Further details regarding the “feeding session” can be conceptually categorized into multiple feeding behaviors (eating and / or drinking behaviors). For example, some feeding behaviors can be count-based, such as licking, licking, or biting; and / or duration-based, such as touching the food bowl, moving the food bowl, smelling food, pausing, biting / eating, licking, licking, or combinations thereof. Notably, for example, licking, licking, or biting can be categorized by count and / or duration. Using these feeding behaviors as an example, features can be developed in the load data for each feeding behavior phase to identify some or all of these specific feeding behaviors occurring by count or duration. The load data can be analyzed in the time domain or frequency domain, or both simultaneously. Time-domain features can include, but are not limited to, mean, median, standard deviation, range, autocorrelation, etc. Frequency-domain features can include, for example, median, energy, power spectral density, etc. In some implementations, the load data can be converted to both time-domain and frequency-domain data. For example, various techniques, such as Fourier transform, can be used to transform time-domain data into frequency-domain data. Similarly, various techniques, such as inverse Fourier transform, can be used to transform frequency-domain data into time-domain data. In some implementations, time-domain features and / or frequency-domain features can be identified based on specific peaks, valleys, and / or flat points within the time-domain and / or frequency-domain data as described herein. Furthermore, time-domain features and / or frequency-domain features can be developed for a single load sensor, individual load sensors in a group, and / or all load sensors. Thus, features can be developed to assist in classifying feeding behavior using machines, one or more machine classifiers, or modeling systems.

[0059] More specifically, additional features may include, but are not limited to: standard deviation of load, length of flat points, cross count of means, count of unique peaks, count of different load values, ratio of different load values ​​to event duration, count of maximum load variation in a single sensor, percentage of medium load bins, percentage of high load bins, high load bin volatility, high load bin variance, autocorrelation function hysteresis or delay, curvature, linearity, peak count, energy, minimum power, power standard deviation, maximum power, maximum variance shift, maximum Kulback-Leibler divergence, Kulback-Leibler divergence time, spectral density entropy, autocorrelation function derivative and / or variants of autoregressive models, etc. Therefore, feeding behavior can be classified based on its relevance to categorical features. For example, selected features can be used as input to a machine classifier to classify feeding behaviors, which can be count-based or duration-based behaviors as described above. Feeding behaviors may include labels indicating the type of behavior and / or confidence measures indicating the likelihood that the labels are correct. For example, unreliable or less reliable data may be reduced or removed. Therefore, for example, a machine classifier can be trained on a variety of training data indicative of animal feeding behaviors and on baseline real labels with features as input. Thus, the training data can be used to train the system, and the evaluation data can be used to evaluate the animal using the knowledge learned from the training data. For example, feeding behaviors can be classified based on a confidence metric that indicates the probability that one or more series of counts or durations have been correctly classified. For example, events can be classified as licking, sucking, sniffing the food bowl, touching the food bowl, pausing, and / or any of a variety of other feeding behaviors as described herein.

[0060] Temporal and / or frequency domain features can be created as input to a computer network or other system that acts as a machine classifier to categorize feeding behaviors within one or more feeding sessions. The machine classifier can be used to analyze payload data to identify and / or label feeding behaviors over a period of time, such as a 3-second lick followed by a 2-second lick, or in some examples, even to identify micro-events within time frames of feeding behaviors in the payload data, such as a single lick. Based on these labels, feeding behaviors (even individual animals) can be classified or categorized.

[0061] A variety of machine classifiers can be utilized, including but not limited to decision trees (e.g., random forests), k-nearest neighbors, support vector machines (SVMs), neural networks (NNs), recurrent neural networks (RNNs), artificial neural networks (ANNs), convolutional neural networks (CNNs), probabilistic neural networks (PNNs), heuristics, regression, lightweight gradient boosting machines (GBMs), and / or similar methods. RNNs may also include (but are not limited to) fully recurrent networks, Hopfield networks, Boltzmann machines, self-organizing maps, learned vector quantization, simple recurrent networks, echo-state networks, long short-term memory networks, bidirectional RNNs, hierarchical RNNs, stochastic neural networks, and / or genetically scaled RNNs. In several implementations, combinations of machine classifiers can be utilized. Using more specific machine classifiers where available and general-purpose machine classifiers at other times can further improve prediction accuracy.

[0062] See now for more details. Figure 4 This flowchart illustrates example data collection and processing methods 200 and related systems that can be implemented in monitoring pet eating behavior. For example, in monitoring eating behavior, steps may include activity classification modeling 230, feeding phase identification 240, and / or feeding repetition modeling 250, performed in any order. For example, activity classification modeling may take the form of building or scoring an activity classification model and may include aggregating data using rolling time windows, such as from about 0.01 seconds to about 5 seconds, from about 0.05 seconds to about 3 seconds, or from about 0.1 seconds to about 1 second. Processing may also include feeding phase identification. Feeding phase identification may include, for example, identifying the start and end points of a feeding session, calculating the food provided and the food remaining after the feeding session, and / or pruning the non-feeding phases of the feeding session. In some examples, processing may also include feeding repetition modeling, which may be based on building or scoring feeding (or drinking) repetition regression modeling. Feeding repetition modeling may, for example, require the provision of aggregated data, such as at the feeding session level.

[0063] When modeling pet eating behavior, several modeling factors can be considered to establish the types of load data features that may help characterize the eating behavior of a particular type of animal (e.g., cats, dogs, etc.). For example, when modeling animal eating behavior, many different animals of the same type are often used. “Real data” can be collected for comparison so that it can be correlated with load sensor data collected for modeling individual eating behavior as well as for the overall modeling of the smart pet food bowl disclosed herein. Real data can be collected in various ways, such as real-time observation, but video recording is good because technicians can pause, slow down, rewind, etc., while carefully considering eating behavior. In some cases, real data may include a combination of video recordings and sensor data once the sensor data is deemed reliable. Therefore, real data can be compared with “training data”, which is based on data collected simultaneously from the same multiple animals using load sensors. Thus, correlations can be established based on load sensor signals of pet eating behavior characteristics that correspond to observable pet eating behavior data collected from video recordings. Real-world data (from video tags) and training data (from load sensors and raw video) can be correlated to “train” a smart pet food bowl, allowing a model to be built. In other words, training can be performed to establish the correct association between load sensor data and raw video with specific feeding behaviors. For example, training data correlated with real-world video data can be used to establish load sensor features for feeding behaviors such as eating, licking, licking, food dropping, etc. Once these feeding behaviors are established through identifiable load sensor signals specific to that behavior, “test data” can be collected again using a set of the same types of animals (e.g., dogs, cats, etc.) to test various models that return results that accurately characterize feeding behaviors. If the real-world data is not aligned with or differs from the data collected using load sensors, that data can be cleaned from the overall dataset, for example, by removing it from the dataset. Similarly, when building an appropriate model, the video “real-world data” can be used again to verify that the model returns good feeding behavior results with sufficient accuracy for usefulness. As an example, any of the several models described elsewhere in this document can be used.

