Wearable animal ingestion seed identification and mark tracking system

Abstract

The invention relates to the technical field of animal behavior monitoring and ecology research, and discloses a wearable animal ingestion seed identification and mark tracking system, which comprises a wearing device module, which is installed through a necklace and is used for collecting animal head posture data, front images and environment parameter information; the ingestion behavior detection module judges whether the animal is ingestion or not and controls the camera to collect an image; the seed identification and marking module identifies seeds and releases markers; the seed tracking and digestion process prediction module tracks the position change of the marker and predicts the seed discharge time; and the data analysis and feedback module uploads data to a cloud platform through wireless communication to realize model training and identification parameter optimization. Through the wearing device module, the ingestion behavior detection module and the seed identification and marking module, seamless connection from accurate sensing of animal ingestion behaviors to dynamic marking of seeds is achieved, and the problems that a traditional method is low in efficiency and high in subjectivity, and specific seeds cannot be accurately marked in real time are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of animal behavior monitoring and ecological research, and in particular to a wearable animal seed-eating identification and marking tracking system. Background Art

[0002] In the fields of ecology, animal behavior, and agricultural and forestry research, a deep understanding of animal feeding behavior and its impact on plant seed dispersal is of vital importance. For example, clarifying which plant seeds specific animals eat in nature, as well as the length of time these seeds stay in the animal's digestive system and their survival status, constitutes crucial information for accurately assessing the functional role of animals in the ecosystem, predicting the dynamic distribution of plant populations, and even effectively curbing the spread of invasive species.

[0003] At present, research on animal feeding behavior and seed dispersal mainly relies on traditional methods, including manual field observation, which records the types and frequency of animals' feeding through the naked eye of researchers; fecal content analysis, which is to collect animal excrement, separate and identify undigested plant seeds from it to infer the animal's feeding habits; and short-term marking and tracing technology, such as feeding animals food with identifiable marks and then estimating the digestion time by simply monitoring their excretions. These methods provide basic data for animal feeding research to a certain extent.

[0004] However, existing technologies have exposed many limitations that are difficult to overcome in practical applications. First, the recognition accuracy and real-time performance of feeding behavior are insufficient. Manual observation is easily affected by environmental interference and subjective factors of the observer. It is difficult to capture the feeding behavior of animals in hidden environments or at night, and it is impossible to accurately distinguish whether subtle head movements are feeding or other behaviors. This leads to the fact that the recording of feeding events is often not accurate enough and cannot be responded to in real time. Secondly, the function of instant recognition and marking of fed seeds is missing. Although traditional fecal analysis can identify excreted seeds, it cannot establish a direct association between specific feeding behavior and specific seed types, nor can it immediately mark the seeds when they enter the animal's body. This makes it difficult for researchers to obtain information on the initial state of seeds entering the digestive tract, thereby limiting the initial state of seed dispersal. In-depth research on the mechanism; secondly, there is a technical gap in the dynamic tracking and accurate prediction of the digestion process of seeds in animals. The existing methods mainly rely on indirect feces collection to judge the digestion time, and cannot provide continuous and refined in vivo data such as the specific movement trajectory, residence time and digestion stage of seeds in the digestive tract. This makes the evaluation of seed digestion efficiency, survival rate and its propagation potential lack of accurate basis; finally, the existing technologies generally lack environmental adaptability and data-driven adaptive optimization capabilities. The existing monitoring system algorithms are usually fixed. When faced with complex and changeable field environments (such as changes in light, temperature, and humidity), the recognition accuracy and system stability will drop significantly, and it is impossible to self-learn and improve performance based on historical data, resulting in long-term monitoring reliability. For this reason, those skilled in the art have proposed a wearable animal-eating seed identification and marking tracking system to solve the above problems. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides a wearable animal feeding seed identification and marking tracking system, which solves the shortcomings of the existing technology in accurate identification of animal feeding behavior, real-time marking of fed seeds, dynamic tracking of seeds in the body, and system adaptive optimization.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a wearable animal seed identification and marking tracking system, comprising:

[0007] A wearable device module, which is installed on the animal's body surface through a collar and collects head posture data and front image information of the animal during activities. The wearable device module is equipped with a camera, a six-axis motion sensor, and an environmental sensor;

[0008] A feeding behavior detection module determines whether the animal is in a feeding state based on the head posture data, and controls the camera to collect image information and the environmental sensor to collect environmental parameter information when it is determined to be in a feeding state;

[0009] A seed recognition and marking module obtains the image information, identifies the seeds in the image, and triggers a marker release mechanism to release a marker of a corresponding concentration based on the recognition result and environmental parameter information;

[0010] A seed tracking and digestion process prediction module obtains the corresponding concentration of the marker, detects the time series signal changes of the marker in the animal body through an implantable sensor, and tracks the position change trajectory of the marker based on the detected signal, thereby predicting the excretion time of the corresponding seed;

[0011] The data analysis and feedback module obtains the position change trajectory, image information and environmental parameter information of the tracking marker, and uploads the information to the cloud platform through the wireless communication device to realize dynamic optimization of model training and recognition parameters.