[0064] In some more detailed examples, data collection and processing 200 for monitoring pet feeding behavior may include performing additional steps that typically occur before monitoring feeding behavior. For example, before processing according to 230, 240, and / or 250 described above, human-computer interaction 210 with the smart pet food bowl may be identified, which may appear in the data as load signals unrelated to animal feeding, and removed from the feeding phase. More specifically, load signal data may be preprocessed 220 to further clean the data, thereby improving the reliability of the feeding phase data. For example, “real data” collected from video recordings may be compared to identify the time periods of human-computer interaction with the smart pet food bowl. When these types of load signals are detected, they may be removed from the dataset, for example, as signals associated with human-computer interaction (e.g., placing the food bowl, picking up the food bowl, pouring food into the food bowl, etc.). These types of interactions are typically very different in frequency and amplitude from those sensed during normal pet feeding activity. Preprocessing may include normalizing load values ​​and / or identifying and cleaning data from the dataset to be analyzed that is not particularly relevant to feeding behavior. This includes identifying and pruning irrelevant collected data, identifying interruptions in feeding sessions, and / or removing data that may be less reliable than any collected data. Similarly, when preprocessing data for training and / or testing smart pet food bowls, “real data” can be compared with training and / or test data to identify load sensor signals corresponding to usable load sensor frequency responses and less reliable load sensor frequency responses.

[0065] Modeling can be based on machine learning or artificial intelligence. Examples of how data can be cleaned to provide a more representative sample for understanding pet eating behavior are described in more detail in Example 1 below. Identifying human-computer interactions can include, for example, identifying pre-meal interactions or other anomalous interactions with the smart pet bowl or its contents, identifying post-meal interactions or other anomalous interactions with the smart pet bowl or its contents, and / or trimming these human-computer interactions or other anomalous interactions from the collected load data. Further details regarding the identification and trimming of human-computer interactions are described in more detail in Example 1 below.

[0066] It should be noted that, although the above Figure 4Other flowcharts are used with the systems and methods described below, but it should be understood that many other methods may also be used to perform the actions associated with these methods. For example, the order of certain boxes may be changed, certain boxes may be combined with other boxes, one or more boxes may be repeated, and / or some of the boxes described may be optional. The method may be executed by processing logic, which may include hardware (circuit, special-purpose logic, etc.), software, or a combination of both. The method or process may be implemented and executed according to instructions on a machine, wherein these instructions are contained on at least one computer-readable medium or a non-transitory machine-readable storage medium.

[0067] See now Figure 5 A more detailed flowchart of data collection and processing 300 for counting the number of times or "times" of drinking is shown as an example. It is worth noting that other organizations count the number of times of drinking, but this example provides an acceptable method that can be implemented. According to this example, raw data 310 is collected, and in some examples, some data preprocessing 320 may be performed, such as based on... Figure 4 For example, identifying human-computer interactions for removal. Preprocessing can also include identifying interruptions based on computed normalized load values ​​and using modeling or machine classifiers, where machine classifiers can include, but are not limited to, decision trees (e.g., random forests), k-nearest neighbors, support vector machines (SVMs), neural networks (NNs), recurrent neural networks (RNNs), artificial neural networks (ANNs), convolutional neural networks (CNNs), probabilistic neural networks (PNNs), heuristic methods, regression, lightweight gradient boosting machines (GBMs), and / or similar methods. For example, heuristic modeling can be used, which can include any form of computer-based problem-solving or discovery method that generates sufficient (even imperfect) results as approximate solutions when pursuing or searching for solutions to a problem. Heuristic methods can allow for accelerated discovery processes to achieve satisfactory results. Therefore, removing identified interruptions, even through rapid approximation, can generally improve the data collected for subsequent processing. In this example, after processing (or a portion thereof) is complete, data processing may include activity classification modeling 330 as an independent component, and licking duration and licking duration 350; and / or in other examples, data processing may include repetition modeling 340 and combined licking and licking repetition counts 360. The processed data may then be normalized by logic 370 to produce data related to the licking repetition counts 380 and licking repetition counts 390.

[0068] Further details regarding modeling, such as the activity categorization modeling and / or repetitive modeling described in the examples above, can also be used to establish categorizations of eating and / or drinking behaviors through other modeling protocols. Regardless of the type of modeling used, smart pet bowls can collect data based on captured different load characteristics, thereby providing information about what type of activity occurs at a given time during a feeding session. This enables the monitoring of feeding behavior and, in some cases, can generate insights related to the pet's health and / or well-being.

[0069] Examples of feeding behaviors that can be captured can be categorized into several broad categories, such as animal interaction, repetition (number of times), consumption, and / or interruption. Table 1 shows some non-restricted examples of behaviors that can be captured using payload data collected from a smart pet food bowl during a feeding session, as follows.

[0070]

[0071] Figures 6 to 9 Several example artificial intelligence (AI) processing protocols are demonstrated that can be used for feeding and / or drinking in multiple types of animals. In these examples, initial setup data input by animal caregivers or pet owners can be combined with more continuous data collected by smart pet food bowls or systems that monitor pet feeding behavior. By definition, Figures 6 to 9 This demonstrates various types of data that can be collected and processed by computers or computer networks. Other types of data collection and / or processing locations can be implemented in ways different from those depicted in these examples. For example, such as Figures 6 to 9 As shown, the types of data collected can include data about an animal feeding session (labeled "s" in the diagram), interactions between the animal and the smart pet bowl (labeled "i"), time increments or time windows (labeled "t"), or a combination of interactions / time increments (labeled "i / t"). More specifically, for any process shown, computation or data processing can occur anywhere, such as, for example, at the smart pet bowl, at a local computer, at a local client device or smartphone (or tablet), at any location suitable for edge computing (closer to the data source than a centralized server or cloud-based location), at a centralized server, or in the cloud. In many cases, the "cloud" symbol is used illustratively to indicate that the processing in this example is cloud-based, but can occur anywhere. Regarding computation or processing in general, Figures 6 to 9 Various computing methodologies are also illustrated, including cloud computing (labeled "c") and / or edge computing (labeled "e").

[0072] See more details Figure 6The system 400 (or smart pet bowl) for monitoring pet eating behavior can be configured via user initial setup 410, a process that includes general user settings. General settings may require input of animal type, breed, weight, physical condition score, reproductive status, age, and / or other information that may be relevant to the collection and processing of feeding-related data. If the smart bowl is for pet food or water, the type used can also be input by the user, or alternatively, the smart bowl can sense the type of food or water (or other liquid) that may be present. If the smart bowl is used for food handling 430, the device registers the event category, and in this example, processing may be used to determine whether the smart bowl's load sensor is triggered by a human or pet event. If the smart bowl is used for water handling 440, the device registers the event category, and in this example, processing may be used to determine whether the smart bowl's load sensor is triggered by a human or pet event. More detailed artificial intelligence and other computer processing are described below. Figure 8 and Figure 9 Further details will be provided later.

[0073] like Figure 7 As shown, by way of example, a smart pet food bowl can be set up for enhanced cat food handling 450. It is worth noting that, as previously mentioned, cloud computing (c) and edge computing (e) are shown only by way of example. This process can be used to replace... Figure 6The pet food bowl processing shown in Figure 430. In this enhanced process (or system), in addition to the load sensors shown and described above, auxiliary sensors may be used for pet detection, such as by scanning or otherwise capturing images and / or a pet face detector. A food type may be selected to fill the bowl, or the bowl may be filled directly without specifying the food type, for example, when only one food type is available. Similarly, human events can be distinguished from pet events, but in this example, the system or method may utilize cat identification (Cat ID), using captured images to match the face of a specific cat as a prerequisite for registering the pet event to that specific cat. In some examples, the smart pet food bowl of this disclosure can provide personalized insights for individual animals (e.g., individual pets in a multi-pet household). Therefore, the smart pet food bowl may be equipped with one or more auxiliary sensors, such as proximity sensors, cameras, microphones, accelerometers, gyroscopes, inertial measurement unit sensors, and / or radar, etc. In other examples, load data during eating and / or drinking may be used to identify the pet, as multiple animals may have different eating behavior characteristics that may be distinguishable in some examples. In some more specific examples, the identification of a cat or other animal can be accomplished by using image recognition (embedding from an image) using a camera, and / or the identification of a dog or other animal can be accomplished by using load sensor data through modeling that closely approximates real-world data collected visually for verification.

[0074] Figure 8 and Figure 9 Additional details were presented regarding AI and / or other protocols that could be used to handle human and / or pet incidents. For example, such as... Figure 8As shown, after the smart pet bowl is set up and the device registers relevant information 510 (e.g., cat or dog, pet food details, animal weight, animal breed, physical condition score, breeding status, age, etc., input by the user or detected by the smart pet bowl or peripheral device), the system 500 (or smart pet bowl) for monitoring pet eating behavior can process human events 520 and pet events 530 separately. Human events detailed in this example can include filling the smart pet bowl, cleaning the smart pet bowl, moving the smart pet bowl (e.g., picking it up), etc. On the other hand, pet events are more fundamental to the invention disclosed herein, as human events are primarily identified for trimming or removing from the collected data to more accurately understand the animal's eating behavior. Pet events that may be related to a dog eating pet food can be categorized using the payload data collected and analyzed as described herein. More specifically, example eating activity categorizations can include eating, licking, sucking, sniffing food (moving food with the nose), dropping food, touching the food bowl, and / or pausing eating. For cats that are eating, eating activities can be categorized as, for example, eating, licking, sniffing food (moving food with the nose), dropping food, scooping food, touching the food bowl or scratching at food, and / or pausing to eat. Eating can be further characterized by, for example, the amount of food consumed, the rate of eating, interaction with food, and / or the duration of eating. When collecting such data, eating activities can be processed, for example, based on the count of activity repetitions and / or the duration of the event.

[0075] like Figure 9 As shown, similarly, after the smart pet bowl is set up and the device registers relevant information 610 (e.g., cat or dog, pet food details, animal weight, animal breed, etc., input by the user or detected by the smart pet bowl or peripheral device), the system 600 (or smart pet bowl) for monitoring pet eating behavior can process human events 620 and pet events 630 separately. Human events detailed in this example can include filling the smart pet bowl, cleaning the smart pet bowl, moving the smart pet bowl (e.g., picking it up), etc. On the other hand, pet events are more fundamental to the invention disclosed herein, as human events are primarily identified for trimming or removing from the collected data to more accurately understand the animal's eating behavior. Pet events that may be related to a dog drinking pet food can be categorized using the payload data collected and analyzed as described herein. More specifically, example drinking activity categorization can include licking, licking, pausing drinking, and / or any other behavior associated with drinking. For cats, drinking activity categorization may also include, for example, licking, licking, pausing while drinking, and / or other behaviors identified that may be related to a cat's drinking. When collecting such data, drinking activities may be processed based, for example, on the count of activity repetitions and / or the duration of the event.

[0076] Figure 10 Examples of feeding activity classifications are shown, which can include identifying activities related to food spillage, smelling food, removing contents (eating), touching the food bowl, and / or any other feeding-related activities. More specifically, Figure 10 The dataset includes collected and categorized data demonstrating how dogs interact with a smart pet food bowl (based on load sensor data), using continuous time increments of 0.3 seconds, such as a 0.3-second rolling time window. In other words, one or more activities are identified and characterized every 0.3 seconds. To simplify the illustration of this example, three possible activities were identified: pausing feeding (p), removing contents (r), and touching the food bowl (t). The dataset does not show any removal of human-computer interaction, which may be indicated by the large load fluctuations shown on the left side of the graph.

[0077] Figure 11 Example systems or methods 700 are demonstrated that can be used to generate health insights, which can be obtained by collecting feeding data using the smart pet food bowl or related systems disclosed herein. In this example, multiple logic systems or integrated logic can be combined to process data related to feeding events 750 and data collected across various time ranges 760. For example, feeding events can be registered based on raw classifications. These feeding events can be registered, for example, using artificial intelligence and / or machine learning (AI / ML) 710. Machine classifiers or other systems can be used to model feeding, drinking, counting repetitions, duration, etc. Example machine classifiers may include decision trees (e.g., random forests), k-nearest neighbors, support vector machines (SVMs), neural networks (NNs), recurrent neural networks (RNNs), convolutional neural networks (CNNs), probabilistic neural networks (PNNs), heuristics, regression, lightweight gradient boosting machines (GBMs), and / or similar methods.

[0078] Normalization logic 720 processes the raw classifications to generate predictions. In some examples, user input and / or labels can be considered, and the predictions of the normalization logic can be overridden, if appropriate. Regarding data collected across various time ranges, the system or method can utilize aggregation logic 730. Through aggregation logic, the normalized predictions are processed to generate aggregated results. In some examples, the aggregated results can then be used in conjunction with certain decision tree logic 740. This decision tree logic can, for example, alert animal caregivers or pet owners to relevant information or insights based on feeding behaviors. In one example, the decision tree logic can be used as part of a decision tree logic system that can be used with other logic systems to transform data into meaningful insights, allowing pets to express their health and well-being and giving pet owners confidence in the decisions made on behalf of their pets.