[0012] Preferably, the wearable device module includes:

[0013] The collar is made of flexible material, suitable for animals of different sizes and conforming to the animal's body surface;

[0014] A six-axis motion sensor, embedded in the collar, collects acceleration and rotation information of the animal's head in real time and outputs head posture data;

[0015] A camera, located on the outside of the collar and working in sync with the six-axis motion sensor, captures image information in front of the animal;

[0016] Environmental sensor, integrating temperature, humidity and light environment parameter detection functions;

[0017] Wireless communication device, used to upload data collected by the wearable device to a local terminal device or cloud platform in real time.

[0018] Preferably, the feeding behavior detection module includes:

[0019] Based on the acceleration data collected by the six-axis motion sensor, the pitch angle of the animal's head is calculated to determine whether the animal is in a feeding state;

[0020] When the pitch angle exceeds a set threshold and remains for a certain period of time, it is determined that the animal is feeding and the camera is triggered to collect image information;

[0021] The feeding behavior detection module further controls the environmental sensor to collect current environmental temperature, humidity and light parameter information to provide reference data for the subsequent seed identification and marking module.

[0022] Preferably, the pitch angle θ is calculated by a six-axis motion sensor, and the specific calculation method is:

[0023]

[0024] Among them: a y Represents the acceleration component of the sensor in the front-back direction of the animal's head; a z Represents the acceleration component in the vertical direction; the pitch angle is used to determine whether the animal is eating.

[0025] Preferably, the seed identification and marking module includes:

[0026] Processing the image captured by the camera, extracting image features and identifying seeds in the image through a pre-trained seed recognition model to determine their category;

[0027] According to the identified seed category and environmental parameter information, the marker release device is controlled to release markers of corresponding concentrations, and the concentration is dynamically adjusted according to the seed category and environmental conditions.

[0028] Preferably, the seed recognition model is trained by a convolutional neural network, the input of the model is the processed image data, and the output is the seed category probability distribution:

[0029] P(c|I)=Softmax(W·GlobalAvgPool(f CNN (I))+b);

[0030] Where: I represents the input image information collected by the camera and pre-processed; c represents a specific category in the preset seed category set; f CNN (·) represents the feature extraction function of the convolutional neural network; GlobalAvgPool(·) represents the global average pooling layer; W and b represent the weight matrix and bias vector of the last fully connected layer of the model, respectively; Softmax(·) is the activation function; P(c|I) is the probability that a seed of category c exists in image I.

[0031] Preferably, the seed tracking and digestion process prediction module includes:

[0032] An implantable sensor is installed in the animal's body and interacts with the marker to detect the signal strength of the marker;

[0033] By analyzing the temporal changes in the marker signal, the movement trajectory of the marker in the animal body is tracked, and the seed excretion time is calculated based on the movement trajectory;

[0034] The time from seed entry to excretion is predicted by combining the animal's weight and the average rate of movement through the digestive tract.

[0035] Preferably, the movement trajectory x(t) of the marker in the animal body depends on the change in its signal intensity in the animal body, and the specific calculation formula is:

[0036]

[0037] Where: x(t) represents the estimated distance traveled by the marker along the digestive tract at time t; v digest represents the average movement rate of the contents of the digestive tract; t0 represents the initial moment when the seed is ingested; t represents the current moment; S(t′) represents the marker signal intensity detected by the implanted sensor at the integral variable moment t′; S0 represents the reference signal intensity detected at the initial moment t0.

[0038] Preferably, the data analysis and feedback module includes:

[0039] By receiving all collected data and storing it in the cloud;

[0040] Based on historical and real-time data, a correlation model between environmental factors and seed identification accuracy is constructed to optimize the environmental impact during the identification process;

[0041] Then, based on the feedback from the cloud platform, the seed recognition model is dynamically updated, and the optimized model is downloaded to the wearable device module via the wireless communication device.

[0042] Preferably, the seed recognition model is optimized by:

[0043]

[0044] in: and Represent the model parameters before and after the update respectively; η represents the learning rate, which controls the step size of each parameter update; Represents the gradient of the model parameter θ; represents the cross entropy loss; λ represents the regularization coefficient; Represents the square of the Frobenius norm of the environmental impact correlation matrix M.