[0079] Examples of information that can be obtained by using data collected through the smart pet bowl and / or system disclosed herein may include event information or insights, daily information or insights, weekly information or insights, monthly information or insights, yearly information or insights, etc. For example, event information that can be provided to pet owners may include the time spent on each activity, the number of times each activity was repeated, the number of food or water interactions, the number of grams consumed per bite, the total number of grams consumed, etc. Examples of daily information may include feeding times throughout the day, total consumption on a given day, average feeding duration, average feeding rate, average number of events, etc. Monthly information may include feeding trends, total consumption, average feeding duration, average feeding rate, average number of events per day, etc.

[0080] Health and behavioral insights that may be associated with such data changes include food or taste preferences, stress and anxiety levels, acute or chronic illnesses, dental problems, and more. By collecting, identifying, and reporting these types of information and insights, pet owners can be assured that they are taking good care of their pets. As an example of how this can work in some situations, pets will use food bowls to develop feeding patterns associated with their eating behavior. Changes in these feeding patterns can be detected and the pet owner notified. Upon receiving the notification, the pet owner may make adjustments that are in the best interests of their pet.

[0081] It should be understood that all the disclosed methods and processes described herein can be implemented using one or more computer programs, components, and / or program modules. These components may be provided as a series of computer instructions on any conventional computer-readable or machine-readable medium, including volatile or non-volatile memories such as RAM, ROM, flash memory, magnetic or optical disks, optical storage, or other storage media. The instructions may be provided as software or firmware and / or may be implemented, wholly or partially, in hardware components such as ASICs, FPGAs, DSPs, or any other similar devices. The instructions may be configured to be executed by one or more processors, which, in executing the series of computer instructions, perform or facilitate the execution of all or part of the disclosed methods and processes. As those skilled in the art will appreciate, the functionality of the program modules may be combined or distributed as needed across various aspects of this disclosure.

[0082] Based on the disclosure herein, the following examples illustrate several embodiments of the present invention.

[0083] 1. An example method for monitoring pet eating behavior under the control of at least one processor, the method comprising:

[0084] As the pet interacts with the contents of the pet food bowl, load data is acquired from a load sensor in the pet food bowl, wherein the load sensor has a sensitivity of + / - 50 grams or finer, and the load data is acquired at a sampling rate of 10 to 150 samples per second.

[0085] The load data is sequentially grouped in time increments ranging from 0.01 seconds to 5 seconds, where a single time increment contains multiple samples; and

[0086] Based on the load data generated from the interaction between the pet and the pet food bowl or the contents of the pet food bowl, feeding behavior occurring within one or more of the time increments is identified.

[0087] 2. The method according to Example 1, the method further includes: if it is determined that the load data is caused by human-computer interaction, accidental triggering or accidental interaction with the pet food bowl or the contents of the pet food bowl, then the load data is excluded when identifying the feeding behavior.

[0088] 3. The method according to any one of Examples 1 to 2, wherein the feeding behavior is a count-based feeding behavior selected from licking, licking or biting.

[0089] 4. The method according to Example 3, the method further comprising: characterizing the count-based eating behavior by counting individual micro-events of the eating behavior over a single time increment or over a time period spanning multiple time increments, wherein the individual micro-events include individual licking, individual licking, or individual biting.

[0090] 5. The method according to Example 4, wherein characterizing the count-based eating behavior includes: characterizing a time period spanning one or more time increments in which the count-based behavior has not occurred.

[0091] 6. The method according to any one of Examples 1 to 5, wherein the feeding behavior is a duration-based behavior selected from the pet touching the food bowl, moving the food bowl, smelling the food, pausing, eating, licking, licking or biting.

[0092] 7. According to the method of Example 6, the method further includes characterizing the duration-based eating behavior by sequentially mapping the time increments in which the duration-based behavior occurs or does not occur.

[0093] 8. The method according to any one of Examples 1 to 7, the method further comprising notifying the pet's caregiver about the feeding behavior or changes in the pet's feeding behavior.

[0094] 9. The method according to Example 8, wherein notifying the caregiver of the pet includes: warning the caregiver that the change in the pet's eating behavior may be related to potential health or behavioral problems.

[0095] 10. The method according to any one of Examples 1 to 9, further comprising: acquiring auxiliary data from auxiliary sensors associated with the pet food bowl, wherein the auxiliary sensors include a proximity sensor, a camera, a microphone, an accelerometer, a gyroscope, an inertial measurement unit sensor, or a combination thereof.

[0096] 11. The method according to any one of Examples 1 to 10, wherein the pet food bowl is a dog water bowl, and wherein the load sensor has a sensitivity of + / -4 grams or finer, the sampling rate is 15 to 75 samples per second, and the time increment is at least about 0.4 seconds.

[0097] 12. The method according to any one of Examples 1 to 10, wherein the pet food bowl is a dog food bowl, and wherein the load sensor has a sensitivity of + / -4 grams or finer, the sampling rate is 15 to 75 samples per second, and the time increment is at least about 0.3 seconds.

[0098] 13. The method according to any one of Examples 1 to 10, wherein the pet food bowl is a cat water bowl, and wherein the load sensor has a sensitivity of + / -2 grams or finer, the sampling rate is 15 to 75 samples per second, and the time increment is at least about 0.4 seconds.

[0099] 14. The method according to any one of Examples 1 to 10, wherein the pet food bowl is a cat food bowl, and wherein the load sensor has a sensitivity of + / -2 grams or finer, the sampling rate is 15 to 75 samples per second, and the time increment is at least about 0.3 seconds.

[0100] 15. The method according to any one of Examples 1 to 10, the method further comprising generating a feeding behavior model for the pet, including identifying the feeding behavior of the pet based on feeding frequency, pet feeding characteristic behaviors, or a combination thereof.

[0101] 16. The method according to any one of Examples 1 to 15, the method further comprising: identifying the pets in a multi-pet household.

[0102] 17. An example non-transitory machine-readable storage medium having instructions thereon, which, when executed, cause a processor to perform a method for monitoring a pet's eating behavior, the method comprising:

[0103] As the pet interacts with the contents of the pet food bowl, load data is acquired from a load sensor in the pet food bowl, wherein the load sensor has a sensitivity of + / - 50 grams or finer, and the load data is acquired at a sampling rate of 10 to 150 samples per second.

[0104] The load data is sequentially grouped in time increments ranging from 0.01 seconds to 5 seconds, where a single time increment contains multiple samples; and

[0105] Based on the load data generated from the interaction between the pet and the pet food bowl or the contents of the pet food bowl, feeding behavior occurring within one or more of the time increments is identified.

[0106] 18. The non-transitory machine-readable storage medium according to Example 17, wherein monitoring the pet's eating behavior includes monitoring count-based eating behavior selected from licking, licking, or biting.

[0107] 19. The non-transitory machine-readable storage medium according to any one of Examples 17 to 18, wherein monitoring the pet's eating behavior includes monitoring duration-based behavior selected from touching the food bowl, moving the food bowl, smelling food, pausing, eating, licking, or combinations thereof.

[0108] 20. The nontransitory machine-readable storage medium according to any one of Examples 17 to 19, wherein monitoring the pet's eating behavior includes notifying the pet's caregiver of the following: the eating behavior, changes in the eating behavior, potential health problems associated with the eating behavior, or a combination thereof.

[0109] 21. The non-transient machine-readable storage medium according to any one of Examples 17 to 20, wherein the load sensor has a sensitivity of + / -4 grams or finer, the sampling rate is 15 to 75 samples per second, and the time increment is at least about 0.3 seconds.

[0110] 22. The non-transitory machine-readable storage medium according to any one of Examples 17 to 21, wherein monitoring the feeding behavior of the pet includes: identifying the pet to be monitored when the pet comes from a multi-pet household.

[0111] definition

[0112] As used herein, “about,” “approximately,” and “substantially” should be understood to refer to numbers within a range, such as -10% to +10% of a reference number, -5% to +5% of a reference number, -1% to +1% of a reference number, or -0.1% to +0.1% of a reference number. All numerical ranges herein should be understood to include all integers or fractions within that range. Furthermore, these numerical ranges should be understood to support claims that involve any number or subset of numbers within that range. For example, disclosures of 1 to 10 should be understood to support ranges of 1 to 8, 3 to 7, 1 to 9, 3.6 to 4.6, 3.5 to 9.9, etc.

[0113] As used in this disclosure and the appended claims, the singular forms “a,” “an,” and “the” include plural references unless the context clearly specifies otherwise. Thus, for example, references to “a component” or “the component” include two or more components.

[0114] The term "including / comprises" will be interpreted as inclusive rather than exclusive. Similarly, the terms "including / comprises" and "or" should be considered inclusive unless the context explicitly prohibits this interpretation. Therefore, the disclosure of embodiments using the term "including / comprises" includes both embodiments that "consist substantially of the specified components" and embodiments that "consist of the specified components".

[0115] The term “and / or” used in the context of “X and / or Y” should be interpreted as “X” or “Y” or “X and Y”. Similarly, “at least one of X or Y” should be interpreted as “X” or “Y” or “X and Y”.

[0116] In the context of this document, the terms “example” and “such as” (especially when followed by a list of terms) are exemplary and illustrative only and should not be considered exclusive or comprehensive.

[0117] The terms "pet" and "animal" are used synonymously herein and refer to any animal capable of using the intelligent pet food bowl or related system of this disclosure, including, but not limited to, cats, dogs, rats, ferrets, hamsters, rabbits, iguanas, pigs, or birds. A pet can be any suitable animal, and this disclosure is not limited to any particular pet animal. The intelligent pet food bowl may be specifically designed for a particular type of animal (e.g., a dog or cat) or may not be specifically designed for it; and / or may be specifically designed for a particular feeding activity (e.g., liquids (water, milk, etc.) or food (dry food, wet food, or semi-wet food, etc.) or may not be specifically designed for it.

[0118] As used herein, ranges are abbreviated to avoid having to list and describe every value within the range. Where appropriate, any suitable value within the range may be chosen as the upper limit, lower limit, or endpoint of the range, and should therefore be flexibly interpreted to include not only the values ​​explicitly listed as range boundaries, but also any individual values ​​or subranges encompassed within the range, as if both values ​​and subranges were explicitly listed. For example, the numerical range of “about 1% to about 5%” should be interpreted to include not only the explicitly listed values ​​of about 1% to about 5%, but also individual values ​​and subranges indicating the range. Thus, included within this numerical range are individual values ​​such as 2, 3.5, and 4; and subranges such as 1-3, 2-4, and 3-5, etc. The same principle applies to listing a range of values. Furthermore, this interpretation should apply regardless of the breadth of the range or characteristic being described.

[0119] The terms “example” or “implementation”, especially when followed by a list of terms, are merely exemplary and illustrative and should not be considered exclusive or comprehensive.

[0120] The methods, storage media, smart pet bowls, and systems disclosed herein are not limited to the specific methodologies, schemes, reagents, etc., described herein, as these may vary, as will be understood by those skilled in the art. Furthermore, the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit, nor will it limit, the scope of disclosure herein or the protection afforded by the claims.

[0121] Unless otherwise defined, all technical terms, specialized terms, and acronyms used herein have the meanings commonly understood by one of ordinary skill in the art to which this invention pertains or in the art of using these terms. While any composition, method, article, or other means or material similar to or equivalent to those described herein may be used in the practice of this invention, certain compositions, methods, articles, or other means or materials are described herein.

[0122] As used herein, for convenience, multiple elements, constituent components, and / or materials may exist in a common list. However, these lists should be understood as individual members of the list being individually identified as separate and distinct members. Therefore, without indication to the contrary, this should not be used to treat any individual member of such a list as a factual equivalent of any other member in the same list solely based on the circumstances in the common group.

[0123] Example

[0124] The features of this disclosure can be further illustrated by the following embodiments, but it should be understood that these embodiments are included herein for illustrative purposes only and are not intended to limit the scope of the invention, unless otherwise specifically stated.

[0125] Example 1 –General Animal Feeding Behavior Modeling Scheme

[0126] The following examples are based on several studies primarily using dogs, which correlate load sensor data with specific feeding behaviors by collecting training data, and then using this training data in additional studies to build acceptable models. In these studies, video footage was used as real-world data for comparison, and the training and test data were correlated to assess accuracy. Several additional studies using cats were also conducted to validate that dog feeding behavior modeling can be applied to cats in a similar manner. In some examples, data processing includes identifying feeding sessions (including feeding phases involving pet interaction with the food bowl (excluding human-computer interaction)), segmenting the feeding session or phase into time increments, building features based on the time increments, and using these features as input to a machine learning model, with real-world video data from that time increment as output. For classification models (category-level output), recall, precision, and / or F1 score are typically used to select the best model. For continuous outcome models, MAE, MAPE, RMSE, and R-squared are used. 2 To determine the optimal model.