[0045] The present invention provides a wearable animal seed identification and marking tracking system. It has the following beneficial effects:

[0046] 1. The present invention achieves a seamless transition from accurate perception of animal feeding behavior to dynamic seed marking through a wearable device module, a feeding behavior detection module, and a seed identification and marking module. Compared with existing solutions that rely on manual observation or a single sensor for behavior judgment and marking, the present invention solves the problems of low efficiency, strong subjectivity, and inability to accurately mark specific seeds in real time with traditional methods, greatly improving the automation level of data collection and the targeting of marking.

[0047] 2. The present invention introduces a seed tracking and digestion process prediction module, and closely couples it with the data analysis and feedback module to construct a complete chain from in vivo signal analysis to excretion time prediction. Compared with the existing method that mainly relies on indirect analysis of feces collection in vitro, the present invention realizes the quantitative tracking of the movement trajectory of seeds in the animal digestive tract and the accurate prediction of excretion time.

[0048] 3. The system of the present invention has strong adaptive and self-learning capabilities. It continuously optimizes the core algorithm through data analysis and feedback modules, and ensures that the recognition accuracy and overall performance of the system can be continuously improved under different environmental conditions and long-term use through a closed-loop feedback mechanism. Compared with the fixed algorithm model and the lack of environmental adaptability in the existing technology, the present invention solves the problem of performance degradation of the model in complex and changing environments, making the system more reliable and more applicable in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 Schematic diagram of the system architecture of the present invention;

[0050] Figure 2 Schematic diagram of the module structure of the wearing device of the present invention.

[0051] Among them, 1. Collar; 2. Six-axis motion sensor; 3. Camera; 4. Environmental sensor. DETAILED DESCRIPTION

[0052] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the present specification. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0053] Please see the attached Figure 1 and attached Figure 2 The embodiment of the present invention provides a wearable animal seed identification and marking tracking system, comprising:

[0054] The wearable device module is installed on the animal's body surface through a collar and collects head posture data and front image information during the animal's activities. The wearable device module is equipped with a camera, a six-axis motion sensor, and an environmental sensor;

[0055] Specifically, this embodiment provides a wearable device module for a wearable animal seed identification and tracking system. As the foundation of the system, this module provides a comfortable wearing experience for animals and, through embedded sensors, enables real-time monitoring of animal feeding behavior and environmental changes. The core function of the wearable device module is to collect animal head posture data and image information. Working in conjunction with other modules, it transmits this collected data to a cloud platform or local terminal device via wireless communication, enabling comprehensive monitoring of animal activities and their surroundings.

[0056] In this embodiment, the wearable device module is designed with consideration given to both animal comfort and sensor efficiency. Specifically, it includes a collar, a six-axis motion sensor, a camera, environmental sensors, and a wireless communication device. These components work together to enable the system to capture animal behavior in real time and provide data support for subsequent seed identification and tracking.

[0057] Collar Design: In this embodiment, the collar is constructed of a flexible material that adjusts to the animal's size, ensuring a snug, comfortable fit. The collar's design offers adaptability and comfort, making it suitable for a wide variety of animals, particularly those with varying body sizes, such as livestock and wildlife. The collar's flexible material ensures portability and durability, making it virtually invisible to the animal during daily activities.

[0058] Six-axis motion sensor: The collar is embedded with a six-axis motion sensor, including an accelerometer and gyroscope, that collects real-time acceleration and rotation information from the animal's head. This sensor outputs data on changes in the animal's head posture and uses this data to determine the animal's activity status. In this embodiment, the acceleration data from the six-axis motion sensor is calibrated using the following formula:

[0059]

[0060] Among them: a raw is the original acceleration value; a corrected is the acceleration value after calibration; N is the number of sampling points.

[0061] This calibration process ensures the accuracy of the data, so that the collected motion data can truly reflect the dynamic changes of the animal's head.

[0062] Camera: The collar features a camera located in front of the animal's head, capturing real-time images of the animal's surroundings. This is particularly important when the animal is feeding, providing clear images of the feed. The camera works in sync with a six-axis motion sensor, triggering image acquisition based on feeding behavior, ensuring the image data aligns with the animal's behavior. The camera is designed to adapt to varying lighting conditions, providing high-quality image data in changing environments.

[0063] Environmental Sensors: The wearable device also includes integrated environmental sensors that monitor temperature, humidity, and light levels. These sensors collect real-time data on various parameters of the animal's environment and transmit this information to other components of the system. This environmental data enables the system to comprehensively assess the conditions surrounding the animal and, combined with collected image information, conduct comprehensive analysis to provide more accurate seed identification and tagging.