[0127] Example 2 – Identifying human-computer interaction

[0128] Figure 12This disclosure demonstrates how such data can be identified and pruned from animal feeding data collected from a smart pet food bowl equipped with one or more load sensors. Possible pre- and post-feeding human-computer interactions can initially be determined based on a percentage of the full duration of the collected dataset. For illustration, in this embodiment, 40% of the feeding session's beginning duration and 20% of its end duration are used. To more carefully identify the start and end of actual feeding sessions and prune human-computer interactions, the moving variance of the collected load values ​​over fixed time values ​​(e.g., 1 second, 2 seconds, 4 seconds, etc.) can be used. In other words, possible pre- and post-feeding phases can be examined stepwise within fixed time frames, and peaks that are several times larger than the average peak occurring during feeding phases not identified as pre- or post-feeding phases can be identified. If a peak in one or more pre- or post-feeding phases is at least 3 times (or at least 5 times, etc.) in magnitude relative to a standard peak occurring during an estimated portion of the feeding phase (or a standard peak of similar size found in possible pre- or post-feeding phases), the signal may indicate a human-computer interaction. Human-computer interactions (HCIs) that most closely approximate the estimated feeding stage can be used as points to prune from the available dataset, leaving only the identified feeding stages that do not contain HCIs. This same or similar process can be used to remove other types of data that were not useful when collecting feeding behavior data.

[0129] Example 3 -Data preprocessing and cleaning

[0130] In addition to removing human-computer interactions that can be identified and removed from the dataset, load sensor data can also be preprocessed and cleaned using modeling logic. For example, real video data is compared with training and test data from the dog population to identify misaligned video data with load sensor data for cleanup from the dataset. More specifically, the collected load sensor data is cleaned to further clarify the relevant load sensor signals associated with various feeding behaviors, such as eating, food dropping, licking, and sniffing food. Based on load thresholds associated with the evaluated animal population, these feeding behaviors are used to collect micro-events (or individual loads applied during the duration of micro-events) of the load sensors during the behavior. For the dogs in this study, the threshold was approximately 1 gram or greater. In this scenario, the logic was programmed to ignore any action if the micro-event was 1 gram or less and met other criteria (indicating poor correlation between the load sensor data and the real data collected from the video).

[0131] Furthermore, preprocessing of the load sensor data may include identifying the start and end points of a complete feeding session (or the start and end points of smaller feeding phases within a complete feeding session). Identifying the start and / or end points of feeding phases involves recognizing human-computer interaction from the session and predicting feeding behaviors (or actions), mapping animal interactions at the timestamp level such that the first and / or last timestamp predicted as eating (or drinking) can be considered as the start and / or end points of feeding, respectively. It is worth noting that these embodiments only show a small portion of a complete feeding session for illustrative purposes, as feeding typically lasts much longer.

[0132] In this embodiment, to determine the starting point, food is provided in a smart pet bowl, and an average baseline load value for a full bowl is established before feeding begins to determine the weight or quality of the food provided. Once the animal begins eating, a changing load signal is collected, and the starting point is determined using this load sensor data. In some cases, the animal being evaluated begins eating immediately after being provided with pet food, which does not provide sufficient time to establish an average baseline load value. In such cases, the weight or quality of the food provided is calculated using other data collected throughout the session, including data collected during feeding interruptions, consumption rates, time increments during shorter feeding phases, etc.

[0133] In this embodiment, to determine the stopping point, one of two scenarios occurs: the food is completely consumed or the animal stops eating before the food is completely consumed. The food is considered completely consumed when the food bowl's load matches the known load before the food was added (e.g., 0 grams of food remaining). Alternatively, when there is still food remaining in the smart pet's food bowl, a calculation is used based on the minimum possible value within a fixed time period at the end of the session (e.g., the last second, the last three seconds, the last five seconds, etc.).

[0134] Example 4 -Eating conversation interactions and time increments

[0135] Figures 13 to 15 Each shows a portion of a single eating session. Figure 13 Boxes are used to display various discrete interaction groups (each group including multiple micro-events, such as individual licking, individual licking, etc.) to show each interaction between the animal and the smart pet bowl or the pet food contained therein based on load data. Figure 14 The time increments that can be used to bundle data are shown by dashed lines. In this embodiment, the time increments are bundled in increments of 0.075 seconds. Figure 15Discrete interaction groups are combined, and these groups are decomposed to include relevant time increments occurring during the discrete interactions. Data segmentation based on the above embodiments and feature construction on the resulting time series improves model performance. Different combinations of session / interaction / time windows are selected for different models to optimize model performance.

[0136] Example 5 - Modeling of repeated eating

[0137] A study of dogs was conducted, including the evaluation of multiple features, 333 training sessions, and 67 animal evaluation sessions. Some of the features in this study include: duration of content removal activity, number of peaks in normalized load, interaction duration, number of peaks, number of troughs per timestamp in normalized load, number of peaks in gradient, lag 2 autocorrelation of gradient RMS, moderate gradient value, number of peaks in gradient, and lag 3 autocorrelation of gradient, etc.

[0138] When processing data from the load sensor of the smart pet food bowl disclosed herein, various machine learning models for continuous variables can be used. The continuous variables used are based on various methods with mean absolute error (MAE), mean absolute percentage error (MAPE), root mean square error (RMSE), and coefficient of determination (R²). 2 The machine learning logic system is used to identify the best model. The data is provided in Table 2, as follows:

[0139]

[0140] As shown in Table 2, there is a strong linear relationship between the predicted number of repetitions and the actual number of repetitions for both the training and evaluation sessions. R0 2 The fact that the value exceeds 90% indicates this. For both the training and evaluation sets, the mean absolute error in predicting the number of feeding repetitions is less than + / - 2.

[0141] Example 6 - Licking and licking load characteristics

[0142] See now Figure 16 and Figure 17 The data shows two sample datasets, including load characteristics that occur when dogs lick water from a smart pet food bowl. Figure 16 ) and the load characteristics that occur when a dog licks water in a smart pet food bowl ( Figure 17The data represents a short timeframe, a few seconds, which may vary from animal to animal. However, these load features, depicting the peaks and troughs of load, are distinguishable from each other, for example, because licking patterns are more rhythmic, while licking patterns are more chaotic. These visual indicators help identify appropriate features to differentiate between licking and licking. For example, some identified features include peak height, distance between peaks, standard deviation of the distance between peaks, number of peaks, features on the peaks and troughs, etc. Summarizing these two examples of load features, licking appears more rhythmic, with the animal placing its tongue into the basin, forming a stream of water, biting the stream, and then the water falling back into the basin. Licking, on the other hand, appears more chaotic, with the animal moving its tongue across the bottom of the basin to collect water remaining on the surface. The two load features depicted in this example are visually consistent with people's understanding of licking and licking, and it can be seen that they are different enough to be distinguishable based on load features alone, without additional input.

[0143] Example 7 - Processing data from the feeding phase

[0144] In this evaluation, the duration (seconds) of feeding and the amount of food consumed (grams) during a single session were assessed for feeding phase identification. The values ​​for pet food provided, pet food remaining (after meal), feeding start, and feeding end were determined, as shown in Table 3A, and then determined using various regression error metrics, as shown in Table 3B. A machine learning model for feeding phase identification was constructed, including: i) the continuous variable duration and ii) the continuous variable consumption. The mean absolute error (MAE), mean absolute percentage error (MAPE), root mean square error (RMSE), and / or coefficient of determination (R²) were investigated. 2 This is used to identify the best model. For example, it was found that MAE helps in understanding the average deviation between predicted and actual values, and R0 was found to be... 2 It helps to assess the linear relationship between actual and predicted values.