[0064] Wireless communication: The wearable device has a built-in wireless communication device (such as Bluetooth), which can upload collected data in real time to a local terminal device (such as a smartphone or tablet) or a cloud platform. Through wireless communication, data can be transmitted to external devices for subsequent processing, analysis, and feedback. The design of the wireless communication device ensures the real-time performance and remote control capabilities of the system, enabling the system to provide continuous monitoring and data feedback in dynamic environments.

[0065] In this embodiment, the working principle of the wearable device module is as follows:

[0066] Device Initialization and Pairing: The wearable device is paired with the local terminal via Bluetooth, completing device ID registration and cloud account binding. Through the Bluetooth connection, users can quickly configure the wearable device and ensure its interoperability with other system modules. After device initialization is complete, the sensor enters the operating state and prepares to collect data.

[0067] Data Collection and Real-Time Monitoring: The wearable device collects the animal's head posture data (via a six-axis motion sensor) and image information (via a camera), while also monitoring ambient temperature, humidity, and light levels. After calibrating the acceleration data, the sensor continuously outputs real-time data, ensuring the system can track and record changes in the animal's posture throughout its activities.

[0068] Data Transmission and Feedback: After data collection, wireless communication devices are used to upload the data in real time to a local terminal or cloud platform. On the cloud platform, the system can analyze the collected data in real time, provide feedback to other modules, and dynamically adjust seed identification and tagging strategies. The high efficiency of wireless communication devices ensures real-time data transmission and smooth system operation.

[0069] The feeding behavior detection module determines whether the animal is in a feeding state based on head posture data. If it is determined to be in a feeding state, it controls the camera to collect image information and the environmental sensor to collect environmental parameter information;

[0070] Specifically, after the wearable device module completes real-time collection of the animal's posture data and image information, the system needs to effectively determine whether the animal is currently feeding to avoid unnecessary operation of the camera and other modules during non-feeding periods. This module, known as the feeding behavior detection module, is primarily used to determine whether the animal is feeding based on the posture information provided by the six-axis motion sensor. If the animal is determined to be feeding, it further triggers the simultaneous collection of image and environmental data.

[0071] In this embodiment, the feeding behavior detection module, as the second stage of the system process, focuses on identifying animal behavior based on the dynamic posture changes of the animal's head. This serves as a prerequisite for subsequent image acquisition and seed recognition. By collaborating with the wearable device module, this module significantly improves the timeliness and relevance of image acquisition, reduces system load, and improves recognition efficiency.

[0072] Generally, when an animal is feeding, its head will exhibit a specific pitch angle characteristic with a certain degree of persistence. To fully utilize this behavioral pattern, this module introduces a quantitative analysis mechanism for head posture data.

[0073] Specifically, this module first calls the three-axis acceleration data provided by the six-axis motion sensor and sets a judgment formula based on the pitch angle change to determine the feeding status.

[0074] In one possible implementation, the pitch angle θ is calculated as follows:

[0075]

[0076] Among them: a y Indicates the acceleration component of the sensor in the front-back direction of the animal's head, in m / s 2 ;a z Indicates the acceleration component in the vertical direction, in m / s 2 ; θ is the pitch angle of the head relative to the horizontal plane, in radians (rad) or degrees (°). The angle unit can be selected according to the application.

[0077] As an option, in some embodiments, to improve the accuracy of feeding state determination, the system also incorporates angular velocity threshold judgment logic. Using angular velocity data provided by the gyroscope, the rate of change of the pitch angle within the detection period is set. If the rate remains stable within a certain angle range within a certain time window, it can be inferred that the animal has entered feeding behavior.

[0078] In a more specific implementation, the system uses a 0.5 second sliding time window to average the pitch angles of consecutive data points and compares them with a set threshold. The judgment criteria are as follows:

[0079]

[0080] Where: θ(t) is the pitch angle at time point t; θ0 is the reference feeding angle set by experience, which is generally set to the average pitch angle when the animal is in a flat head state; ∈ is the feeding state angle fluctuation tolerance, in degrees or radians; T represents the total number of data points in the sliding window.

[0081] In some embodiments, the absolute value change of acceleration in the front-to-back direction of the head is also used as an auxiliary judgment basis, that is:

[0082] |a y (t)-a y (t-1)|<δ;

[0083] Where: δ represents the acceleration difference tolerance value, which is used to confirm the stability of the head position change; a y (t) represents the acceleration component of the sensor in the front-back direction of the animal's head, the value at time t.

[0084] Once the detection module determines that an animal has entered a feeding state, it immediately sends a trigger signal to the camera and environmental sensors. The camera begins continuous image acquisition, while the environmental sensors simultaneously collect current temperature, humidity, and light parameters. This mechanism ensures that the system captures images and environmental data during feeding, providing the raw information for subsequent seed identification and environmental condition matching analysis.