[0145]

[0146]

[0147] For MAE and RMSE, smaller values ​​are better in the range [0,∞); while for R2, larger values ​​are better in the range [0,∞).

[0148] The collected data show that the estimated duration averaged approximately + / - 9 seconds of the actual duration, while the actual duration averaged approximately 222 seconds (+ / - 4%). The estimated food consumption averaged + / - 6 grams of the actual consumption, while the actual consumption averaged approximately 193 grams (+ / - 3%). The predicted values ​​for duration and consumption showed an R² value exceeding 98%, indicating that the duration and consumption models are highly accurate.

[0149] Example 8 -Categorization of eating activities within a mealtime conversation (eating and / or drinking)

[0150] Training schemes can be used to improve prediction logic associated with various eating and / or drinking behaviors. In this study, six eating behavior parameters, including eating, pausing eating, touching the food bowl, licking, sniffing food, and food falling, were evaluated and compared with the predicted values ​​of individual micro-events. As mentioned in Example 1, real-world data in the form of video recordings was used in two different studies: one related to the training of the smart pet food bowl, i.e., training data; and the other related to the testing of the smart pet food bowl, i.e., test data. For example, multiple machine learning classifiers were used. The training data was used to establish eating behaviors, and the test data was used to identify an acceptable classification model, which in this case was considered the best classification model. Confusion matrices, along with recall, precision, and F1 score, were used to compare the real-world data with the predicted data.

[0151] Table 4 below shows some example results demonstrating that the modeling performed can have high predictive power for categories with high support. The model performs well in predicting eating, pausing, touching the food bowl, and licking. Using the parameters chosen for this study, the predictions for smelling food and food dropping are poor, but could be improved, for example, by different load sensitivities and / or additional training data.

[0152]

[0153] *Recall rate refers to the proportion of actual positive predictions out of all actual positives.

[0154] **Precision rate refers to the proportion of truly positive predictions among all positive predictions (actual values ​​obtained via the video's true label).

[0155] The F1 score is a metric that combines precision and recall, providing an overall measure of model performance. The maximum F1 score is 100% (representing a perfect model). The model used in this example was selected based on the highest F1 score.

[0156] As shown in Table 4, the model performed exceptionally well in identifying feeding and pausing, with an F1 score exceeding 90%. It also performed well in identifying licking and touching the food bowl, with an F1 score exceeding 80%. The model performed poorly in sniffing and food dropping (a category with low support), but this is likely to continue to improve with additional support, as we did see that adding data improved the results; however, these behaviors are simply rarer in dogs' feeding sessions.

[0157] Example 9Modeling performance during drinking sessions

[0158] Further details regarding characterizing licking and licking during drinking sessions can be obtained by collecting and processing data from load sensors using various modeling or logic methods. For example, machine learning logic can be used to collect data related to various drinking behaviors, such as licking, licking, other, pausing, etc. (evaluating the classification or grouping of four drinking behaviors). In this embodiment, data was collected based on 190 test sessions, with a time increment (or time window) of 0.4 seconds per window used to collect the dataset. The results of this study are presented in Table 5, as follows:

[0159]

[0160] As shown in Table 5, the model performs exceptionally well in identifying licking, sucking, or other behaviors, with an F1 score exceeding 90%. The model also performs well in identifying licking, with an F1 score exceeding 80%.

[0161] Example 10 – Modeling using a combination of licking and feeding data

[0162] Considering that a combined model of licking and repetition frequency is also useful for modeling pet eating behavior for monitoring and / or discovering health insights, Table 6 provides data collected from a combined licking and repetition model of dogs, along with the actual repetition (ϒ) distribution, and Table 7 provides model performance data.

[0163]

[0164]

[0165] MAE = Mean Absolute Error

[0166] RMSE = Root Mean Square Error

[0167] R 2 =Determination coefficient

[0168] As shown in Tables 6 and 7, the model performs well, with a strong linear relationship between the number of predictions and observations, as evidenced by the fact that values ​​on both the training and test sets exceed 90%. The mean absolute error on the test set is ±5 (+ / -19%) of the mean of 26.1. For this merged licking and feeding model, the features that provide the best performance are: i) interaction duration, number of peaks on the load, the ratio of peaks to interaction duration, the total number of peaks on the normalized load, and the ratio of negative peaks to interaction duration.

[0169] Example 11 – Repeated modeling using normalization logic

[0170] The repetition model using normalized logic provides a way to combine data, such as combining outputs from activity classification models and combined repetition models, to determine the number of repetitions of licking and licking. Figure 18 Example load data collected during a single drinking session is shown, including licking and licking (as well as two touches of the feeding bowl). Figure 19 This demonstrates how normalization logic can be applied to, for example... Figure 18 Data collected as shown or in other similar embodiments.

[0171] Example 12 –Performance of single-feeding behavior model

[0172] To illustrate the performance of modeling a single feeding behavior in dogs (rather than two or more combinations of feeding behaviors), namely licking, using a standalone, repetitive model, data was collected and processed as shown in Tables 8 through 10. For this data, 190 test sessions were used, with the machine learning model set to default parameters. In this embodiment, Table 8 provides the results using the standalone licking model with a repetitive (ϒ) distribution, Table 9 provides the performance results of the licking model, and Table 10 provides the normalized licking performance results.

[0173]

[0174]

[0175] MAE = Mean Absolute Error

[0176] RMSE = Root Mean Square Error

[0177] R 2 =Determination coefficient

[0178]

[0179] MAE = Mean Absolute Error

[0180] RMSE = Root Mean Square Error

[0181] R 2 =Determination coefficient

[0182] The data shows that the normalized licking output model in Table 10 performs better than the licking model in Table 9.

[0183] Example 13 –Pet Health Insights

[0184] Regarding health insights, various health issues can be correlated with collected feeding behavior data, based on the animal's interaction with the smart pet bowl disclosed herein. Health insights obtainable from the smart pet bowl and related systems and methods can include identifying conditions related to general animal health, such as dental health, diabetes, kidney health, and digestion. For example, changes in feeding behavior, such as increases or decreases in the frequency of eating or drinking, changes in the duration of eating or drinking, changes in the rate of eating or drinking, changes in the number of repetitions of biting or licking, changes in the number of pauses, and changes in the volume consumed, can provide insights into animal health.