[0085] In addition, to enhance the robustness of the feeding state detection algorithm, some embodiments also introduce an adaptive threshold update strategy. This strategy dynamically adjusts the threshold parameters ∈, δ, and the reference feeding angle θ0 based on historical feeding behavior data. The update formula can use the exponentially weighted moving average (EWMA) method:

[0086]

[0087] Where: θ new is the average pitch angle in the current judgment period; α is the smoothing factor 0<α<1, which is used to control the weight ratio of historical values to new values; It is the updated value of the reference feeding angle at time t+1.

[0088] The design of this module can effectively reduce the misjudgment rate, especially in dynamic environments such as when animals are running or shaking their heads. The system still has a certain ability to distinguish and avoid false triggering of image acquisition operations.

[0089] The seed recognition and marking module obtains image information, identifies the seeds in the image, and triggers the marker release mechanism to release the corresponding concentration of marker based on the recognition result and environmental parameter information;

[0090] Specifically, after the feeding behavior detection module confirms that the animal has entered the feeding state, the system immediately activates the seed identification and marking module. This module, following instructions triggered by the preceding module, acquires the image and environmental information captured in real time by the wearable device module. Its core task is to accurately identify seeds in the image and, based on the identification results and environmental parameters, dynamically control the marker release mechanism to complete the marking process for specific seeds. This module is the key link between animal behavior perception and subsequent in-vivo tracking. The accuracy of its recognition and the effectiveness of its marking directly determine the data quality of the entire system.

[0091] In this embodiment, the seed recognition and marking module integrates advanced image processing technology with an intelligent marker release strategy. It not only determines seed type but also dynamically adjusts the marking scheme based on environmental factors, ensuring that markers remain effectively attached and tracked by subsequent modules under varying conditions.

[0092] Generally speaking, the original image information captured by the camera may contain complex background and noise interference. Therefore, before seed recognition, the image needs to be preprocessed. Specifically, the first step in image processing is to reduce noise and enhance the image. In some embodiments, Gaussian filtering or median filtering algorithms can be used to remove random noise. Subsequently, the target region of interest (ROI) suspected of being a seed is extracted from the background through edge detection or color segmentation technology, thereby focusing computing resources and improving recognition efficiency.

[0093] After image preprocessing, the system uses a pre-trained seed recognition model to classify the target region. Alternatively, this system employs a lightweight convolutional neural network (CNN), such as the MobileNetV3 architecture. This architecture maintains high recognition accuracy while requiring minimal computational resources, making it suitable for deployment in edge computing scenarios such as wearable devices. The model's task is to output a probability distribution representing the likelihood that a seed in the image belongs to each pre-defined category.

[0094] In one possible implementation, the seed class probability distribution P(c|I) is calculated as follows:

[0095] P(c|I)=Softmax(W·GlobalAvgPool(f CNN (I))+b);

[0096] Where: I represents the input image information collected by the camera and pre-processed; c represents a specific category in the preset seed category set; f CNN (·) represents the feature extraction function of the convolutional neural network, which maps the input image I into a high-dimensional feature vector; GlobalAvgPool(·) represents the global average pooling layer, which is used to reduce the spatial dimension of the feature map to one dimension to reduce the number of parameters; W and b represent the weight matrix and bias vector of the last fully connected layer of the model, respectively; Softmax(·) is the activation function, which converts the output score of the model into a probability distribution with a sum of 1; P(c|I) is the probability that a seed of category c exists in image I.

[0097] When the recognition probability P(c|I) of a certain category exceeds the preset confidence threshold, the seed recognition and marking module will trigger the marker release mechanism.

[0098] Specifically, the release of the markers is not static but is dynamically adjusted based on the identified seed type c and environmental parameters collected by environmental sensors. This is to address the impact of different environments on the marker's adhesion and stability. For example, high temperatures may cause the marker to evaporate or degrade too quickly.

[0099] In this example, the marker release concentration C marker The calculation method is:

[0100]

[0101] Where: C marker represents the concentration of marker released at the final decision; k c represents the reference calibration coefficient associated with seed category c. Different seeds require different reference concentrations to ensure effective labeling due to their different surface physical and chemical properties; T env represents the current ambient temperature measured by the environmental sensor in degrees Celsius; T0 represents a reference base temperature, such as 25 degrees Celsius; α represents the temperature adjustment factor, which is a dimensionless parameter used to quantify the degree of influence of temperature on the concentration of the marker.

[0102] In some embodiments, in order to further improve the adaptability of the labeling strategy, other environmental parameters such as humidity can also be taken into consideration. For example, the humidity factor can be added to the concentration calculation formula:

[0103]

[0104] Among them: H env is the current ambient humidity; H0 is the reference humidity; γ is the humidity adjustment factor.

[0105] The seed tracking and digestion process prediction module obtains the corresponding concentration of markers, detects the changes in the timing signal of the markers in the animal body through implanted sensors, and tracks the trajectory of the marker position changes based on this, and then predicts the excretion time of the corresponding seeds;

[0106] Specifically, after the seed identification and labeling module successfully applies a dynamic concentration of markers to the seeds consumed by the animal, the system's task enters the stage of tracking these marked seeds within the animal's digestive system. This module, the seed tracking and digestion process prediction module, is responsible for inheriting the labeling results of the previous link. Through sensors pre-implanted in the animal's body, it monitors the timing signals of the markers in real time, analyzes the movement trajectory of the seeds based on the signal changes, and ultimately makes a scientific prediction of the seed excretion time. This module is the core of achieving a complete closed loop from in vitro monitoring to in vivo tracking, providing direct data support for studying the digestive physiology of animals and the changes in the state of seeds after passing through the digestive tract.

[0107] In this example, the Seed Tracking and Digestion Prediction Module uses a non-invasive signal detection method combined with a mathematical model to quantitatively analyze the dynamic processes of seeds in the digestive tract. The module is designed to obtain precise spatiotemporal data on the markers within the animal body and convert this data into understandable predictions about the digestive process.

[0108] Typically, after a marker enters an animal's digestive tract, its signal changes regularly as it moves through the digestive system. This system exploits this principle, capturing these changes with an implanted sensor.

[0109] Specifically, this module includes one or more implantable sensors. These sensors are strategically placed within the animal's body, for example, subcutaneously or near the wall of the digestive tract, to detect specific signals emitted by the marker, such as fluorescence, weak radioactive signals, or pH changes. The intensity of the signal collected by the sensor is a function of time, reflecting the change in the relative distance between the marker and the sensor.

[0110] In one possible implementation, the system analyzes the temporal changes in the marker's signal intensity to infer its trajectory within the digestive tract. The signal processing module processes the received discrete signal sequence to estimate the marker's position. The cumulative distance x(t) the marker has traveled within the digestive tract can be estimated by integrating the changes in signal intensity using the following calculation method:

[0111]

[0112] Where: x(t) represents the estimated distance traveled by the marker along the digestive tract at time t, in meters (m); v digest represents the average rate of movement of the contents of the digestive tract. This parameter can be pre-calibrated according to the species, age and physiological state of the animal, and the unit is meter per second (m / s); t0 represents the initial moment when the seed is ingested; t represents the current moment; S(t′) represents the marker signal strength detected by the implanted sensor at the integral variable moment t′; S0 represents the reference signal strength detected at the initial moment t0, which serves as the normalization benchmark.

[0113] By correlating the relative rate of change of signal intensity with the movement speed of the marker, the total movement distance can be obtained by time accumulation.

[0114] After obtaining the marker's position trajectory, this module further enables the prediction of expulsion time. This prediction relies not only on the tracked position but also incorporates the animal's own physiological characteristics to improve prediction accuracy.

[0115] Alternatively, the estimated discharge time t corresponding to the seed exit The calculation can be combined with the animal's weight information:

[0116]

[0117] Where: t exit represents the predicted moment of seed excretion; t ingest represents the time when the seed is ingested, which is consistent with the aforementioned t0; m represents the weight of the animal in kilograms (kg), which makes the model individualized; β represents a dimensionless empirical correction coefficient, which is used to correct the nonlinear effect of different physiological conditions or food types on the total digestion time; v digest It also represents the average rate of movement of digestive tract contents.

[0118] In some embodiments, in order to make the prediction model more refined, the average moving speed v digest It is not a constant. It can be modeled as a dynamic variable whose value is affected by the seed category (identified by module three) and the animal's activity state (indirectly reflected by the movement data of module two). For example, a dynamic rate model v can be established. digest (c,A), where c is the seed category and A is the animal's activity intensity index.

[0119] Furthermore, to address the tracking challenges posed by the complex structure of the digestive tract, some embodiments may employ multiple implantable sensor arrays. By analyzing the differences in signal strength and time delays received by different sensors, the system can employ triangulation or more complex source localization algorithms to obtain a more precise trajectory of the marker in three-dimensional space, rather than just the one-dimensional distance along the digestive tract.

[0120] The data analysis and feedback module obtains the tracking marker position change trajectory, image information and environmental parameter information, and uploads the information to the cloud platform through wireless communication devices to achieve dynamic optimization of model training and recognition parameters.