[0185] Example 14 - Cat load data collection and processing

[0186] Much of the data shown and described in the preceding embodiments relates to studies on dogs. However, according to this disclosure, one or more load sensors associated with a smart pet bowl can also be used to collect data on cat eating behavior. Load data can be collected and processed in a manner similar to that described above regarding dog studies. It is worth noting, however, that cats tend to eat and drink faster than dogs, and, on average, cats typically eat and drink in a more graceful and refined manner than dogs. Therefore, for smart pet bowls designed for cats, for example, a load sensor with higher sensitivity may be more helpful compared to one used for dogs. That said, as long as the load sensor is sensitive enough and responds quickly enough to capture the micro-behaviors of interest that require load sensor data capture, a load sensor with sensitivity exceeding the necessary level (too high) may not be a particularly concerning issue.

[0187] As an example of data collection from cat studies, a smart pet bowl with four (4) individual load sensors equidistantly distributed around the bottom of the bowl was used. Notably, fewer or more load sensors can be used, for example, from one to ten. In this example, four independent loads from the four load sensors were collected and then summed to approximate the load that a smart pet bowl with a single load sensor might obtain. The total count, the average load recorded, the standard deviation, the minimum load recorded, the percentage of counts below a specific load value, and the maximum load data were recorded and evaluated. In this study, the cumulative total load sensor readings recorded were below approximately 75 grams, accounting for 99.5% of the counts. Based on these cat data, load sensors with appropriate load sensitivity were selected for use.

[0188] Certain embodiments of the invention have been disclosed in this specification. While specific terminology has been used, it is used in a general and descriptive sense only and not for limiting purposes. The scope of the invention is set forth in the claims. Many modifications and variations of the invention are possible based on the foregoing teachings. Therefore, it should be understood that the invention may be practiced in forms other than those specifically described within the scope of the appended claims.

Claims

1. A method for monitoring pet eating behavior under the control of at least one processor, the method comprising: As the pet interacts with the contents of the pet food bowl, load data is acquired from a load sensor in the pet food bowl, wherein the load sensor has a sensitivity of + / - 50 grams or finer, and the load data is acquired at a sampling rate of 10 to 150 samples per second. The load data is sequentially grouped in time increments ranging from 0.01 seconds to 5 seconds, where a single time increment contains multiple samples; and Based on the load data generated from the interaction between the pet and the pet food bowl or the contents of the pet food bowl, feeding behavior occurring within one or more of the time increments is identified.

2. The method according to claim 1, further comprising: If it is determined that the load data is caused by human-computer interaction, accidental triggering, or accidental interaction with the pet food bowl or the contents of the pet food bowl, then the load data is excluded when identifying the feeding behavior.

3. The method of claim 1, wherein the feeding behavior is a count-based feeding behavior selected from licking, licking, or biting.

4. The method according to claim 3, further comprising: The count-based eating behavior is characterized by counting individual micro-events of the eating behavior over a single time increment or over a period of time spanning multiple time increments, wherein the individual micro-events include individual licking, individual licking, or individual biting.

5. The method of claim 4, wherein characterizing the count-based eating behavior comprises: This represents a time period spanning one or more time increments during which the count-based behavior has not occurred.

6. The method of claim 1, wherein the feeding behavior is a duration-based behavior selected from the pet touching the food bowl, moving the food bowl, smelling the food, pausing, eating, licking, licking or biting.

7. The method of claim 6, further comprising characterizing the duration-based eating behavior by sequentially mapping the time increments in which the duration-based behavior occurred or did not occur.

8. The method of claim 1, further comprising notifying the pet's caregiver of the feeding behavior or changes in the pet's feeding behavior.

9. The method of claim 8, wherein notifying the caregiver of the pet comprises: The caregiver is warned that such changes in the pet's eating behavior may be related to underlying health or behavioral problems.

10. The method according to claim 1, further comprising: Auxiliary data is acquired from auxiliary sensors associated with the pet food bowl, wherein the auxiliary sensors include proximity sensors, cameras, microphones, accelerometers, gyroscopes, inertial measurement unit sensors, or combinations thereof.

11. The method of claim 1, wherein the pet food bowl is a dog water bowl, and wherein the load sensor has a sensitivity of + / -4 grams or finer, the sampling rate is 15 to 75 samples per second, and the time increment is at least about 0.4 seconds.

12. The method of claim 1, wherein the pet food bowl is a dog food bowl, and wherein the load sensor has a sensitivity of + / -4 grams or finer, the sampling rate is 15 to 75 samples per second, and the time increment is at least about 0.3 seconds.

13. The method of claim 1, wherein the pet food bowl is a cat water bowl, and wherein the load sensor has a sensitivity of + / -2 grams or finer, the sampling rate is 15 to 75 samples per second, and the time increment is at least about 0.4 seconds.

14. The method of claim 1, wherein the pet food bowl is a cat food bowl, and wherein the load sensor has a sensitivity of + / -2 grams or finer, the sampling rate is 15 to 75 samples per second, and the time increment is at least about 0.3 seconds.

15. The method of claim 1, further comprising generating a feeding behavior model for the pet, including identifying the feeding behavior of the pet based on feeding frequency, pet feeding characteristic behaviors, or a combination thereof.

16. The method according to claim 1, further comprising: Identify pets in multi-pet households.

17. A non-transitory machine-readable storage medium having instructions thereon, which, when executed, cause a processor to perform a method for monitoring a pet's eating behavior, the method comprising: As the pet interacts with the contents of the pet food bowl, load data is acquired from a load sensor in the pet food bowl, wherein the load sensor has a sensitivity of + / - 50 grams or finer, and the load data is acquired at a sampling rate of 10 to 150 samples per second. The load data is sequentially grouped in time increments ranging from 0.01 seconds to 5 seconds, where a single time increment contains multiple samples; and Based on the load data generated from the interaction between the pet and the pet food bowl or the contents of the pet food bowl, feeding behavior occurring within one or more of the time increments is identified.

18. The non-transitory machine-readable storage medium of claim 17, wherein monitoring the pet's eating behavior includes monitoring count-based eating behavior selected from licking, biting, or chewing.

19. The non-transitory machine-readable storage medium of claim 17, wherein monitoring the pet's eating behavior includes monitoring duration-based behavior selected from touching the food bowl, moving the food bowl, smelling food, pausing, eating, licking, or combinations thereof.

20. The non-transitory machine-readable storage medium of claim 17, wherein monitoring the pet's eating behavior includes notifying the pet's caregiver of the following: the eating behavior, changes in the eating behavior, potential health problems associated with the eating behavior, or a combination thereof.

21. The non-transient machine-readable storage medium of claim 17, wherein the load sensor has a sensitivity of + / -4 grams or finer, the sampling rate is 15 to 75 samples per second, and the time increment is at least about 0.3 seconds.

22. The non-transitory machine-readable storage medium of claim 17, wherein monitoring the pet's eating behavior comprises: When the pet comes from a multi-pet household, identify the pet to be monitored.

Images