[0121] Specifically, after the Seed Tracking and Digestion Prediction Module completes the analysis of the time-series signals of the markers within the animal's body, the system has accumulated multidimensional data covering the entire chain from animal feeding behavior to seed digestion. This data includes image information, environmental parameters, and the movement trajectory of the seeds within the body. To fully tap the value of this data and enhance the overall intelligence of the system, the Data Analysis and Feedback Module, as the highest-level functional unit of the system, has the core responsibility of conducting in-depth analysis and collaborative modeling of all data uploaded to the cloud. Based on the analysis results, it dynamically optimizes the recognition models and related parameters in the front-end equipment, thereby building an adaptive, self-learning closed-loop intelligent system.

[0122] In this embodiment, the data analysis and feedback module is the hub connecting the data collection end and the intelligent decision-making end. It is not only a data aggregation center, but also a knowledge generation and feedback center, ensuring that the system can continuously evolve as the amount of data grows, improving its robustness and accuracy in complex and changing environments.

[0123] Typically, various data collected by wearable devices are transmitted in real time or in batches via wireless communication to a cloud platform. This platform provides powerful computing power and massive storage space, making it an ideal environment for performing complex data analysis and model training.

[0124] Specifically, this module receives the data streams generated by the previous modules, including but not limited to: the seed recognition results and corresponding original images generated by module three, the environmental parameter information (temperature, humidity, light) collected by module two, and the seed body movement trajectory analyzed by module four.

[0125] In one possible implementation, to quantify the impact of environmental factors on seed recognition accuracy, this module first constructs an environmental impact correlation matrix. This is intended to identify which environmental conditions significantly interfere with the performance of the seed recognition model. The correlation matrix M can be constructed as follows:

[0126]

[0127] Where: M ij represents the element in the correlation matrix, representing the overall influence of the jth environmental factor on the seed identification task; P(c k |I) is the output of the seed recognition model in module 3, about category c k The posterior probability in image I; E j Represents the jth environmental factor, whose set is {T env ,H env ,L env}, corresponding to ambient temperature, humidity and light intensity respectively; Represents the partial derivative of the recognition probability to changes in environmental factors, that is, sensitivity, which is used to measure the impact of small changes in the environment on the confidence of the recognition results; is the weight set for the jth environmental factor, allowing researchers to adjust the importance of different environmental factors based on prior knowledge.

[0128] After calculating the correlation matrix M, the system obtains quantitative information about the model's "vulnerabilities" in the current environment. The next key step is to use this information to guide the optimization process of the seed recognition model, making it more resistant to these interference factors.

[0129] As an option, the model optimization process in this embodiment introduces a regularization term related to environmental influences based on the traditional loss function. The parameters of the seed recognition model (such as the CNN model mentioned in Module 3) The update rules are as follows:

[0130]

[0131] in: and Represent the model parameters before and after the update respectively; η represents the learning rate, which controls the step size of each parameter update; Represents the gradient of the model parameter θ; stands for Cross-Entropy Loss, which is a standard loss function that measures the difference between the model's classification prediction and the true label; λ stands for the regularization coefficient, which is a hyperparameter used to balance the relationship between classification accuracy and the model's robustness to the environment; Represents the square of the Frobenius norm of the environmental impact correlation matrix M.

[0132] By minimizing this term in the optimization objective, we can effectively force the model to adjust its parameters θ to reduce its sensitivity to high-impact environmental factors.

[0133] Once the cloud platform completes model parameter optimization, the updated model file is pushed back to the wearable device module. Upon receiving the new model, the wearable device replaces the existing recognition model. This process completes a complete feedback loop from data collection, cloud analysis, model optimization, and front-end deployment.

[0134] In some embodiments, this module can also perform deeper data mining. For example, the seed discharge time t tracked by module 4 exit By performing correlation analysis with the seed category c identified in module three and the animal activity data collected in module two, a complex predictive model of "specific seeds-animal activity intensity-digestion rate" was established, providing richer insights for nutritional and ecological research.

[0135] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A wearable animal seed identification and marking tracking system, characterized in that: include: A wearable device module, which is installed on the animal's body surface through a collar and collects head posture data and front image information of the animal during activities. The wearable device module is equipped with a camera, a six-axis motion sensor, and an environmental sensor; A feeding behavior detection module determines whether the animal is in a feeding state based on the head posture data, and controls the camera to collect image information and the environmental sensor to collect environmental parameter information when it is determined to be in a feeding state; A seed recognition and marking module obtains the image information, identifies the seeds in the image, and triggers a marker release mechanism to release a marker of a corresponding concentration based on the recognition result and environmental parameter information; A seed tracking and digestion process prediction module obtains the corresponding concentration of the marker, detects the time series signal changes of the marker in the animal body through an implantable sensor, and tracks the position change trajectory of the marker based on the detected signal, thereby predicting the excretion time of the corresponding seed; The data analysis and feedback module obtains the position change trajectory, image information and environmental parameter information of the tracking marker, and uploads the information to the cloud platform through the wireless communication device to realize dynamic optimization of model training and recognition parameters.

2. A wearable animal seed identification and marking tracking system according to claim 1, characterized in that: The wearing device module includes: The collar is made of flexible material, suitable for animals of different sizes and conforming to the animal's body surface; A six-axis motion sensor, embedded in the collar, collects acceleration and rotation information of the animal's head in real time and outputs head posture data; A camera, located on the outside of the collar and working in sync with the six-axis motion sensor, captures image information in front of the animal; Environmental sensor, integrating temperature, humidity and light environment parameter detection functions; The data collected by the wearable device module is uploaded to the local terminal device or cloud platform in real time through wireless communication.

3. A wearable animal seed identification and marking tracking system according to claim 1, characterized in that: The feeding behavior detection module includes: Based on the acceleration data collected by the six-axis motion sensor, the pitch angle of the animal's head is calculated to determine whether the animal is in a feeding state; When the pitch angle exceeds a set threshold and remains for a certain period of time, it is determined that the animal is feeding and the camera is triggered to collect image information; The feeding behavior detection module further controls the environmental sensor to collect current environmental temperature, humidity and light parameter information to provide reference data for the subsequent seed identification and marking module.

4. A wearable animal seed identification and marking tracking system according to claim 3, characterized in that: The pitch angle θ is calculated by a six-axis motion sensor, and the specific calculation method is: Among them: a y Represents the acceleration component of the sensor in the front-back direction of the animal's head; a z Represents the acceleration component in the vertical direction; the pitch angle is used to determine whether the animal is eating.

5. The wearable animal seed identification and marking tracking system according to claim 1, characterized in that: The seed identification and marking module includes: Processing the image captured by the camera, extracting image features and identifying seeds in the image through a pre-trained seed recognition model to determine their category; According to the identified seed category and environmental parameter information, the marker release device is controlled to release markers of corresponding concentrations, and the concentration is dynamically adjusted according to the seed category and environmental conditions.

6. A wearable animal seed identification and marking tracking system according to claim 5, characterized in that: The seed recognition model is trained by a convolutional neural network. The input of the model is the processed image data, and the output is the probability distribution of seed categories: P(c|I)=Softmax(W·GlobalAvgPool(f CNN (I))+b); Where: I represents the input image information collected by the camera and pre-processed; c represents a specific category in the preset seed category set; f CNN (·) represents the feature extraction function of the convolutional neural network; GlobalAvgPool(·) represents the global average pooling layer; W and b represent the weight matrix and bias vector of the last fully connected layer of the model, respectively; Softmax(·) is the activation function; P(c|I) is the probability that a seed of category c exists in image I.

7. The wearable animal seed identification and marking tracking system according to claim 1, characterized in that: The seed tracking and digestion process prediction module includes: An implantable sensor is installed in the animal's body and interacts with the marker to detect the signal strength of the marker; By analyzing the temporal changes in the marker signal, the movement trajectory of the marker in the animal body is tracked, and the seed excretion time is calculated based on the movement trajectory; The time from seed entry to excretion is predicted by combining the animal's weight and the average rate of movement through the digestive tract.

8. The wearable animal seed identification and marking tracking system according to claim 7, characterized in that: The movement trajectory x(t) of the marker in the animal body depends on the change in its signal intensity in the animal body. The specific calculation formula is: Where: x(t) represents the estimated distance traveled by the marker along the digestive tract at time t; v digest represents the average movement rate of the contents of the digestive tract; t0 represents the initial moment when the seed is ingested; t represents the current moment; S(t′) represents the marker signal intensity detected by the implanted sensor at the integral variable moment t′; S0 represents the reference signal intensity detected at the initial moment t0.

9. The wearable animal seed identification and marking tracking system according to claim 1, characterized in that: The data analysis and feedback module includes: By receiving all collected data and storing it in the cloud; Based on historical and real-time data, a correlation model between environmental factors and seed identification accuracy is constructed to optimize the environmental impact during the identification process; Then, based on the feedback from the cloud platform, the seed recognition model is dynamically updated, and the optimized model is downloaded to the wearable device module via the wireless communication device.

10. The wearable animal seed identification and marking tracking system according to claim 9, characterized in that: The seed recognition model is optimized in the following ways: in: and Represent the model parameters before and after the update respectively; η represents the learning rate, which controls the step size of each parameter update; Represents the gradient of the model parameter θ; represents the cross entropy loss; λ represents the regularization coefficient; Represents the square of the Frobenius norm of the environmental impact correlation matrix M.

Images